textstringThe person's words, up to 300 characters.
min length 1 · max length 300
Look up any past season, game, record or betting trend, with its sources, coverage and sample size.
| Method | Path | Title | Access | Price | MCP tool |
|---|---|---|---|---|---|
| GET | /v1/entities:resolve | Resolve a name to IDs | public | 1 | resolve_entity |
| GET | /v1/metrics | Describe stats | public | 1 | describe_metrics |
| GET | /v1/coverage | Get a stat's coverage | public | 1 | get_coverage |
| POST | /v1/stats:find | Find stats | public | 1 | find_stats |
| GET | /v1/players/{player_id}/history | Get a player's history | public | 1 | get_player_history |
| GET | /v1/teams/{team_id}/history | Get a team's history | public | 1 | get_team_history |
| GET | /v1/games/{game_id}/history | Get a past game | public | 1 | get_game_history |
| GET | /v1/records | Get records | public | 1 | get_records |
| POST | /v1/lines:history | Get a betting trend | public | 1 | get_line_history |
| POST | /v1/views:query | Get a stats table | public | 1 | get_view |
| POST | /v1/views:parse | Turn words into a stats table | public | 0 | parse_view |
| POST | /v1/rules:test | Test a betting rule | public | 1 | test_rule |
| POST | /v1/charts:export | Export a chart | public | 0 | export_chart |
| POST | /v1/rules:compare_source | Compare a rule with its source's record | public | 1 | compare_rule_to_source |
| POST | /v1/rules:test_portfolio | Test betting rules together | public | 1 | test_portfolio |
| GET | /v1/rules/templates | List system templates | public | 0 | list_rule_templates |
| POST | /v1/rules/templates:fill | Fill in a system template | public | 0 | fill_rule_template |
| POST | /v1/rules:interview | Build a rule by questions | public | 1 | build_rule |
| POST | /v1/data-requests:draft | Draft a data request letter | public | 0 | create_data_request |
| POST | /v1/rules:test_exchange | Test an exchange trading strategy | public | 1 | test_exchange_strategy |
| POST | /v1/rules:test_batch | Test many betting rules at once | public | 1 | test_rules_batch |
| GET | /v1/rules/runs/{run_id} | Get a rule test you ran | public | 0 | get_rule_run |
| GET | /v1/rules/{rule_id}/versions | List a saved rule's versions | public | 0 | list_rule_versions |
| GET | /v1/rules/{rule_id}/diff | Compare two versions of a saved rule | public | 0 | compare_rule_versions |
| POST | /v1/rules:translate | Write a rule from words | public | 0 | translate_rule |
| GET | /v1/rules/{rule_id}/record | Get a saved rule's record since it was saved | public | 0 | get_rule_record |
| POST | /v1/rules/{rule_id}:register | Register a saved rule in public | public | 0 | register_rule |
| GET | /v1/registrations/{registration_id} | Get a registered rule | public | 0 | get_registration |
| POST | /v1/registrations/{registration_id}:withdraw | Withdraw a registration | public | 0 | withdraw_registration |
| GET | /v1/rules/{rule_id}/matches | List a saved rule's matches now | public | 0 | list_rule_matches |
| POST | /v1/rules:sweep | Sweep a betting rule's numbers | public | 1 | sweep_rule |
| GET | /v1/rules/runs/{run_id}/export | Export a rule test's bets | public | 0 | export_rule_run |
| POST | /v1/rules:compare | Compare rule tests | public | 0 | compare_rules |
| POST | /v1/rules:share | Share a rule test | public | 0 | share_rule_run |
| GET | /v1/rules/reports/{report_id} | Get a shared rule test | public | 0 | get_rule_report |
| GET | /v1/rules/runs | List the rule tests you ran | public | 0 | list_rule_runs |
| GET | /v1/trends | List trends that survived screening | public | 0 | list_trends |
| POST | /v1/trends:audit | Audit a published trend | public | 1 | audit_trend |
| POST | /v1/rules:import | Import a betting system | write:formulas | 40 | import_rule |
| GET | /v1/rules/imports/{import_id} | Get an imported system | read:formulas | 0 | get_rule_import |
| POST | /v1/rules:import/{import_id}:confirm | Save an imported system | write:formulas | 1 | confirm_rule_import |
| POST | /v1/rules:import/{import_id}:undo | Undo a saved import | write:formulas | 0 | undo_rule_import |
| GET | /v1/rules/imports | List your imported systems | read:formulas | 0 | list_rule_imports |
GET /v1/entities:resolve
public, no key needed · 1 credit, not charged during the beta · MCP tool resolve_entity · status live
A player's, team's, coach's or game's name to Kelly IDs: ranked candidates with team, seasons, position and birth date, and the ID when exactly one is meant. When two match, the result asks which, never guesses.
qstring · queryrequiredThe name as the person said it, e.g. 'Gary Payton'.
min length 1 · max length 80
kindstring · queryplayer, team, game or competition; default any.
One of: player, team, game, competition (open: new values may be added).
sportstring · queryA sport slug, e.g. basketball.
seasonstring · queryA season they played in, e.g. 1995-96.
pattern ^[0-9]{4}(-[0-9]{2})?$
team_idstring · queryA team they played for (tm_…).
positionstring · queryA position, e.g. C.
max length 8
limitinteger · queryCandidates, 1 to 10; default 5.
min 1 · max 10
200 The entities facts.
objectconstrequiredAlways entities.
Always "entities".
resolvedstringrequirednullableThe ID when exactly one candidate is meant; null when the result asks.
candidatesarray of objectrequiredCandidates, best first.
idstringrequiredIts Kelly ID.
kindstringrequiredplayer, team, game or competition.
namestringrequiredIts name.
sportstringnullableIts sport.
matchstringrequiredexact, alias, prefix, trigram or vector.
One of: exact, alias, prefix, trigram, vector (open: new values may be added).
scorenumberrequiredName similarity, 0 to 1.
teamsarray of objectTeams, latest first.
team_idstringnullableIts Kelly ID.
namestringnullableIts name.
fromstringnullableFirst season.
tostringnullableLast season.
seasonsobjectFirst and last seasons Kelly holds.
firststringnullableFirst.
laststringnullableLast.
positionstringnullablePosition.
bornstringnullableBirth date.
format date
competitionstringnullableA team's competition.
contextobjectA game's kickoff, week and teams.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled
GET https://api.askkelly.ai/v1/entities:resolve?q=Chiefs&kind=team
Authorization: Bearer $KELLY_API_KEY{
"object": "entities",
"resolved": "tm_0xcwddhkbjaqggrxg4xg",
"candidates": [
{
"id": "tm_0xcwddhkbjaqggrxg4xg",
"kind": "team",
"name": "Kansas City Chiefs",
"sport": "spt_bzcnprg8js5azynzmsyd",
"match": "prefix",
"score": 1,
"seasons": {
"first": "2026",
"last": "2026"
},
"competition": "nfl"
},
{
"id": "tm_r46ra6zqhcwmgn5c8b46",
"kind": "team",
"name": "Chicago Bears",
"sport": "spt_bzcnprg8js5azynzmsyd",
"match": "trigram",
"score": 0.429,
"seasons": {
"first": "2026",
"last": "2026"
},
"competition": "nfl"
},
{
"id": "tm_szswdq5mys2cx8dhz3ep",
"kind": "team",
"name": "Los Angeles Chargers",
"sport": "spt_bzcnprg8js5azynzmsyd",
"match": "trigram",
"score": 0.286,
"seasons": {
"first": "2026",
"last": "2026"
},
"competition": "nfl"
}
],
"ambiguity": null,
"summary": "Chiefs is Kansas City Chiefs (tm_0xcwddhkbjaqggrxg4xg).",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_98a43d8632dc3d0cb5394fd6",
"live_topics": [],
"rg_line": null
}GET /v1/metrics
public, no key needed · 1 credit, not charged during the beta · MCP tool describe_metrics · status live
A competition's stats from Kelly's catalogue: name, synonyms and slang, unit, definition, formula, version, grains, splits, defaults, qualifiers and coverage.
competitionstring · queryrequiredThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
qstring · queryWords for a stat, slang allowed: 'dimes', 'Ks', 'ATS'.
max length 80
metric_idsarray of string · queryStat IDs or names.
min items 1 · max items 8
entitystring · queryplayer or team.
One of: player, team (open: new values may be added).
includearray of string · queryAdd: coverage_summary, qualifiers, splits.
200 The metrics facts.
objectconstrequiredAlways metrics.
Always "metrics".
catalogue_versionstringrequiredThe catalogue's version.
metricsarray of objectrequiredThe stats.
metric_idstringrequired{competition}.{stat_key}.
namestringrequiredIts name.
synonymsarray of stringOther words for it, slang included.
unitstringnullableIts unit.
definitionstringrequiredWhat it counts.
formulastringnullableHow it's computed, for a derived stat.
versionstringrequiredIts definition's version.
dimensionsarray of stringrequiredFields filters may name besides stats.
splitsarray of stringrequiredSplits find_stats takes.
defaultsobjectrequiredDefaults applied when a call doesn't say.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · unknown_metric
GET https://api.askkelly.ai/v1/metrics?competition=nba&q=dimes
Authorization: Bearer $KELLY_API_KEY{
"object": "metrics",
"catalogue_version": "1.0.0",
"metrics": [
{
"metric_id": "nba.assists",
"name": "Assists",
"synonyms": [
"ast",
"dimes",
"apg",
"dishes"
],
"unit": "ast",
"definition": "Passes leading directly to a made field goal.",
"kind": "count",
"formula": null,
"inputs": [],
"version": "1.0.0",
"catalogue_version": "1.0.0",
"tolerance": 0,
"entities": [
"player",
"team"
],
"grains": [
"game",
"season",
"career"
],
"defaults": {
"season_type": "regular",
"since_year_includes": "season_starting",
"missing_is": "null",
"era_adjust": "when_careers_do_not_overlap",
"small_sample_n": 30
},
"plausible_range": {
"game": [
0,
60
]
},
"precedence": [
"balldontlie",
"kelly"
],
"rights": {
"field": "player.stats.game.*",
"class": "redistribute_derived",
"surfaces": [
"site",
"wire_first_party",
"api",
"mcp_plain",
"mcp_premium",
"mcp_cards"
]
},
"dont_confuse_with": [],
"better": "high",
"decimals": null,
"per": [
"game",
"season",
"rate"
]
}
],
"dimensions": [
"season",
"season_type",
"game_date",
"home",
"neutral",
"opponent_id",
"team_id",
"player_id",
"result",
"month",
"day_of_week",
"started",
"position",
"rest_days",
"back_to_back",
"career_game_number",
"age"
],
"splits": [
"home_away",
"opponent",
"month",
"result",
"season",
"day_of_week",
"started",
"rest"
],
"defaults": {
"season_type": "regular",
"since_year_includes": "season_starting",
"missing_is": "null",
"era_adjust": "when_careers_do_not_overlap",
"small_sample_n": 30
},
"summary": "1 nba stat in Kelly's catalogue (version 1.0.0) matching 'dimes'.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_734c32275ef84e3849106d3a",
"live_topics": [],
"rg_line": null
}GET /v1/coverage
public, no key needed · 1 credit, not charged during the beta · MCP tool get_coverage · status live
Which seasons each stat exists for and its status (official, unofficial, reconstructed, incomplete or absent), with games expected and present, so a missing stat is never read as zero.
competitionstring · queryrequiredThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
metric_idsarray of string · queryStat IDs or names; default all.
min items 1 · max items 8
from_seasonstring · queryFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstring · queryLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
entitystring · queryplayer or team; default both.
One of: player, team (open: new values may be added).
as_ofstring · queryAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstring · queryA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstring · queryconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitinteger · queryRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstring · queryA result's next value, for the next page.
season_typestring · queryregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The coverage facts.
objectconstrequiredAlways coverage.
Always "coverage".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
GET https://api.askkelly.ai/v1/coverage?competition=nba&metric_ids=points
Authorization: Bearer $KELLY_API_KEY{
"object": "coverage",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "season_label",
"label": "Season",
"type": "string"
},
{
"name": "metric_id",
"label": "Metric id",
"type": "string"
},
{
"name": "entity",
"label": "Entity",
"type": "string"
},
{
"name": "season_type",
"label": "Season type",
"type": "string"
},
{
"name": "status",
"label": "Status",
"type": "string"
},
{
"name": "games_expected",
"label": "Games expected",
"type": "string"
},
{
"name": "games_present",
"label": "Games present",
"type": "string"
}
],
"rows": [
[
"2025-26",
"nba.points",
"player",
"regular",
"official",
3,
3
],
[
"2025-26",
"nba.points",
"team",
"regular",
"incomplete",
3,
1
]
],
"row_count": 2,
"total_rows": 2
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"summary": "Coverage of 1 nba stat: 2 season rows, statuses incomplete, official.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_8c636e3a52e0a466b23480ad",
"live_topics": [],
"rg_line": null
}POST /v1/stats:find
public, no key needed · 1 credit, not charged during the beta · MCP tool find_stats · status live
A finder over every season Kelly holds: metrics at game, season or career grain for players or teams, with filters, a split, a sort and a limit, as columns plus rows with sources, coverage notes and the snapshot read. Ranked by the sort you choose.
Body, application/json (required):
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
entitystringplayer or team.
One of: player, team (open: new values may be added).
grainstringgame, season or career.
One of: game, season, career (open: new values may be added).
metricsarray of string1 to 8 stat IDs or names; games, starts, wins, losses and ties count rows. Teams also take points_for, points_against, point_diff, win_pct, the ATS and over/under records (ats_wins, ats_pct, overs, over_pct…) and opp_<key> for the opponent's figure.
min items 1 · max items 8
filtersarray of stringGame rows to keep before totals; 'active = true' keeps players or teams with a season in the latest season (false: retired), each '<field> <op> <value>' with op = != > >= < <= in between, e.g. 'passing_yards >= 300', 'home = false', 'spread between 6.5..7.5'.
max items 12
havingarray of stringTotals to keep after, same form, e.g. 'games >= 10'.
max items 12
splitstringA split from describe_metrics, e.g. home_away.
max length 32
entity_idsarray of stringOnly these players or teams (plr_, tm_).
max items 20
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
sortstringA metric, '-' first for most first; default the first metric, most first.
max length 64
min_sampleintegerOverride the catalogue's qualifier (a minimum of games).
min 0
metric_versionstringPin the catalogue's version (semver).
max length 16
as_ofstringAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstringA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstringconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitintegerRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstringA result's next value, for the next page.
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The stats facts.
objectconstrequiredAlways stats.
Always "stats".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
queryobjectThe query as run, normalised; send it back with a cursor to page.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/stats:find
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"entity": "player",
"grain": "career",
"metrics": [
"points"
],
"limit": 2
}
{
"object": "stats",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "player",
"label": "Player",
"type": "string"
},
{
"name": "points",
"label": "Points",
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0"
},
{
"name": "games",
"label": "Games",
"type": "integer"
}
],
"rows": [
[
"Pam Two",
900,
70
],
[
"Pat One",
96,
3
]
],
"row_count": 2,
"total_rows": 3
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"query": {
"competition": "nba",
"entity": "player",
"grain": "career",
"metrics": [
"nba.points"
],
"filters": [],
"having": [],
"split": null,
"entity_ids": [],
"from_season": null,
"to_season": null,
"sort": "-nba.points",
"min_sample": null,
"season_type": "regular",
"metric_version": null
},
"summary": "3 player rows for nba.points (career grain); first: Pam Two, as of 2026-11-06T00:00:00Z.",
"assumptions": [
"Regular season only (season_type=regular).",
"Every season Kelly covers for these stats.",
"Each season's line from the source Kelly serves first by precedence (balldontlie, kelly), else the sum of its games.",
"All seasons Kelly holds.",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total."
],
"has_more": true,
"next_cursor": "eyJvIjoyLCJzIjoic25hcF81MiJ9",
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_c99de431a1085e376fa11fcd",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/stats:find
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"entity": "player",
"grain": "decade",
"metrics": [
"points"
]
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "grain: 'decade' is not one of ['game', 'season', 'career']",
"param": "grain",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_b1df340af500bc88454d3e3c",
"retryable": false,
"display_message": "Something in that request was not understood."
}GET /v1/players/{player_id}/history
public, no key needed · 1 credit, not charged during the beta · MCP tool get_player_history · status live
A player's career totals, season lines, splits or game log, with team stints, each value with its source.
player_idstring · pathrequiredA player ID (plr_…) from resolve_entity.
viewstring · queryseasons (default), career, splits or game_log.
One of: seasons, career, splits, game_log (open: new values may be added).
metricsarray of string · queryStat IDs or names; default the catalogue's set for the position.
min items 1 · max items 8
splitstring · queryWith view splits, e.g. home_away.
max length 32
opponent_idstring · queryOnly games against this team (tm_…).
last_ninteger · queryWith view game_log: the last n games.
min 1 · max 500
filtersarray of string · queryRows to keep, each '<field> <op> <value>' with op = != > >= < <= in between, e.g. 'passing_yards >= 300', 'home = false', 'spread between 6.5..7.5'.
max items 12
from_seasonstring · queryFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstring · queryLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
as_ofstring · queryAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstring · queryA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstring · queryconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitinteger · queryRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstring · queryA result's next value, for the next page.
season_typestring · queryregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The player history facts.
objectconstrequiredAlways player_history.
Always "player_history".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
playerobjectThe player.
idstringKelly ID.
namestringName.
positionstringnullablePosition.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
GET https://api.askkelly.ai/v1/players/plr_znvh1bpfvdj2d79szye4/history
Authorization: Bearer $KELLY_API_KEY{
"object": "player_history",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "player",
"label": "Player",
"type": "string"
},
{
"name": "season",
"label": "Season",
"type": "string"
},
{
"name": "games",
"label": "Games",
"type": "integer"
},
{
"name": "games_played",
"label": "Games played",
"type": "integer",
"unit": "games",
"metric_id": "nba.games_played",
"metric_version": "1.0.0"
},
{
"name": "games_started",
"label": "Games started",
"type": "integer",
"unit": "games",
"metric_id": "nba.games_started",
"metric_version": "1.0.0"
},
{
"name": "minutes",
"label": "Minutes",
"type": "integer",
"unit": "min",
"metric_id": "nba.minutes",
"metric_version": "1.0.0"
},
{
"name": "points",
"label": "Points",
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0"
},
{
"name": "rebounds",
"label": "Rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.rebounds",
"metric_version": "1.0.0"
},
{
"name": "offensive_rebounds",
"label": "Offensive rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.offensive_rebounds",
"metric_version": "1.0.0"
},
{
"name": "defensive_rebounds",
"label": "Defensive rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.defensive_rebounds",
"metric_version": "1.0.0"
}
],
"rows": [
[
"Pat One",
"2025-26",
3,
3,
null,
114,
96,
10,
null,
null
]
],
"row_count": 1,
"total_rows": 1
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [
{
"metric_id": "nba.defensive_rebounds",
"seasons": "2025-26",
"status": "incomplete",
"text": "3 of 3 games have no value"
},
{
"metric_id": "nba.offensive_rebounds",
"seasons": "2025-26",
"status": "incomplete",
"text": "3 of 3 games have no value"
},
{
"metric_id": "nba.rebounds",
"seasons": "2025-26",
"status": "incomplete",
"text": "2 of 3 games have no value"
}
],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"player": {
"id": "plr_znvh1bpfvdj2d79szye4",
"name": "Pat One",
"position": "G"
},
"summary": "Pat One's seasons: 1 rows, as of 2026-11-06T00:00:00Z.",
"assumptions": [
"Regular season only (season_type=regular).",
"The catalogue's default stats for a nba player.",
"Every season Kelly covers for these stats.",
"Each season's line from the source Kelly serves first by precedence (balldontlie, kelly), else the sum of its games.",
"All seasons Kelly holds.",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_83516e3943bf7e73af4fff52",
"live_topics": [],
"rg_line": null
}GET /v1/teams/{team_id}/history
public, no key needed · 1 credit, not charged during the beta · MCP tool get_team_history · status live
A franchise's seasons (record, points, finish, playoffs, coach), head-to-head against one opponent, coaches or games, with the franchise's names and cities over time.
team_idstring · pathrequiredA team ID (tm_…) from resolve_entity.
viewstring · queryseasons (default), head_to_head, coaches or games.
One of: seasons, head_to_head, coaches, games (open: new values may be added).
opponent_idstring · queryThe opponent (tm_…), required with head_to_head.
from_seasonstring · queryFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstring · queryLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
as_ofstring · queryAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstring · queryA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstring · queryconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitinteger · queryRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstring · queryA result's next value, for the next page.
season_typestring · queryregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The team history facts.
objectconstrequiredAlways team_history.
Always "team_history".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
teamobjectThe team.
idstringKelly ID.
namestringName.
competitionstringnullableIts competition.
franchisearray of objectThe franchise's names and cities over time.
head_to_headobjectTotals against the opponent: games, wins, losses, ties, points for and against.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
GET https://api.askkelly.ai/v1/teams/tm_b4w5y9xy5f9sgg1cmkvy/history?view=head_to_head&opponent_id=tm_205kvyzpxcr3et677thx
Authorization: Bearer $KELLY_API_KEY{
"object": "team_history",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "season",
"label": "Season",
"type": "string"
},
{
"name": "games",
"label": "Games",
"type": "integer"
},
{
"name": "wins",
"label": "Wins",
"type": "integer"
},
{
"name": "losses",
"label": "Losses",
"type": "integer"
},
{
"name": "ties",
"label": "Ties",
"type": "integer"
},
{
"name": "score_for",
"label": "Score for",
"type": "integer"
},
{
"name": "score_against",
"label": "Score against",
"type": "integer"
}
],
"rows": [
[
"2025-26",
3,
2,
0,
1,
335,
319
]
],
"row_count": 1,
"total_rows": 1
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"team": {
"id": "tm_b4w5y9xy5f9sgg1cmkvy",
"name": "Home Hawks",
"competition": "nba"
},
"franchise": [
{
"name": "Home Hawks",
"from_season": "2025-26",
"to_season": "2025-26",
"games": 3
}
],
"head_to_head": {
"games": 3,
"wins": 2,
"losses": 0,
"ties": 1,
"score_for": 335,
"score_against": 319
},
"summary": "Home Hawks against that opponent: 2-0-1 in 3 games.",
"assumptions": [
"Regular season only (season_type=regular)."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_bb01d8a7b03b179cf1434175",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/teams/tm_b4w5y9xy5f9sgg1cmkvy/history
Authorization: Bearer $KELLY_API_KEY{
"object": "team_history",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "season",
"label": "Season",
"type": "string"
},
{
"name": "games",
"label": "Games",
"type": "integer"
},
{
"name": "wins",
"label": "Wins",
"type": "integer"
},
{
"name": "losses",
"label": "Losses",
"type": "integer"
},
{
"name": "ties",
"label": "Ties",
"type": "integer"
},
{
"name": "win_pct",
"label": "Win %",
"type": "number",
"unit": "fraction"
},
{
"name": "score_for",
"label": "Score for",
"type": "integer"
},
{
"name": "score_against",
"label": "Score against",
"type": "integer"
},
{
"name": "margin",
"label": "Margin",
"type": "integer"
}
],
"rows": [
[
"2025-26",
3,
2,
0,
1,
0.833,
335,
319,
16
]
],
"row_count": 1,
"total_rows": 1
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"team": {
"id": "tm_b4w5y9xy5f9sgg1cmkvy",
"name": "Home Hawks",
"competition": "nba"
},
"franchise": [
{
"name": "Home Hawks",
"from_season": "2025-26",
"to_season": "2025-26",
"games": 3
}
],
"summary": "Home Hawks's seasons: 1 rows, as of 2026-11-06T00:00:00Z.",
"assumptions": [
"Regular season only (season_type=regular)."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_a62840c10a1d308fe4c7a2fe",
"live_topics": [],
"rg_line": null
}GET /v1/games/{game_id}/history
public, no key needed · 1 credit, not charged during the beta · MCP tool get_game_history · status live
Any past game's line score, box score, closing lines and context.
game_idstring · pathrequiredA game ID (gm_…) from resolve_entity.
includearray of string · queryParts; default line_score and box_score.
as_ofstring · queryAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstring · queryA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstring · queryconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitinteger · queryRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstring · queryA result's next value, for the next page.
season_typestring · queryregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The game history facts.
objectconstrequiredAlways game_history.
Always "game_history".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
gameobjectThe game: name, start, status, competition, season, teams, final score, venue.
line_scoreobjectnullableA table for this part of the game.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
box_scoreobjectnullableA table for this part of the game.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
contextobjectnullableA table for this part of the game.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
linesobjectnullableA table for this part of the game.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
scoring_summaryarray of objectDerived scoring summaries; play rows stay internal.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
GET https://api.askkelly.ai/v1/games/gm_1hnx5nrmk4xjpz6fzyfz/history
Authorization: Bearer $KELLY_API_KEY{
"object": "game_history",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": null,
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"game": {
"game_id": "gm_1hnx5nrmk4xjpz6fzyfz",
"name": "Away Owls at Home Hawks",
"date": "2025-10-21",
"start_time": "2025-10-21T23:30:00Z",
"competition": "nba",
"season_label": "2025-26",
"season_type": "regular",
"neutral_site": false,
"home_team_id": "tm_b4w5y9xy5f9sgg1cmkvy",
"away_team_id": "tm_205kvyzpxcr3et677thx",
"home_team": "Home Hawks",
"away_team": "Away Owls",
"home_score": 110,
"away_score": 100
},
"box_score": {
"columns": [
{
"name": "competition_id",
"label": "Competition id",
"type": "string"
},
{
"name": "competition",
"label": "Competition",
"type": "string"
},
{
"name": "sport",
"label": "Sport",
"type": "string"
},
{
"name": "season_id",
"label": "Season id",
"type": "string"
},
{
"name": "season_label",
"label": "Season",
"type": "string"
},
{
"name": "season_type",
"label": "Season type",
"type": "string"
},
{
"name": "game_date",
"label": "Date",
"type": "date"
},
{
"name": "start_time",
"label": "Start time",
"type": "string"
},
{
"name": "home",
"label": "Home",
"type": "boolean"
},
{
"name": "neutral",
"label": "Neutral",
"type": "string"
},
{
"name": "result",
"label": "Result",
"type": "string"
},
{
"name": "started",
"label": "Started",
"type": "string"
},
{
"name": "valid_at",
"label": "Valid at",
"type": "string"
},
{
"name": "source_ts",
"label": "Source ts",
"type": "string"
},
{
"name": "recorded_at",
"label": "Recorded at",
"type": "string"
},
{
"name": "assumed_known_at",
"label": "Assumed known at",
"type": "string"
},
{
"name": "known_at",
"label": "Known at",
"type": "string"
},
{
"name": "minutes",
"label": "Minutes",
"type": "integer",
"unit": "min",
"metric_id": "nba.minutes",
"metric_version": "1.0.0"
},
{
"name": "points",
"label": "Points",
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0"
},
{
"name": "rebounds",
"label": "Rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.rebounds",
"metric_version": "1.0.0"
},
{
"name": "offensive_rebounds",
"label": "Offensive rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.offensive_rebounds",
"metric_version": "1.0.0"
},
{
"name": "defensive_rebounds",
"label": "Defensive rebounds",
"type": "integer",
"unit": "reb",
"metric_id": "nba.defensive_rebounds",
"metric_version": "1.0.0"
},
{
"name": "assists",
"label": "Assists",
"type": "integer",
"unit": "ast",
"metric_id": "nba.assists",
"metric_version": "1.0.0"
},
{
"name": "steals",
"label": "Steals",
"type": "integer",
"unit": "stl",
"metric_id": "nba.steals",
"metric_version": "1.0.0"
},
{
"name": "blocks",
"label": "Blocks",
"type": "integer",
"unit": "blk",
"metric_id": "nba.blocks",
"metric_version": "1.0.0"
},
{
"name": "turnovers",
"label": "Turnovers",
"type": "integer",
"unit": "tov",
"metric_id": "nba.turnovers",
"metric_version": "1.0.0"
},
{
"name": "fouls",
"label": "Personal fouls",
"type": "integer",
"unit": "pf",
"metric_id": "nba.fouls",
"metric_version": "1.0.0"
},
{
"name": "fg_made",
"label": "Field goals made",
"type": "integer",
"unit": "fg",
"metric_id": "nba.fg_made",
"metric_version": "1.0.0"
},
{
"name": "fg_attempted",
"label": "Field goals attempted",
"type": "integer",
"unit": "fga",
"metric_id": "nba.fg_attempted",
"metric_version": "1.0.0"
},
{
"name": "fg_pct",
"label": "Field goal percentage",
"type": "number",
"unit": "fraction",
"metric_id": "nba.fg_pct",
"metric_version": "1.0.0"
},
{
"name": "fg3_made",
"label": "Three-pointers made",
"type": "integer",
"unit": "fg3",
"metric_id": "nba.fg3_made",
"metric_version": "1.0.0"
},
{
"name": "fg3_attempted",
"label": "Three-pointers attempted",
"type": "integer",
"unit": "fg3a",
"metric_id": "nba.fg3_attempted",
"metric_version": "1.0.0"
},
{
"name": "fg3_pct",
"label": "Three-point percentage",
"type": "number",
"unit": "fraction",
"metric_id": "nba.fg3_pct",
"metric_version": "1.0.0"
},
{
"name": "ft_made",
"label": "Free throws made",
"type": "integer",
"unit": "ft",
"metric_id": "nba.ft_made",
"metric_version": "1.0.0"
},
{
"name": "ft_attempted",
"label": "Free throws attempted",
"type": "integer",
"unit": "fta",
"metric_id": "nba.ft_attempted",
"metric_version": "1.0.0"
},
{
"name": "ft_pct",
"label": "Free throw percentage",
"type": "number",
"unit": "fraction",
"metric_id": "nba.ft_pct",
"metric_version": "1.0.0"
},
{
"name": "plus_minus",
"label": "Plus-minus",
"type": "integer",
"unit": "pts",
"metric_id": "nba.plus_minus",
"metric_version": "1.0.0"
},
{
"name": "effective_fg_pct",
"label": "Effective field goal percentage",
"type": "number",
"unit": "fraction",
"metric_id": "nba.effective_fg_pct",
"metric_version": "1.0.0"
},
{
"name": "true_shooting_pct",
"label": "True shooting percentage",
"type": "number",
"unit": "fraction",
"metric_id": "nba.true_shooting_pct",
"metric_version": "1.0.0"
},
{
"name": "assist_to_turnover",
"label": "Assist-to-turnover ratio",
"type": "number",
"unit": "ratio",
"metric_id": "nba.assist_to_turnover",
"metric_version": "1.0.0"
}
],
"rows": [
[
"cmp_mm0mp7dsd2retmqgfps5",
"nba",
"basketball",
"ssn_7zj2aaf4knqg0p3bzrj8",
"2025-26",
"regular",
"2025-10-21",
"2025-10-21T23:30:00Z",
false,
false,
"L",
true,
"2025-10-21T23:30:00Z",
"2025-10-21T23:30:00Z",
"2025-11-01T09:00:00Z",
null,
"2025-11-01T09:00:00Z",
40,
28,
null,
null,
null,
null,
null,
null,
null,
null,
10,
22,
null,
null,
null,
null,
null,
6,
null,
null,
null,
null,
null
],
[
"cmp_mm0mp7dsd2retmqgfps5",
"nba",
"basketball",
"ssn_7zj2aaf4knqg0p3bzrj8",
"2025-26",
"regular",
"2025-10-21",
"2025-10-21T23:30:00Z",
true,
false,
"W",
true,
"2025-10-21T23:30:00Z",
"2025-10-21T23:30:00Z",
"2025-11-01T09:00:00Z",
null,
"2025-11-01T09:00:00Z",
30,
12,
null,
null,
null,
null,
null,
null,
null,
null,
5,
9,
null,
null,
null,
null,
null,
2,
null,
null,
null,
null,
null
],
[
"cmp_mm0mp7dsd2retmqgfps5",
"nba",
"basketball",
"ssn_7zj2aaf4knqg0p3bzrj8",
"2025-26",
"regular",
"2025-10-21",
"2025-10-21T23:30:00Z",
true,
false,
"W",
true,
"2025-10-21T23:30:00Z",
"2025-10-21T23:30:00Z",
"2025-11-01T09:00:00Z",
null,
"2025-11-01T09:00:00Z",
36,
30,
10,
null,
null,
null,
null,
2,
null,
null,
11,
20,
null,
null,
null,
null,
null,
8,
null,
null,
null,
null,
null
]
],
"row_count": 3,
"total_rows": 3
},
"summary": "Away Owls at Home Hawks on 2025-10-21: 100-110.",
"assumptions": [
"Kelly's lake doesn't hold the line score yet."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_6a3ef0d12db67b8853eda7ce",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/games/gm_zzzzzzzzzzzzzzzzzzzz/history
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No game with that ID; resolve_entity finds one by teams and date.",
"param": "game_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_b702bda388f13c4c3ea39778",
"retryable": false,
"display_message": "That could not be found."
}GET /v1/records
public, no key needed · 1 credit, not charged during the beta · MCP tool get_records · status live
All-time (career), season and single-game records for a stat, with the qualifier applied; withheld when the span has games without the stat.
competitionstring · queryrequiredThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
metric_idstring · queryrequiredThe stat's ID or name.
min length 1 · max length 64
scopestring · queryrequiredsingle_game, season or career.
One of: single_game, season, career (open: new values may be added).
entitystring · queryplayer (default) or team.
One of: player, team (open: new values may be added).
filtersarray of string · queryRows to keep, each '<field> <op> <value>' with op = != > >= < <= in between, e.g. 'passing_yards >= 300', 'home = false', 'spread between 6.5..7.5'.
max items 12
from_seasonstring · queryFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstring · queryLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
min_sampleinteger · queryOverride the qualifier.
min 0
active_onlyboolean · queryOnly players or teams active now.
as_ofstring · queryAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstring · queryA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstring · queryconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitinteger · queryRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstring · queryA result's next value, for the next page.
season_typestring · queryregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The records facts.
objectconstrequiredAlways records.
Always "records".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
completebooleanEvery game in the span has the stat.
qualifierstringnullableThe qualifier applied.
withheld_reasonstringnullableWhy no records are given.
follow_upsarray of objectCalls that would answer.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
GET https://api.askkelly.ai/v1/records?competition=nba&metric_id=points&scope=single_game
Authorization: Bearer $KELLY_API_KEY{
"object": "records",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "player",
"label": "Player",
"type": "string"
},
{
"name": "season",
"label": "Season",
"type": "string"
},
{
"name": "game_date",
"label": "Date",
"type": "date"
},
{
"name": "opponent",
"label": "Opponent",
"type": "string"
},
{
"name": "home",
"label": "Home",
"type": "boolean"
},
{
"name": "result",
"label": "Result",
"type": "string"
},
{
"name": "points",
"label": "Points",
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0"
},
{
"name": "games",
"label": "Games",
"type": "integer"
}
],
"rows": [
[
"Pat One",
"2025-26",
"2025-10-25",
"Away Owls",
true,
"T",
41,
1
],
[
"Pip Three",
"2025-26",
"2025-10-22",
"Home Hawks",
true,
"L",
31,
1
],
[
"Pat One",
"2025-26",
"2025-10-21",
"Away Owls",
true,
"W",
30,
1
],
[
"Pip Three",
"2025-26",
"2025-10-21",
"Home Hawks",
false,
"L",
28,
1
],
[
"Pat One",
"2025-26",
"2025-10-22",
"Away Owls",
false,
"W",
25,
1
],
[
"Pam Two",
"2025-26",
"2025-10-21",
"Away Owls",
true,
"W",
12,
1
]
],
"row_count": 6,
"total_rows": 6
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": null,
"read_from": "lake",
"complete": true,
"qualifier": null,
"summary": "Single game Points records (nba): first Pat One, as of 2026-11-06T00:00:00Z.",
"assumptions": [
"Regular season only (season_type=regular).",
"Every season Kelly covers for these stats.",
"One row per game: the source Kelly serves first by precedence (balldontlie, kelly).",
"All seasons Kelly holds.",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_e58a33b9a51abe061e391757",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/records?competition=nba&metric_id=dunks_per_moon&scope=season
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/unknown_metric",
"title": "Unknown stat",
"status": 422,
"code": "unknown_metric",
"detail": "'dunks_per_moon' isn't a nba stat in Kelly's catalogue. describe_metrics lists them.",
"param": "metric_id",
"doc_url": "https://docs.askkelly.ai/problems/unknown_metric",
"request_id": "req_28fec2a8301ca00dcff899ca",
"retryable": false,
"display_message": "Kelly doesn't track that stat."
}POST /v1/lines:history
public, no key needed · 1 credit, not charged during the beta · MCP tool get_line_history · status live
How a kind of bet did over past games (against the spread, over/under or straight up), with the sample size, a 95% Wilson interval, a baseline and how many filters made it.
Body, application/json (required):
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
marketstringspread, total or moneyline.
One of: spread, total, moneyline (open: new values may be added).
sidestringfavourite, underdog, home, away, over or under.
One of: favourite, underdog, home, away, over, under (open: new values may be added).
team_idstringOnly this team's games (tm_…).
linestringclose (default) or open.
One of: open, close (open: new values may be added).
filtersarray of stringGames to keep (spread, total, price, rest_days, divisional, primetime...), each '<field> <op> <value>' with op = != > >= < <= in between, e.g. 'passing_yards >= 300', 'home = false', 'spread between 6.5..7.5'.
max items 12
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
from_datestringFirst date (YYYY-MM-DD).
format date
to_datestringLast date.
format date
include_gamesbooleanAdd the games, one row each.
as_ofstringAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstringA snapshot_id an earlier history result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
response_formatstringconcise (default: key columns, 25 rows) or detailed (IDs, every column, up to 200 rows).
One of: concise, detailed (open: new values may be added).
limitintegerRows per page, 1 to 200; default 25.
min 1 · max 200
cursorstringA result's next value, for the next page.
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The line history facts.
objectconstrequiredAlways line_history.
Always "line_history".
snapshot_idstringrequirednullableThe lake snapshot read; pass it back to read the same data.
pattern ^snap_[0-9]{1,18}$
read_fromstringlake, lake+neon or neon (the fallback).
One of: lake, lake+neon, neon (open: new values may be added).
data_throughobjectrequiredThe newest final game included, per sport.
stalebooleanrequiredA freshness target was missed.
tableobjectrequirednullableRows as columns plus rows, units in the columns. Concise results leave out ID columns; detailed ones keep them.
columnsarray of objectrequiredThe columns, in row order.
namestringrequiredThe column's name.
labelstringrequiredIts label, as a person reads it.
typestringrequiredstring, integer, number, boolean, date or id.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringThe stat's unit (yd, pts, fraction).
metric_idstringThe stat's catalogue ID ({competition}.{stat_key}).
metric_versionstringThe stat definition's version (semver).
source_keystringThe key in sources its values come from.
rowsarray of arrayrequiredThe rows on this page.
row_countintegerrequiredRows on this page.
total_rowsintegerrequiredRows in the whole result.
withheldarray of objectColumns this surface may not carry.
columnstringrequiredIts name.
reasonstringrequiredWhy: the Rights Register entry that keeps it off this surface.
sourcesarray of objectrequiredEvery source the values came from, by precedence.
keystringrequiredThe key columns point at (s1, s2...).
sourcestringrequiredThe source's name in Kelly's Rights Register.
licencestringrequiredIts licence or terms, in short.
attributionstringnullableThe credit line it requires.
snapshot_idstringrequirednullableThe lake snapshot read.
pattern ^snap_[0-9]{1,18}$
recorded_atstringrequirednullableWhen Kelly recorded the values.
format date-time
officialbooleanrequiredThe league's official figures.
reconstructedbooleanrequiredRebuilt from play-by-play rather than a box score.
disagreesbooleanrequiredAnother source differs beyond tolerance.
precedenceintegerrequiredIts rank when sources disagree; 1 is served.
coverage_notesarray of objectrequiredWhat's missing or unofficial in the span.
metric_idstringrequiredThe stat's catalogue ID.
seasonsstringrequiredThe seasons, one or a range.
statusstringrequiredabsent, incomplete, unofficial, reconstructed or tier_locked.
One of: official, unofficial, reconstructed, incomplete, absent, tier_locked (open: new values may be added).
textstringrequiredOne sentence to state in the answer.
ambiguityobjectrequirednullableSet when the input could mean more than one thing: no rows, and the question to ask the person.
paramstringrequiredThe input that's unclear.
questionstringrequiredThe question to ask.
candidatesarray of objectrequiredThe choices.
idstringIts ID.
labelstringHow a person tells it apart.
sampleobjectrequirednullableA trend's sample: the count, its 95% Wilson interval, a baseline and how many filters made it.
nintegerrequiredGames, pushes included.
winsintegerrequiredWins (covers, overs).
lossesintegerrequiredLosses.
pushesintegerrequiredPushes.
ratenumberrequirednullableWins over decided games.
ci_lownumberrequirednullableThe interval's low end.
ci_highnumberrequirednullableThe interval's high end.
ci_methodconstrequiredAlways wilson_95.
Always "wilson_95".
baselineobjectrequirednullableThe same kind of bet with no filters, over the same seasons.
labelstringrequiredWhat the baseline counts.
nintegerrequiredDecided games in it.
ratenumberrequirednullableIts rate.
filters_appliedintegerrequiredFilters that narrowed the sample.
small_samplebooleanrequiredToo few games to read much into.
queryobjectThe trend's filters as run.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/lines:history
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"market": "spread",
"side": "home",
"include_games": true
}
{
"object": "line_history",
"snapshot_id": "snap_52",
"data_through": {},
"stale": false,
"table": {
"columns": [
{
"name": "game_date",
"label": "Date",
"type": "date"
},
{
"name": "season",
"label": "Season",
"type": "string"
},
{
"name": "home",
"label": "Home",
"type": "boolean"
},
{
"name": "points_for",
"label": "Points for",
"type": "integer",
"unit": "pts"
},
{
"name": "points_against",
"label": "Points against",
"type": "integer",
"unit": "pts"
},
{
"name": "spread_close",
"label": "Closing spread",
"type": "number"
},
{
"name": "total_close",
"label": "Closing total",
"type": "number"
},
{
"name": "result",
"label": "Result",
"type": "string"
},
{
"name": "ats_result",
"label": "Against the spread",
"type": "string"
},
{
"name": "ou_result",
"label": "Over/under",
"type": "string"
},
{
"name": "graded",
"label": "Graded",
"type": "string"
}
],
"rows": [
[
"2025-10-25",
"2025-26",
true,
120,
120,
null,
null,
"T",
null,
null,
null
],
[
"2025-10-22",
"2025-26",
true,
99,
105,
null,
null,
"L",
null,
null,
null
],
[
"2025-10-21",
"2025-26",
true,
110,
100,
-4.5,
215.5,
"W",
"cover",
"under",
"win"
]
],
"row_count": 3,
"total_rows": 3
},
"sources": [
{
"key": "s1",
"source": "kelly",
"licence": "Kelly's own computation and metadata",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"ambiguity": null,
"sample": {
"n": 1,
"wins": 1,
"losses": 0,
"pushes": 0,
"rate": 1,
"ci_low": 0.2065493143772375,
"ci_high": 1,
"ci_method": "wilson_95",
"baseline": {
"label": "Every nba close spread (home), same seasons",
"n": 1,
"rate": 1
},
"filters_applied": 0,
"small_sample": true
},
"read_from": "lake",
"query": {
"competition": "nba",
"market": "spread",
"side": "home",
"team_id": null,
"from_season": null,
"to_season": null,
"from_date": null,
"to_date": null,
"season_type": "regular",
"spread_min": null,
"spread_max": null,
"total_min": null,
"total_max": null,
"rest_min": null,
"rest_max": null,
"home": null,
"back_to_back": null
},
"summary": "1 of 1 covered (100.0%, 95% interval 20.7% to 100.0%); baseline 100.0% — a small sample.",
"assumptions": [
"Regular season only (season_type=regular).",
"Closing lines (Kelly's consensus close)."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_d8d33be84363c069528ba7ea",
"live_topics": [],
"rg_line": null
}POST /v1/views:query
public, no key needed · 1 credit, not charged during the beta · MCP tool get_view · status live
One table from a view spec (registry/contracts/views/view-spec.schema.json): players or teams by game, season or career with the columns, filters, totals tests, split, sort and mode asked for; rank, percentile, sample size, small-sample flags and 95% intervals beside each number; total, per-game, league-average and difference rows computed; streaks and best stretches; sources, coverage notes and the snapshot read.
Body, application/json (required):
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
columnsarray of stringStat keys (passing_yards) or dimensions (opponent, game_date), in display order, 1 to 40. Teams also take points_for, points_against, point_diff, wins, losses, ties, win_pct, ats_wins, ats_losses, ats_pushes, ats_pct, overs, unders, ou_pushes, over_pct and the opponent's figure for any team stat as opp_<key> (opp_passing_yards).
min items 1 · max items 40
calcsarray of stringUp to 20 formula columns over the rows, each 'key:formula', e.g. 'ypa:passing_yards / passing_attempts'; 'key~Label:formula' names its header. A formula reads totals; a stat it names that isn't a column is read for it. list_functions has the functions. Sort by a calc's key.
max items 20
kindstringtable (default), streak or span.
One of: table, streak, span (open: new values may be added).
entitystringplayer (default) or team.
One of: player, team (open: new values may be added).
grainstringgame, season (default) or career.
One of: game, season, career (open: new values may be added).
entity_idsarray of stringUp to 20 players or teams (plr_, tm_); none ranks every one (a finder).
max items 20
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
sortarray of stringUp to 3 columns, '-' first for highest first.
max items 3
filtersarray of stringGames to keep, each '<field> <op> <value>' (op = != > >= < <= in between), e.g. 'passing_yards >= 300', 'home = false'.
max items 12
havingarray of stringTests on a row's own totals, e.g. 'passing_yards >= 4000' for 4,000-yard seasons.
max items 8
splitstringA split from describe_metrics, e.g. home_away.
max length 32
modestringtot (default), pg (per game), season (per full season) or p100 (per 100 of the rate base).
One of: tot, pg, season, p100 (open: new values may be added).
contextstringnone (default), rank or pct beside each stat.
One of: none, rank, pct (open: new values may be added).
heatstringoff (default), league or rows.
One of: off, league, rows (open: new values may be added).
diffarray of stringTwo row keys: adds a difference row.
min items 2 · max items 2
conditionarray of stringkind streak: up to 4 tests every game must meet.
max items 4
lengthintegerkind span: games in the stretch, 2 to 162.
min 2 · max 162
min_sampleintegerOverride the qualifier for the comparison set (a minimum of games).
min 1 · max 10000
limitintegerRows per page, 1 to 200; default 50.
min 1 · max 200
cursorstringA result's next value, for the next page.
max length 64
as_ofstringAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstringA snapshot_id an earlier result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
200 The view facts.
objectconstrequiredAlways view.
Always "view".
vconstThe view contract version.
Always 1.
default 1
specobjectrequiredThe spec as the engine read it (defaults filled, filters normalised).
vconstThe contract version.
Always 1.
default 1
kindstringtable: rows of stats. streak: runs of games in a row that meet every condition. span: the best stretch of length games by the first sort column.
One of: table, streak, span (open: new values may be added).
default "table"
competitionstringrequiredThe competition slug, e.g. nfl, nba, ncaaf.
entitystringRows are players or teams; in racing (horse_racing, greyhounds, harness) runners, horses, jockeys, trainers or tracks.
One of: player, team, runner, horse, jockey, trainer, track (open: new values may be added).
default "player"
grainstringOne row per game, per season, or per career.
One of: game, season, career (open: new values may be added).
default "season"
entity_idsarray of stringUp to 20 IDs: plr_ or tm_; in racing rnr_ (runner), anm_ (horse), rpn_ (jockey, trainer) or ven_ (track). Empty: every player or team (a finder).
default []
from_seasonstringnullableFirst season, a label like 2025 or 2025-26.
default null
to_seasonstringnullableLast season.
default null
season_typestringregular, postseason or both.
One of: regular, postseason, both (open: new values may be added).
default "regular"
columnsarray of stringrequiredCatalogue stat keys (passing_yards) or dimensions (opponent), in display order. 1 to 40.
calcsarray of objectUp to 20 formula columns over the view's rows, shown after the columns. Each is an object or its text 'key[.mode][~label]:formula'. Errors come back in calc_errors, not as a refusal.
default []
keystringrequiredThe calc's column key: lowercase letters, digits and _, not one of the columns. Sort by it like any column.
formulastringrequiredThe formula, e.g. passing_yards / passing_attempts or ROLLING_AVG(passing_yards, 3). Excel syntax unless mode says otherwise.
labelstringnullableThe column header; the formula when null. No ':' (the URL form uses it).
default null
modestringThe formula's input language (ADR-K591 D1): excel unless the formula is written in another.
One of: excel, sheets, sql, polars, pandas, dplyr, dax, tableau, pine (open: new values may be added).
default "excel"
sortarray of stringUp to 3 columns, '-' first for highest first. Multi-sort in order.
default []
filtersarray of stringUp to 12 filters in the history grammar: '<field> <op> <value>', op one of = != < <= > >= in between. 'active = true' keeps players or teams with a season in the latest season ('active = false': retired).
default []
havingarray of stringUp to 8 tests on a row's own totals, after grouping: 'passing_yards >= 4000' keeps 4,000-yard seasons (filters test each game).
default []
splitstringnullableA catalogue split (home_away, opponent, month…): one row per split value.
default null
modestringtot: totals. pg: per game. season: per full season (17 games in the NFL, 82 in the NBA; the engine names it). p100: per 100 of the competition's rate base (dropbacks, possessions; the engine names it).
One of: tot, pg, season, p100 (open: new values may be added).
default "tot"
contextstringBeside each stat: nothing, its rank, or its percentile, in the comparison set the result names.
One of: none, rank, pct (open: new values may be added).
default "none"
heatstringHeat position for each stat cell: off, against the league comparison set, or against the other rows shown.
One of: off, league, rows (open: new values may be added).
default "off"
diffarray of stringTwo row keys: adds a difference row (first minus second).
default []
conditionarray of stringkind=streak: up to 4 filters each game must meet.
default []
lengthintegernullablekind=span: games in the stretch, 2 to 162.
default null
min_sampleintegernullableOverrides the catalogue's qualifier for the comparison set.
default null
limitintegerRows per page, 1 to 200.
default 50
cursorstringnullableA result's next value, unchanged.
default null
as_ofstringnullableRead the data as of this time (ISO 8601).
format date-time · default null
snapshot_idstringnullableRead one lake snapshot, e.g. snap_1234.
default null
querystringrequiredThe spec's readable URL query against the page's base view.
linkstringnullableThe view on askkelly.ai.
default null
canonicalstringnullableThe base page path: the canonical link (G3).
default null
mode_labelstringrequiredTotals, Per game, Per 17, Per 100 dropbacks…
columnsarray of objectrequiredThe columns, in the spec's order.
keystringrequiredThe column key from the spec.
labelstringrequiredIts label, as a person reads it.
shortstringnullableA short header (Yds, Cmp%), when the catalogue has one.
default null
typestringrequiredThe value type of its cells.
One of: string, integer, number, boolean, date, id (open: new values may be added).
unitstringnullableThe unit (yards, percent…), when the catalogue has one.
default null
metric_idstringnullableThe catalogue ID ({competition}.{key}); null for a dimension.
default null
metric_versionstringnullableThe catalogue version of the metric's definition.
default null
kindstringnullableThe catalogue kind: count, measure, ratio, rate, derived…
default null
scaledbooleanTrue when the mode changes it (a count shown per game).
default false
betterone of several shapesWhich way is better, for rank, heat and sort arrows; null when neither.
default null
denominatorstringnullableThe column key of its count (attempts behind completion %).
default null
definitionstringnullableThe catalogue's one-line definition, for the header menu.
default null
lockedbooleanTrue when the Rights Register withholds it on this surface: no values.
default false
locked_reasonstringnullableWhy the column is locked, as a person reads it.
default null
no_exportbooleanTrue when the page may show it but it never goes into a download, a cell copy or a bulk export (per-book prices under licence); set from the Rights Register, and the API and MCP apply the same rule to their exports.
default false
coveragestringnullableA coverage note's text when some seasons have no value.
default null
formulastringnullableThe catalogue's formula over summed parts (passing_completions / passing_attempts), for 'Explain this number'.
default null
sourcestringnullableThe key in the result's sources its values come from, when one source serves the whole column.
default null
sincestringnullableThe first season with values in Kelly's history (from coverage).
default null
sortablebooleanFalse when the column can't be sorted or filtered on (a calc built on an edge, D16).
default true
decimalsone of several shapesFraction digits to show: 3 for a fraction or an average, 1 for a per-600 rate, 2 for a count shown per game or per season; null to show the value as sent. Display only: copies and downloads keep the full value.
default null
rowsarray of objectrequiredThis page of rows, in the spec's sort.
keystringrequiredStable row key (plr_x, plr_x:2024, gm_y, a split value); diff names two.
labelstringnullableSummary rows: Total, Per game, Per 17, League average, Difference.
default null
cellsarray of objectrequiredOne value per column, in the result's column order; null = no value (never zero).
ctxone of several shapesOne context per cell, or null when none apply.
default null
decimalsone of several shapesFraction digits per cell where a summary row is in another mode than the table (a Per game row under totals shows 2, a Total row under per game shows 0); a null cell, or a row without the list, follows its column's decimals.
default null
hrefstringnullableThe row's own page on askkelly.ai (/players/josh-allen, a game page), when it has one.
default null
groupstringnullableThe group the row belongs to (a season in a game log, a split family); a page keeps sorting within groups.
default null
summary_rowsarray of objectrequiredSummary rows, keys from total, per_game, per_season, league, diff, in that order, each only when it applies.
keystringrequiredStable row key (plr_x, plr_x:2024, gm_y, a split value); diff names two.
labelstringnullableSummary rows: Total, Per game, Per 17, League average, Difference.
default null
cellsarray of objectrequiredOne value per column, in the result's column order; null = no value (never zero).
ctxone of several shapesOne context per cell, or null when none apply.
default null
decimalsone of several shapesFraction digits per cell where a summary row is in another mode than the table (a Per game row under totals shows 2, a Total row under per game shows 0); a null cell, or a row without the list, follows its column's decimals.
default null
hrefstringnullableThe row's own page on askkelly.ai (/players/josh-allen, a game page), when it has one.
default null
groupstringnullableThe group the row belongs to (a season in a game log, a split family); a page keeps sorting within groups.
default null
total_rowsintegerrequiredRows matching the view across every page.
nextstringnullablePass as cursor for the next page; null on the last page.
default null
compare_setone of several shapesThe set the rank and heat compare against.
default null
sourcesarray of objectrequiredhistory.shape Source objects: licence, attribution, precedence.
coverage_notesarray of objecthistory.shape coverage notes: seasons or fields with gaps.
calc_errorsarray of objectOne entry per calc that couldn't be computed; its column is kept with null cells.
keystringrequirednullableThe calc's key; null for the whole set (too many calcs).
codestringrequiredThe formula language's error code: syntax, not_in_view, type, shape, budget…
messagestringrequiredWhat went wrong, as a person reads it.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
snapshot_idstringnullableThe lake snapshot the view was read from.
default null
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
data_throughobjectThe last date each source covers.
stalebooleanTrue when a source is behind its schedule.
default false
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · unknown_metric · invalid_filter · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/views:query
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"entity": "player",
"grain": "career",
"columns": [
"points"
],
"calcs": [
"ppg~Points per game:points / games"
],
"sort": [
"-ppg"
],
"limit": 2
}
{
"object": "view",
"v": 1,
"spec": {
"v": 1,
"kind": "table",
"competition": "nba",
"entity": "player",
"grain": "career",
"entity_ids": [],
"from_season": null,
"to_season": null,
"season_type": "regular",
"columns": [
"points"
],
"calcs": [
{
"key": "ppg",
"formula": "points / games",
"label": "Points per game",
"mode": "excel"
}
],
"sort": [
"-ppg"
],
"filters": [],
"having": [],
"split": null,
"mode": "tot",
"context": "none",
"heat": "off",
"diff": [],
"condition": [],
"length": null,
"min_sample": null,
"limit": 2,
"cursor": null,
"as_of": null,
"snapshot_id": null
},
"query": "g=career&calc=ppg~Points%20per%20game:points%20%2F%20games&sort=-ppg&n=2",
"link": null,
"canonical": null,
"mode_label": "Totals",
"columns": [
{
"key": "points",
"label": "Points",
"short": null,
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0",
"kind": "count",
"scaled": false,
"better": "high",
"denominator": null,
"definition": "Points scored: 2 a field goal inside the arc, 3 beyond it, 1 a free throw.",
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": "s1",
"since": "1946-47",
"sortable": true,
"decimals": null
},
{
"key": "ppg",
"label": "Points per game",
"short": null,
"type": "number",
"unit": "pts/games",
"metric_id": null,
"metric_version": null,
"kind": "calc",
"scaled": false,
"better": null,
"denominator": null,
"definition": "Your formula: points / games",
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": "points / games",
"source": null,
"since": null,
"sortable": true,
"decimals": null
}
],
"rows": [
{
"key": "plr_znvh1bpfvdj2d79szye4",
"label": "Pat One",
"cells": [
96,
32
],
"ctx": null,
"href": "/players/pat-one-znvh1bpfvdj2d79szye4",
"group": null
},
{
"key": "plr_d9b91wfaa4ym1y788aq8",
"label": "Pip Three",
"cells": [
59,
29.5
],
"ctx": null,
"href": "/players/pip-three-d9b91wfaa4ym1y788aq8",
"group": null
}
],
"summary_rows": [],
"total_rows": 3,
"next": "eyJvIjoyLCJzIjoic25hcF81MiJ9",
"compare_set": null,
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"calc_errors": [],
"assumptions": [
"Each season's line from the source Kelly serves first by precedence (balldontlie, kelly), else the sum of its games.",
"All seasons Kelly holds.",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total.",
"Read for the calcs, not shown: games."
],
"snapshot_id": "snap_52",
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"stale": false,
"summary": "3 player career rows, most points per game first (totals): Pat One, 32.",
"has_more": true,
"next_cursor": "eyJvIjoyLCJzIjoic25hcF81MiJ9",
"source": "kelly",
"attribution": null,
"request_id": "req_66e61ed3eefabd8d1d36be88",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/views:query
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"entity": "player",
"grain": "career",
"columns": [
"points",
"games"
],
"sort": [
"-points"
],
"limit": 2
}
{
"object": "view",
"v": 1,
"spec": {
"v": 1,
"kind": "table",
"competition": "nba",
"entity": "player",
"grain": "career",
"entity_ids": [],
"from_season": null,
"to_season": null,
"season_type": "regular",
"columns": [
"points",
"games"
],
"calcs": [],
"sort": [
"-points"
],
"filters": [],
"having": [],
"split": null,
"mode": "tot",
"context": "none",
"heat": "off",
"diff": [],
"condition": [],
"length": null,
"min_sample": null,
"limit": 2,
"cursor": null,
"as_of": null,
"snapshot_id": null
},
"query": "g=career&sort=-points&n=2",
"link": null,
"canonical": null,
"mode_label": "Totals",
"columns": [
{
"key": "points",
"label": "Points",
"short": null,
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0",
"kind": "count",
"scaled": false,
"better": "high",
"denominator": null,
"definition": "Points scored: 2 a field goal inside the arc, 3 beyond it, 1 a free throw.",
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": "s1",
"since": "1946-47",
"sortable": true,
"decimals": null
},
{
"key": "games",
"label": "Games",
"short": null,
"type": "integer",
"unit": null,
"metric_id": null,
"metric_version": null,
"kind": null,
"scaled": false,
"better": null,
"denominator": null,
"definition": null,
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": null,
"since": null,
"sortable": true,
"decimals": null
}
],
"rows": [
{
"key": "plr_emrxad012xnt6nbe4vy4",
"label": "Pam Two",
"cells": [
900,
70
],
"ctx": null,
"href": "/players/pam-two-emrxad012xnt6nbe4vy4",
"group": null
},
{
"key": "plr_znvh1bpfvdj2d79szye4",
"label": "Pat One",
"cells": [
96,
3
],
"ctx": null,
"href": "/players/pat-one-znvh1bpfvdj2d79szye4",
"group": null
}
],
"summary_rows": [],
"total_rows": 3,
"next": "eyJvIjoyLCJzIjoic25hcF81MiJ9",
"compare_set": null,
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"calc_errors": [],
"assumptions": [
"Each season's line from the source Kelly serves first by precedence (balldontlie, kelly), else the sum of its games.",
"All seasons Kelly holds.",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total."
],
"snapshot_id": "snap_52",
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"stale": false,
"summary": "3 player career rows, most points first (totals): Pam Two, 900.",
"has_more": true,
"next_cursor": "eyJvIjoyLCJzIjoic25hcF81MiJ9",
"source": "kelly",
"attribution": null,
"request_id": "req_257348adf3bb68434c3ddc93",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/views:query
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"grain": "game",
"columns": [
"points"
]
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "every game of every player is too broad: add entity_ids, a filter or from_season",
"param": "spec",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_a7a069fff56f97236196e9db",
"retryable": false,
"display_message": "Something in that request was not understood."
}POST https://api.askkelly.ai/v1/views:query
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"entity": "player",
"grain": "season",
"columns": [
"season",
"points",
"games"
],
"sort": [
"-points"
],
"context": "rank",
"min_sample": 1
}
{
"object": "view",
"v": 1,
"spec": {
"v": 1,
"kind": "table",
"competition": "nba",
"entity": "player",
"grain": "season",
"entity_ids": [],
"from_season": null,
"to_season": null,
"season_type": "regular",
"columns": [
"season",
"points",
"games"
],
"calcs": [],
"sort": [
"-points"
],
"filters": [],
"having": [],
"split": null,
"mode": "tot",
"context": "rank",
"heat": "off",
"diff": [],
"condition": [],
"length": null,
"min_sample": 1,
"limit": 50,
"cursor": null,
"as_of": null,
"snapshot_id": null
},
"query": "sort=-points&ctx=rank&min=1",
"link": null,
"canonical": null,
"mode_label": "Totals",
"columns": [
{
"key": "season",
"label": "Season",
"short": null,
"type": "string",
"unit": null,
"metric_id": null,
"metric_version": null,
"kind": null,
"scaled": false,
"better": null,
"denominator": null,
"definition": null,
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": null,
"since": null,
"sortable": true,
"decimals": null
},
{
"key": "points",
"label": "Points",
"short": null,
"type": "integer",
"unit": "pts",
"metric_id": "nba.points",
"metric_version": "1.0.0",
"kind": "count",
"scaled": false,
"better": "high",
"denominator": null,
"definition": "Points scored: 2 a field goal inside the arc, 3 beyond it, 1 a free throw.",
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": "s1",
"since": "1946-47",
"sortable": true,
"decimals": null
},
{
"key": "games",
"label": "Games",
"short": null,
"type": "integer",
"unit": null,
"metric_id": null,
"metric_version": null,
"kind": null,
"scaled": false,
"better": null,
"denominator": null,
"definition": null,
"locked": false,
"locked_reason": null,
"no_export": false,
"coverage": null,
"formula": null,
"source": null,
"since": null,
"sortable": true,
"decimals": null
}
],
"rows": [
{
"key": "plr_emrxad012xnt6nbe4vy4:2025-26",
"label": "Pam Two",
"cells": [
"2025-26",
900,
70
],
"ctx": [
null,
{
"n": null,
"small": false,
"rank": 1,
"of": 3,
"pct": 100,
"lo": null,
"hi": null,
"heat": null
},
null
],
"href": "/players/pam-two-emrxad012xnt6nbe4vy4",
"group": null
},
{
"key": "plr_znvh1bpfvdj2d79szye4:2025-26",
"label": "Pat One",
"cells": [
"2025-26",
96,
3
],
"ctx": [
null,
{
"n": null,
"small": false,
"rank": 2,
"of": 3,
"pct": 50,
"lo": null,
"hi": null,
"heat": null
},
null
],
"href": "/players/pat-one-znvh1bpfvdj2d79szye4",
"group": null
},
{
"key": "plr_d9b91wfaa4ym1y788aq8:2025-26",
"label": "Pip Three",
"cells": [
"2025-26",
59,
2
],
"ctx": [
null,
{
"n": null,
"small": false,
"rank": 3,
"of": 3,
"pct": 0,
"lo": null,
"hi": null,
"heat": null
},
null
],
"href": "/players/pip-three-d9b91wfaa4ym1y788aq8",
"group": null
}
],
"summary_rows": [
{
"key": "league",
"label": "League average",
"cells": [
null,
351.666667,
null
],
"ctx": null,
"decimals": [
null,
1,
null
],
"href": null,
"group": null
}
],
"total_rows": 3,
"next": null,
"compare_set": {
"label": "NBA players in the same season",
"n": 3,
"qualifier": "At least 1 games"
},
"sources": [
{
"key": "s1",
"source": "balldontlie",
"licence": "BALLDONTLIE Terms of Service, last updated September 17, 2026, §6: Data may be used for lawful commercial purposes, incl",
"snapshot_id": "snap_52",
"recorded_at": null,
"official": true,
"reconstructed": false,
"disagrees": false,
"precedence": 1
}
],
"coverage_notes": [],
"calc_errors": [],
"assumptions": [
"Each season's line from the source Kelly serves first by precedence (balldontlie, kelly), else the sum of its games.",
"All seasons Kelly holds.",
"At least 1 games (min_sample).",
"Stat catalogue 1.0.0; a missing value is not zero, and games without it are left out of its total."
],
"snapshot_id": "snap_52",
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"stale": false,
"summary": "3 player season rows, most points first (totals): Pam Two, 900.",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_cf5911a9a448cf448dc0f487",
"live_topics": [],
"rg_line": null
}POST /v1/views:parse
public, no key needed · 0 credits, not charged during the beta · MCP tool parse_view · status live
Plain words to a view spec, starting from the current table's fields: the spec, what was understood and the words it couldn't place, never guessed.
Body, application/json (required):
textstringThe person's words, up to 300 characters.
min length 1 · max length 300
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
columnsarray of stringStat keys (passing_yards) or dimensions (opponent, game_date), in display order, 1 to 40. Teams also take points_for, points_against, point_diff, wins, losses, ties, win_pct, ats_wins, ats_losses, ats_pushes, ats_pct, overs, unders, ou_pushes, over_pct and the opponent's figure for any team stat as opp_<key> (opp_passing_yards).
min items 1 · max items 40
calcsarray of stringUp to 20 formula columns over the rows, each 'key:formula', e.g. 'ypa:passing_yards / passing_attempts'; 'key~Label:formula' names its header. A formula reads totals; a stat it names that isn't a column is read for it. list_functions has the functions. Sort by a calc's key.
max items 20
kindstringtable (default), streak or span.
One of: table, streak, span (open: new values may be added).
entitystringplayer (default) or team.
One of: player, team (open: new values may be added).
grainstringgame, season (default) or career.
One of: game, season, career (open: new values may be added).
entity_idsarray of stringUp to 20 players or teams (plr_, tm_); none ranks every one (a finder).
max items 20
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
sortarray of stringUp to 3 columns, '-' first for highest first.
max items 3
filtersarray of stringGames to keep, each '<field> <op> <value>' (op = != > >= < <= in between), e.g. 'passing_yards >= 300', 'home = false'.
max items 12
havingarray of stringTests on a row's own totals, e.g. 'passing_yards >= 4000' for 4,000-yard seasons.
max items 8
splitstringA split from describe_metrics, e.g. home_away.
max length 32
modestringtot (default), pg (per game), season (per full season) or p100 (per 100 of the rate base).
One of: tot, pg, season, p100 (open: new values may be added).
contextstringnone (default), rank or pct beside each stat.
One of: none, rank, pct (open: new values may be added).
heatstringoff (default), league or rows.
One of: off, league, rows (open: new values may be added).
diffarray of stringTwo row keys: adds a difference row.
min items 2 · max items 2
conditionarray of stringkind streak: up to 4 tests every game must meet.
max items 4
lengthintegerkind span: games in the stretch, 2 to 162.
min 2 · max 162
min_sampleintegerOverride the qualifier for the comparison set (a minimum of games).
min 1 · max 10000
limitintegerRows per page, 1 to 200; default 50.
min 1 · max 200
cursorstringA result's next value, for the next page.
max length 64
as_ofstringAnswer as Kelly knew it at this time (RFC 3339); default now.
format date-time
snapshot_idstringA snapshot_id an earlier result returned, to read the same data again.
pattern ^snap_[0-9]{1,18}$
200 The view parse facts.
objectconstrequiredAlways view_parse.
Always "view_parse".
specobjectrequiredThe spec to send to get_view.
vconstThe contract version.
Always 1.
default 1
kindstringtable: rows of stats. streak: runs of games in a row that meet every condition. span: the best stretch of length games by the first sort column.
One of: table, streak, span (open: new values may be added).
default "table"
competitionstringrequiredThe competition slug, e.g. nfl, nba, ncaaf.
entitystringRows are players or teams; in racing (horse_racing, greyhounds, harness) runners, horses, jockeys, trainers or tracks.
One of: player, team, runner, horse, jockey, trainer, track (open: new values may be added).
default "player"
grainstringOne row per game, per season, or per career.
One of: game, season, career (open: new values may be added).
default "season"
entity_idsarray of stringUp to 20 IDs: plr_ or tm_; in racing rnr_ (runner), anm_ (horse), rpn_ (jockey, trainer) or ven_ (track). Empty: every player or team (a finder).
default []
from_seasonstringnullableFirst season, a label like 2025 or 2025-26.
default null
to_seasonstringnullableLast season.
default null
season_typestringregular, postseason or both.
One of: regular, postseason, both (open: new values may be added).
default "regular"
columnsarray of stringrequiredCatalogue stat keys (passing_yards) or dimensions (opponent), in display order. 1 to 40.
calcsarray of objectUp to 20 formula columns over the view's rows, shown after the columns. Each is an object or its text 'key[.mode][~label]:formula'. Errors come back in calc_errors, not as a refusal.
default []
keystringrequiredThe calc's column key: lowercase letters, digits and _, not one of the columns. Sort by it like any column.
formulastringrequiredThe formula, e.g. passing_yards / passing_attempts or ROLLING_AVG(passing_yards, 3). Excel syntax unless mode says otherwise.
labelstringnullableThe column header; the formula when null. No ':' (the URL form uses it).
default null
modestringThe formula's input language (ADR-K591 D1): excel unless the formula is written in another.
One of: excel, sheets, sql, polars, pandas, dplyr, dax, tableau, pine (open: new values may be added).
default "excel"
sortarray of stringUp to 3 columns, '-' first for highest first. Multi-sort in order.
default []
filtersarray of stringUp to 12 filters in the history grammar: '<field> <op> <value>', op one of = != < <= > >= in between. 'active = true' keeps players or teams with a season in the latest season ('active = false': retired).
default []
havingarray of stringUp to 8 tests on a row's own totals, after grouping: 'passing_yards >= 4000' keeps 4,000-yard seasons (filters test each game).
default []
splitstringnullableA catalogue split (home_away, opponent, month…): one row per split value.
default null
modestringtot: totals. pg: per game. season: per full season (17 games in the NFL, 82 in the NBA; the engine names it). p100: per 100 of the competition's rate base (dropbacks, possessions; the engine names it).
One of: tot, pg, season, p100 (open: new values may be added).
default "tot"
contextstringBeside each stat: nothing, its rank, or its percentile, in the comparison set the result names.
One of: none, rank, pct (open: new values may be added).
default "none"
heatstringHeat position for each stat cell: off, against the league comparison set, or against the other rows shown.
One of: off, league, rows (open: new values may be added).
default "off"
diffarray of stringTwo row keys: adds a difference row (first minus second).
default []
conditionarray of stringkind=streak: up to 4 filters each game must meet.
default []
lengthintegernullablekind=span: games in the stretch, 2 to 162.
default null
min_sampleintegernullableOverrides the catalogue's qualifier for the comparison set.
default null
limitintegerRows per page, 1 to 200.
default 50
cursorstringnullableA result's next value, unchanged.
default null
as_ofstringnullableRead the data as of this time (ISO 8601).
format date-time · default null
snapshot_idstringnullableRead one lake snapshot, e.g. snap_1234.
default null
querystringrequiredThe spec as a readable URL query beside the base view (ADR-K575 §4).
completebooleanrequiredEvery word was placed and the spec is valid.
understoodarray of stringrequiredThe phrases it placed, in order.
unknownarray of stringrequiredWords it couldn't place; ask, never guess.
problemstringrequirednullableWhy the words can't make a valid view, when they can't.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large
POST https://api.askkelly.ai/v1/views:parse
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"text": "seasons with at least 500 points",
"columns": [
"points",
"games"
]
}
{
"object": "view_parse",
"spec": {
"v": 1,
"kind": "table",
"competition": "nba",
"entity": "player",
"grain": "season",
"entity_ids": [],
"from_season": null,
"to_season": null,
"season_type": "regular",
"columns": [
"points",
"games"
],
"calcs": [],
"sort": [
"-points"
],
"filters": [],
"having": [
"points >= 500"
],
"split": null,
"mode": "tot",
"context": "none",
"heat": "off",
"diff": [],
"condition": [],
"length": null,
"min_sample": null,
"limit": 50,
"cursor": null,
"as_of": null,
"snapshot_id": null
},
"query": "sort=-points&h=points>=500",
"complete": true,
"understood": [
"at least 500 points",
"seasons"
],
"unknown": [],
"problem": null,
"summary": "Showing at least 500 points; seasons.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_02da57d435bb8d505ae6d1f3",
"live_topics": [],
"rg_line": null
}POST /v1/rules:test
public, no key needed · 1 credit, not charged during the beta · MCP tool test_rule · status live
A betting rule tested on past games, point in time: every game the rule's formula matched, bet at Kelly's consensus price at the weekly sheet's edition or the close and graded. Verdict v2 (nine checks in three groups, each with a zone), the honest forward ROI beside the past ROI, the record, units, ROI intervals, the exact and clustered p-value adjusted for the rules tried, the deflated sample, the rule against its own base, CLV, stress tests (half a point worse, -110, the other side), splits, each condition's impact and every bet in date order. Racing (horse_racing, greyhounds, harness): every runner the rule matches, bet at the off to win, place, each way or to lay, settled at SP, Betfair SP or the tote with dead heats, non-runners, Rule 4 and commission. A rule that reads anything not known at the bet moment, or Kelly's own edge, is refused as rule_not_testable with fixes.
Body, application/json (required):
formulastringThe rule: a formula true or false on each team-game (or each runner in a race, for horse_racing, greyhounds and harness), starting with =.
min length 2 · max length 2000
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...; racing: horse_racing, greyhounds or harness (seasons are calendar years).
pattern ^[a-z0-9_]{2,24}$
seasonsarray of stringSeasons to test, oldest first (1 to 25), e.g. ["2021", "2022"]; or use from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
marketstringspread (default), moneyline, over, under, draw or teaser (each match is a teased leg; legs from the same week go on tickets; see the teaser_ options). Soccer (epl, laliga, bundesliga, seriea, ligue1, mls, ucl): moneyline is three-way (a draw loses), spread is the Asian handicap (quarter lines settle half won or half lost) and draw bets the draw in the matched game. Racing: win (default), place, each_way, lay or lay_place, every runner the rule matches is a bet at the off.
One of: spread, moneyline, over, under, draw, teaser, win, place, each_way, lay, lay_place (open: new values may be added).
sidestringrule (default) bets the team the rule matched; fade bets the other side.
One of: rule, fade (open: new values may be added).
bet_momentstringclose (default), edition (football: the weekly sheet's Monday 10 AM ET line) or day_before (other sports: 24 hours before the start).
One of: close, edition, day_before (open: new values may be added).
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
stakestringrisk (default): 1 unit risked a bet; to_win: each bet sized to win 1 unit.
One of: risk, to_win (open: new values may be added).
pricestringconsensus (default): Kelly's price at the bet moment; flat_110: every bet at -110.
One of: consensus, flat_110 (open: new values may be added).
testsintegerRules tried on this data outside Kelly, this one included (default 1). Kelly counts the account's own; this may only raise that count.
min 1 · max 100000
detailstringfull (default) or summary (no bets list, splits or condition impact).
One of: full, summary (open: new values may be added).
base_formulastringThe base the rule filters, e.g. =AND(NOT(home), spread_close > 0); it must hold on every game the rule bets. Default: Kelly takes the rule's who-or-when conditions.
min length 2 · max length 2000
plan_roinumberAlso report the bets needed to prove this ROI (a fraction, 0.05 is 5%).
ci_levelnumberThe ROI interval's level, 0.8 to 0.99 (default 0.95).
min 0.8 · max 0.99
max_stale_minutesintegerLeave out a price whose snapshot is older than this at the bet moment (default 60 at the close, 360 earlier; never under 10 for games before October 2022).
min 5 · max 10080
bankroll_unitsnumberBankroll in units for the ruin chance and the staking replays (default 100).
max 1000000
horizon_betsintegerBets ahead the ruin chance looks (default 500).
min 1 · max 1000
pathsintegerSimulated futures for the ruin chance (default 5000).
min 1000 · max 10000
staking_plansarray of stringStaking plans to replay: flat, to_win, percent, kelly (default all four).
min items 1 · max items 4
percentnumberShare of the bankroll the percent plan bets (default 0.02).
max 0.5
kelly_fractionnumberMultiple of the Kelly criterion the kelly plan bets; omitted, the honest multiple from the record's sample size.
max 1
kelly_capnumberMost of the bankroll the kelly plan puts on one bet (default 0.05).
max 1
max_bets_per_dayintegerCap: bets a day (ET).
min 1 · max 1000
max_bets_per_gameintegerCap: bets a game.
min 1 · max 10
max_bets_per_team_weekintegerCap: bets on a team a week.
min 1 · max 50
max_open_stakenumberCap: share of the bankroll at risk at once.
max 1
daily_stop_loss_unitsnumberCap: no more bets that day once this many units are lost.
max 100000
cap_prioritystringWhich bets a cap keeps: earliest (default) or random.
One of: earliest, random (open: new values may be added).
warm_upstringas_written (default) grades every bet; require_full grades only bets whose windows had all the earlier games they read (ROLLING_AVG(margin, 5) needs 5).
One of: as_written, require_full (open: new values may be added).
settlestringRacing only: the price bets settle at: sp (industry SP; win default, each_way), bsp (Betfair SP; place, lay and lay_place default) or tote.
One of: sp, bsp, tote (open: new values may be added).
commissionnumberRacing only: Betfair commission on a winning BSP bet's winnings, as a share (default 0.05).
min 0 · max 0.2
ew_placesintegerRacing each_way only: places paid, instead of the standard terms by field size.
min 1 · max 8
ew_fractionnumberRacing each_way only: the place fraction of the win odds (0.2 is 1/5, 0.25 is 1/4), instead of the standard terms.
max 1
outputstringrule (default): the formula is true or false on each game. probability: it gives the side's chance from 0 to 1 (moneyline or spread, back side, edition or close), scored as calibration and bet at edge_threshold.
One of: rule, probability (open: new values may be added).
edge_thresholdnumberoutput probability only: the expected return a bet needs, chance × decimal price − 1; 0.02 (default) bets where the formula sees a 2% edge.
robustness_offarray of stringRobustness blocks to leave out (all of them with "all"); every block runs by default.
max items 10
ablationstringshapley (default): each condition's share of the extra units (Shapley values); leave_one_out: only the rule without each condition.
One of: shapley, leave_one_out (open: new values may be added).
break_datestringA date to test for a change in the edge (a Chow test), e.g. 2023-09-01.
format date
random_baseline_simsintegerRandom rules drawn for the baseline (default 2,000).
min 100 · max 10000
monte_carlo_simsintegerReorderings of the bets (default 10,000; fewer, never under 2,000, for long records).
min 100 · max 20000
monte_carlo_modestringblock (default: whole bet-days resampled in runs of about a week), iid or permutation.
One of: block, iid, permutation (open: new values may be added).
skip_probnumberChance each bet is missed in the Monte Carlo and noise runs (default 0).
min 0 · max 0.5
trim_winnersnumberShare of the biggest winners removed before the Monte Carlo (default 0).
min 0 · max 0.2
noise_runsintegerRuns a noise family (default 200).
min 10 · max 1000
input_scalenumberNoise on the numbers the rule compares, as a share of their spread (default 0.10).
min 0 · max 0.5
line_probnumberChance each bet's line moves half a point either way in the line noise (default 0.5).
min 0 · max 1
savebooleantrue keeps this rule as a version of a saved rule named name (or of rule_id): the same rule again keeps its version, a different one is the next.
namestringThe saved rule's name, for a new one (1 to 120 characters).
min length 1 · max length 120
rule_idstringAdd this test as a version of a rule already saved (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
transferbooleanAlso run the rule on its other market (spread and moneyline; over and under) and on other leagues, with its numbers moved to the same percentile there, pooled across leagues (transfer). Evidence beside the verdict; these runs don't count toward rules tried.
transfer_leaguesarray of stringLeagues to carry it to (default: sibling leagues first, then the rest).
min items 1 · max items 13
transfer_marketsstringauto (default) adds the other market of the rule's own league; none leaves it out.
One of: auto, none (open: new values may be added).
transfer_thresholdsstringboth (default), literal (the rule's numbers as written) or quantile (each number moved to the same percentile of its field in the other league).
One of: both, literal, quantile (open: new values may be added).
translation_idstringThe translation_id a translate_rule result gave, when the formula came from the person's words; it is kept with the test.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
chartsbooleantrue adds the equity, drawdown and rolling ROI charts as Kelly Chart Specs (detail full).
200 The rule test facts.
objectconstrequiredAlways rule_test.
Always "rule_test".
verdictobjectrequiredVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
overallobjectrequiredEvery graded bet.
recordstringWins-losses(-pushes).
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
betsintegerGraded bets.
decisionsintegerWins plus losses.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_intervalarray of numbernullableROI's 95% interval at 1 unit risked a bet.
win_ratenumbernullableWins over decisions.
break_evennumbernullableThe win rate that breaks even at these prices.
significanceobjectnullableThe record test (kmath.records.record_test) and verdict v2's statistics (kmath.significance): the p-value, adjusted for the rules tried, and the anytime-valid test.
p_valuenumbernullableThe larger of the exact and the clustered one-sided p against no skill at these prices (F12, F23).
methodstringnullableexact_binomial, exact_simulation, saddlepoint or z_score.
znumbernullableThe score z: units over the square root of Σ(d − 1).
t_statnumbernullableThe t statistic with the sample variance.
p_value_exactnumbernullableThe exact (or z) tail alone.
p_value_clusterednumbernullableThe tail with bets clustered by slate and team-season (F23).
p_value_vs_vignumbernullableAgainst a random bet at the no-vig close (needs closes).
p_adjustednumbernullableThe p-value adjusted for the rules tried (Romano-Wolf on the ledger's rules, Bonferroni on the count otherwise; F14).
adjust_methodstringnullableromano_wolf, holm or none.
testsintegernullableRules tried in this family, this one included.
tests_effectivenumbernullableEffective tests (Li-Ji): near-duplicate rules count as fewer.
reality_check_pnumbernullableWhite's Reality Check p (when the ledger holds the rules' units).
spa_pnumbernullableHansen's SPA p (when the ledger holds the rules' units).
clears_t3booleannullablez at or above 3 (Harvey-Liu-Zhu's bar).
haircutnumbernullableHarvey-Liu haircut: the share of the ROI the rules tried take away.
haircut_roinumbernullableROI after the haircut, per decision.
anytime_validobjectnullableThe record as a bet against no skill (F17): e-value, anytime p, ROI range.
e_valuenumbernullableThe e-value now.
p_valuenumbernullable1 over the largest e-value so far.
max_enumbernullableThe largest e-value along the path.
roi_cs95array of numbernullableAn ROI range valid at every bet.
roi_ci95array of numbernullableROI's 95% score interval, clustered (F13, F23).
units_ci95array of numbernullableUnits' 95% interval: decisions times the ROI interval.
roi_ciobjectnullableROI intervals at ci_level: score (the one shown), BCa, block bootstrap over slates and anytime-valid (F13).
deflatedobjectnullablePer-bet Sharpe, PSR and DSR at the effective tests, and the bets needed: deflated (sample_needed), single test and 80% power, each times the design effect (F15).
sample_neededintegernullableDecisions needed, deflated for the rules tried; null with no edge to prove.
sample_needed_singleintegernullableDecisions needed for one rule tried.
sample_needed_power80integernullableDecisions for 80% power at this ROI.
sample_needed_planintegernullableDecisions for 80% power at plan_roi.
dsrnumbernullableDeflated Sharpe ratio.
psrnumbernullableProbabilistic Sharpe ratio.
sample_neededobjectnullableBets needed to tell this win rate from break-even (one-sided, with 80% power, ...).
small_samplebooleanUnder 30 decisions.
few_decisionsbooleanUnder 5 decisions.
honestobjectnullableThe honest forward ROI (F22, the headline ROI): empirical Bayes toward a prior mean (the rule's own closing-line value when at least half its bets have a close, else the usual two-way margin), less the winner's curse for the rules tried and the out-of-sample decay.
roinumbernullableHonest forward ROI.
ci80array of numbernullable80% range.
raw_roinumbernullableThe past ROI per decision, beside it.
prior_meannumbernullableThe prior mean the ROI is shrunk toward (prior.source says which).
prior_sdnumbernullableThe spread of true ROIs (0.02 until fitted on the ledger).
shrinknumbernullableWeight on the prior, se²/(se² + τ²).
selection_adjnumbernullableWinner's-curse correction for the rules tried.
decay_factornumbernullableOut-of-sample decay on the excess (0.74).
stabilizes_at_betsintegernullableDecisions at which the record and the prior weigh the same.
uses_clvbooleanThe prior mean is the rule's own closing-line value.
prior_basisstringfitted (prior_sd from the ledger's fit) or seed.
priorobjectWhere the prior mean comes from.
sourcestringclv (the rule's mean closing-line value), margin (the two-way margin, -4.55%) or fitted.
meannumbernullableThe prior mean used.
closesintegerDecided bets with a no-vig close.
cappedbooleanA closing-line value above 0 was held at 0: under 50 bets have a close.
wordsstringThe prior, said plainly.
dependenceobjectnullableBets clustered by slate (the date; the week in the NFL and college football) and team-season (F23).
clustersintegerSlates.
design_effectnumberClustered over iid variance.
effective_betsintegerDecisions over the design effect (at least 1).
methodstringcr1_twoway, cr1_slate, cr1_larger_oneway, wild_bootstrap or wild_bootstrap_webb.
baseobjectnullableThe rule against the rest of its own base (F19): the who-or-when conditions alone, or base_formula.
formulastringnullableThe base, or null for the whole market.
sourcestringauto, user or market.
One of: auto, user, market (open: new values may be added).
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
roinumbernullableThe base's ROI per decision.
liftnumbernullableRule ROI less base ROI.
p_value_vs_basenumbernullableSeason-stratified permutation p of the rule within its base.
units_from_basenumbernullableUnits the base alone would have made on the rule's stakes.
units_from_filternumbernullableThe rule's units less units_from_base.
clvobjectrequiredClosing-line value as evidence (F21): each bet's no-vig CLV, q_close x price - 1, on bets Kelly holds a close for. Empty when betting at the close.
bets_with_closeintegerBets with a close.
bets_without_closeintegerBets without one.
referencestringThe close used: consensus (the no-vig average across books).
devigstringHow the vig was removed: multiplicative.
lines_convertedintegerBets whose close was carried from the closing line to the bet's line (NFL spreads and totals).
coveragenumbernullableShare of bets with a close probability (null with no bets).
closes_staleintegerCloses left out as older than 60 minutes at the start, or seen after it (audit F9): those bets keep their result but have no CLV.
points_coveragenumbernullableShare of bets with points beaten (spreads and totals).
points_testobjectnullableThe clustered test on points beaten (audit F3).
betsintegerBets with points beaten.
meannumbernullableMean points the bet's line beat the close by.
std_error_clusterednumbernullableStandard error clustered by slate and team-season.
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean points beaten > 0.
cluster_methodstringHow the standard error was clustered.
basisstringWhat Beats the close is judged on: probability, or points when under 80% of bets have a close probability and more have points beaten (audit F3).
One of: probability, points (open: new values may be added).
meannumbernullableMean CLV (fraction): the bets' expected ROI if the close is right.
ci95array of numbernullableClustered 95% interval on the mean.
min items 2 · max items 2
p_valuenumbernullableOne-sided clustered p of mean CLV > 0.
beat_ratenumbernullableShare of bets that beat the close, ties left out.
beat_rate_ci95array of numbernullableWilson 95% interval on beat_rate.
min items 2 · max items 2
sign_test_pnumbernullableExact one-sided sign test against half.
mean_probnumbernullableMean of the close's no-vig probability less the bet moment's.
mean_pointsnumbernullableMean points the bet's line beat the closing line by (spreads and totals).
points_betsintegerBets mean_points is over.
profit_slopenumbernullablePer-bet units regressed on CLV: near 1 when CLV turned into profit.
profit_slope_senumbernullableIts slate-clustered standard error.
sample_neededintegernullableBets CLV needs for a 95% interval to clear 0 at this mean and spread (null unless the mean is above 0).
sample_needed_roiintegernullableBets profit would need at the same edge.
significanceobjectnullableThe CLV mean's tests (null under two bets).
tnumbernullableMean over its iid standard error.
ci95array of numbernullableStudent 95% interval (iid).
min items 2 · max items 2
std_errornumbernullableiid standard error.
bets_needednumbernullableBets a t of 2 needs at this mean and spread.
std_error_clusterednumbernullableStandard error clustered by slate and team-season (F23).
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean CLV > 0 (wild cluster bootstrap under 42 slates, else Student t); beats_close reads it.
cluster_methodstringHow the standard error was clustered.
design_effectnumbernullableClustered variance over iid variance.
notestringWhy the block is empty, when it is.
luckobjectThe luck-adjusted result (F20): the units the market's fair probabilities expected the decided bets to make against what they made. Pushes count zero both ways.
readingstringluck (bets before the close: the gap is outcome noise) or versus_the_market (bets at the close: the gap is the rule's edge on the closing market).
One of: luck, versus_the_market (open: new values may be added).
close_sourcestringconsensus.
devigstringmultiplicative.
betsintegerDecided bets with a fair probability.
bets_without_probintegerDecided bets without one.
left_outobjectBets left out.
pushintegerPushes.
halfintegerHalf results (quarter lines).
lines_convertedintegerBets whose close was carried to the bet's line.
expected_unitsnumbernullableSum of stake x (q x (price - 1) - (1 - q)).
actual_unitsnumbernullableUnits the bets made.
luck_unitsnumbernullableactual_units less expected_units.
luck_se_unitsnumbernullableluck's standard error given the probabilities.
expected_winsnumbernullableSum of q.
actual_winsintegernullableWins.
luck_percentilenumbernullableWhere the wins sit among the market's (Poisson-binomial, mid-p): 0.95 is luckier than (95) of (100) identical paths.
expected_roinumbernullableexpected_units over units risked.
actual_roinumbernullableactual_units over units risked.
thetanumbernullablePer-bet units regressed on expected units (30 bets or more): near 1 when the close is calibrated on these bets.
by_seasonarray of objectEach season, in calendar order; luck_units add up to the total.
seasonstringSeason.
betsintegerDecided bets with a probability.
expected_unitsnumberUnits the market expected.
actual_unitsnumberUnits made.
luck_unitsnumberactual_units less expected_units.
capacityobjectnullableF8: the limits Kelly measured at the bet moment (ADR-K783); evidence beside the verdict, never in it. No book publishes its limits, so only measured ones are shown.
availablebooleanWhether measured limits are shown for this market and moment.
marketstringnullableThe market as line history names it (spreads, totals, h2h).
before_startstringnullableThe time before the start the bet moment falls in.
One of: 0_1h, 1_12h, 12_36h, 36_72h, 72h_plus (open: new values may be added).
booksarray of objectPer book, highest median first.
bookstringnullableThe book as the vendor names it.
vendorstringnullableWho measured it.
currencystringnullableThe stake currency, null when the vendor doesn't say.
mediannumbernullableThe median limit.
p25numbernullableIts 25th percentile.
p75numbernullableIts 75th percentile.
minnumbernullableThe smallest.
maxnumbernullableThe largest.
snapshotsintegerLimit readings behind it.
gamesintegerGames behind it.
first_seenstringnullableThe first reading's date.
last_seenstringnullableThe last reading's date.
max_stake_per_betobjectnullableThe highest book median: the most one book typically took on one bet.
bookstringnullableThe book.
amountnumbernullableThe median limit.
currencystringnullableIts currency.
per_season_at_limitnumbernullablemax_stake_per_bet times the rule's bets a season.
bets_per_seasonnumbernullableThe rule's bets a season.
limits_heldintegerLimit readings Kelly holds for this market and moment.
reasonstringnullableWhy none are shown, when available is false.
stressobjectrequiredThe same bets graded harder.
half_point_worseobjectnullableEvery line half a point worse (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
flat_110objectnullableEvery bet at -110 (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
other_sideobjectThe other side of every bet.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
break_evenobjectnullableF7: how much worse the price or line can get before the profit is gone.
price_multipliernumbernullableThe share of each payout (d - 1) that breaks even (under 1: the room to spare).
american_avgnumbernullableThe flat American price that breaks even.
decimal_avgnumbernullableThe same, decimal.
line_pointsnumbernullableThe most points worse (in halves, up to 3) the line could be with a profit left; 0 when none; null for moneylines.
gridarray of objectnullableF7: the bets at -105 to -120 with the line 0 to 1 point worse (moneylines: payouts cut 0-6%); left out at detail summary.
vigintegernullableEvery bet at this American price.
line_shiftnumbernullablePoints the line moved (negative: worse).
price_multipliernumbernullableMoneylines: the share of each payout kept.
unitsnumbernullableUnits.
roinumbernullableROI.
slippage_mcobjectnullableF7: each bet's line worsened by a draw from the market's own Monday-to-close moves, 1,000 seeded runs; null without such moves.
runsintegerRuns.
roi_p05numbernullable5th percentile ROI.
roi_p50numbernullableMedian ROI.
moves_fromintegerLine moves the draws come from.
move_mediannumbernullableTheir median size, points.
price_worseobjectnullableTeasers: every ticket ten cents worse (price_american). Racing: every bet's winnings 5% smaller.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
legsobjectnullableTeasers: every teased leg, judged against the per-leg break-even (1/D)^(1/n).
recordstringLegs won-lost(-pushed).
decisionsintegerLegs won plus lost.
cover_ratenumbernullableLegs won over legs decided.
break_even_per_legnumberThe cover rate a leg needs at the ticket's price.
significanceobjectnullableThe legs' record test against the per-leg break-even.
sample_neededobjectnullableLegs needed to tell this cover rate from the break-even.
random_baselineobjectnullableF18: the rule against random bets of the same count on the same games, matched by season and price bucket.
simsintegerRandom rules drawn.
betsintegerThe rule's bets.
poolintegerBets the market offered.
percentilenumbernullableShare of random rules with a lower ROI.
beats_per_100integerRandom rules out of 100 the rule beats.
p_valuenumbernullable(1 + random rules at or above) / (sims + 1).
analytic_pnumbernullableThe finite-population normal cross-check.
random_roi_mediannumbernullableMedian random ROI (about minus the vig).
random_roi_p05numbernullable5th percentile random ROI.
random_roi_p95numbernullable95th percentile.
best_of_testsobjectWould the best of M random rules (M = rules tried) do this well?
testsintegerM.
p_valuenumbernullable1 - F(ROI)^M.
histogramobjectnullableRandom ROIs in 30 bins (edges, counts); full detail only.
decayobjectnullableF24: is the rule still making money? Rolling ROI and CLV, one unknown break (sup-F, block permutation p), a CUSUM alarm, a known-date test, the trend and the decay ratio.
betsintegerGraded bets.
windowintegernullableRolling window, bets.
rollingarray of objectnullableUp to 200 points; full detail only.
breakobjectnullableThe most likely date the edge changed.
atstringDate.
bet_indexintegerBets before it.
sup_fnumbernullablesup-F.
p_valuenumbernullableBlock permutation p-value.
at_edgebooleanWithin the first or last 15%: not a confirmed break.
roi_beforenumbernullableROI before.
roi_afternumbernullableROI after.
bets_afterintegerBets after.
cusumobjectnullablePage's lower CUSUM alarm.
known_breakobjectnullableA Chow test at break_date.
trend_per_100_betsnumbernullableProfit per bet's change per 100 bets (1 unit risked).
decay_rationumbernullableROI after the break (or the newest 30%) over ROI before.
clv_breakobjectnullableThe break test on CLV.
roi_newest_30numbernullableROI of the newest 30% of bets.
ablationobjectnullableF27: each top-level AND condition's Shapley share of the units beyond the base's ROI.
baseobjectEvery game with a line: decisions and ROI.
totalnumbernullableThe rule's units beyond the base's ROI (the Shapley values add up to it).
clausesarray of objectOne a condition, in the order written.
conditionstringThe condition.
shapley_unitsnumbernullableIts share of the extra units.
labelstringdead, harmful, load_bearing or helps.
One of: dead, harmful, load_bearing, helps (open: new values may be added).
fixobjectnullableRemove it: the rule without it.
pairsarray of objectPairs (up to 6 conditions).
methodstringexact or sampled.
One of: exact, sampled (open: new values may be added).
nearbyobjectnullableF28: every number in the rule moved two steps each way (steps fixed by the field).
numbersarray of objectThe rule's numbers.
variantsintegerVariants in the grid.
distinct_bet_setsintegerVariants with their own bets.
roi_mediannumbernullableMedian ROI of the variants.
roi_p05numbernullable5th percentile.
roi_p95numbernullable95th percentile.
spike_rationumbernullable(ROI - median) / IQR of the variants.
readingstringnullableplateau, slope or spike.
nastringnullableWhy there are no numbers to move.
monte_carloobjectnullableF31: drawdowns and losing runs the same bets give in other orders.
simsintegerSimulations.
modestringblock, iid or permutation.
max_drawdown_unitsobjectMax drawdown, units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
longest_losing_runobjectLongest losing run, bets.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
final_unitsobjectFinal units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
prob_below_zeronumbernullableShare of runs finishing below zero.
worst_100_bets_units_p05numbernullable5th percentile of the worst 100-bet stretch.
noiseobjectnullableF32: jittered inputs, lines moved half a point either way, and an earlier bet moment.
inputsobjectnullableThe rule's compared numbers jittered.
runsintegerRuns.
profitable_runsintegerRuns with a profit.
roi_mediannumbernullableMedian ROI.
roi_p05numbernullable5th percentile ROI.
overlap_mediannumbernullableMedian Jaccard overlap of the bets with the rule's.
linesobjectnullableLines (moneylines: payouts) moved at random.
delayobjectnullableThe rule bet at the earlier moment.
concentrationobjectnullableF34: the profit without the best season, team, opponent, month and weekday.
byarray of objectGroupings in a fixed order.
groupstringseason, team, opponent, month or weekday.
bestobjectnullableThe best group (3+ bets).
without_bestobjectnullableThe rule without it.
effective_groupsnumbernullable1 / Herfindahl of the profitable groups' shares.
p_more_concentratednumbernullablePermutation p-value (teams and opponents).
trim_top_0_05objectnullableWithout the top 5% of bets (moneylines) or bet-days.
split_dimensionsarray of objectnullableF35: the rule by fixed dimensions with shrunk ROI and intervals.
dimensionstringThe dimension.
heterogeneity_pnumbernullableChi-square p-value of win rates across its levels.
heterogeneity_qnumbernullableBenjamini-Hochberg q across the dimensions.
real_differencebooleanq < 0.10.
where_profitstringnullableThe level with the most shrunk extra units, when real.
cellsarray of objectnullableLevels in a fixed order; full detail only.
robustnessobjectThe suite's run: the verdict v2 checks it feeds and its seed.
checksarray of objecthalf_point_worse, nearby_holds, not_concentrated and still_working.
keystringrequiredThe check.
One of: half_point_worse, nearby_holds, not_concentrated, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na.
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumbernullableThe figure the status compares.
thresholdnumbernullableWhat it is compared with.
fast_pathbooleanThe rule's numbers were re-bet from its compiled columns.
seedintegerThe seed (a hash of the rule and its options).
evaluationsintegerFormula evaluations the suite ran.
offarray of stringBlocks switched off.
pathobjectThe units path in date order: drawdown and longest losing run; at detail full its series (F47).
max_drawdown_unitsnumberThe deepest fall from a peak, units.
longest_losing_runintegerMost losses in a row (pushes don't break it).
seriesobjectnullableBet-indexed lines of the graded bets (detail full): index the bet numbers kept (past 600 bets thinned by LTTB, the deepest drawdown always kept), dates, units, drawdown, base_units (the market base's ROI at the same stakes, null unless measured), band_p025 and band_p975 (the no-skill range: 1,000 seeded runs of the same bets at their fair chances; null under 20 bets or for racing), rolling_roi_100 with its no-skill range rolling_p025 and rolling_p975, and markers (season starts, the holdout).
xstringAlways bet_index.
One of: bet_index (open: new values may be added).
indexarray of integerBet numbers (1 is the first graded bet).
datesarray of stringEach kept bet's date (YYYY-MM-DD, UTC).
unitsarray of numberUnits to date.
drawdownarray of numberUnits below the running peak (0 or below).
base_unitsarray of numbernullableThe market base at the same stakes.
band_p025array of numbernullableNo-skill 2.5th percentile.
band_p975array of numbernullableNo-skill 97.5th percentile.
rolling_roi_100array of numberROI over the last 100 bets (null before).
rolling_p025array of numberIts no-skill 2.5th percentile.
rolling_p975array of numberIts no-skill 97.5th percentile.
markersarray of objectWhere each season and the holdout start.
typestringseason or holdout.
One of: season, holdout (open: new values may be added).
atintegerThe bet number it starts at.
labelstringnullableThe season, or newest 30%.
downsampledbooleanThe series were thinned to 600 points.
expected_max_drawdown_zero_edgenumbernullableThe median deepest drawdown of the no-skill runs, units.
holdoutobjectnullableThe latest 30% of bets against the earlier 70%.
seasonsarray of objectSeasons in calendar order.
splitsarray of objectnullableHome, away, favorite, underdog and line bands, in a fixed order; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
withoutarray of objectnullableOne row a top-level AND condition: the rule's record without it; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
p_value_vs_rulenumbernullableThe rule against the bets the condition removes (F19).
p_value_vs_rule_adjustednumbernullablep_value_vs_rule, Holm-adjusted over the conditions shown.
betsarray of objectEvery bet in date order (left out at detail summary).
keystring{game_id}:{team_id} (racing: {race_id}:{runner_id}).
starts_atstringKickoff (RFC 3339).
format date-time
seasonstringThe season.
labelstringDate, team and opponent.
priceobjectnullableThe price at the bet moment, decimal.
linenumbernullableThe bet's line as graded (the team's spread, the total).
resultstringnullablewin, loss or push (soccer quarter lines: half_win or half_loss); null until played.
close_probnumbernullableThe fair closing probability of the same side, for CLV.
units_to_datenumberUnits after this bet, in date order.
how_gradedstringnullableHow the bet was graded, in words.
unitsnumbernullableUnits the bet won or lost at the run's staking (racing: its exact profit or loss with dead heats, each way, Rule 4 and commission); null until played.
drawdown_to_datenumbernullableUnits to date less their running peak (0 or below); null until played.
sidestringnullablehome or away (spread, moneyline), over, under or draw.
marketstringThe market graded (the other total when fading a total).
bet_momentstringWhen the bet was taken: edition, day_before or close.
clv_probnumbernullableThe fair closing chance less the fair chance at the bet moment, when Kelly holds both.
samplestringnullableholdout for the latest 30% of graded bets by time, else in_sample; null with no holdout or until played.
One of: in_sample, holdout (open: new values may be added).
whyarray of objectWhy it matched: up to six conditions (F46).
clausestringThe condition as written.
valuenumber | boolean | stringnullableThe compared expression's number for a comparison (rest_days - opponent.rest_days for rest_days - opponent.rest_days >= 3), else true or false.
settlementarray of stringThe settlement rules that decided the bet (IDs in settlement.rules).
leak_testobjectThe leak test (F3): the rule and each part of it evaluated again on a sample of rows with the data cut at the bet moment. A leak refuses the test (422 rule_not_testable, problem leak_found), so a report always says clean or not_run.
statusstringrequiredclean, leak_found or not_run.
One of: clean, leak_found, not_run (open: new values may be added).
rows_checkedintegerrequiredRows rebuilt at the bet moment.
matched_checkedintegerOf them, rows the rule matched.
matches_flippedintegerRows whose match changed with the data cut.
partsarray of objectrequiredParts that changed, in the formula's order (empty when clean).
namestringThe part, as written (rule is the whole rule).
rows_changedintegerRows where it changed.
max_changenumbernullableThe largest change in a number.
examplestringOne changed row, {game_id}:{team_id}.
warm_upobjectThe warm-up check (F4): bets taken before the rule's windows had the earlier games they read (one season at a time; windows never reach into last season).
policystringThe policy graded in the report.
One of: as_written, require_full (open: new values may be added).
games_neededintegerEarlier games every window needs (0: the rule has no window).
windowsarray of objectEach window in the formula's order.
exprstringThe window, as written.
games_neededintegernullableEarlier games it reads (null: season to date).
short_rowsintegerRows in scope with fewer earlier games this season.
short_betsintegerBets the rule took with fewer.
short_betsintegerBets taken with a window short.
by_policyarray of objectnullableas_written, then require_full; null when the rule has no window.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
warningbooleanShort bets are over 10% of the bets, or leaving them out moves ROI by more than half its 95% interval's half-width.
settlementobjectHow bets were settled (F9): Kelly's own grading rules as data, and each rule's bets.
profilestringThe rules profile (kelly_standard).
rules_versionstringThe rules table's date.
appliedarray of objectRules that decided at least one bet, in table order.
rulestringRule ID.
betsintegerBets.
rulesarray of objectThe rules for this league and market.
idstringRule ID.
textstringThe rule in words.
effective_fromstringThe date Kelly's grader took it up.
coverageobjectCoverage (F10): from every game in scope to the bets graded, each season's priced share, blank inputs and whether the rows left out for no line differ from the bets kept.
flowarray of objectThe steps in order; rule_true rows = decisions + every row out after it.
stepstringin_scope, rule_true, one_bet_a_game, full_windows, priced, graded or decisions.
gamesintegerGames (in_scope only).
rowsintegerRows.
outobjectRows that left at this step, by reason.
by_seasonarray of objectEvery season asked for, in order.
seasonstringSeason.
gamesintegerGames in scope.
rowsintegerRows.
rule_trueintegerRows the rule matched.
betsintegerBets graded.
inputs_missingarray of objectInputs read bare on the row, with blanks.
attributestringThe input.
rows_blankintegerRows with no value.
left_out_differsobjectRows left out for no line against the bets kept: standardized mean difference over home, neutral site and rest days; flagged over 0.1 with 10 or more left out.
flagbooleanThey differ.
max_smdnumbernullableThe largest difference.
covariatestringnullableWhere it is.
left_outintegerRows.
keptintegerRows.
by_covariatearray of objectEach covariate's difference.
momentobjectThe bet moment's clock (F5): its target time, the prices' snapshot age at it and where the lines came from.
bet_momentstringedition, day_before or close.
targetstringnullableThe moment's time for a game, in words.
max_stale_minutesintegerOldest price kept, in minutes before the target.
floor_minutes_before_oct_2022integerThe limit's floor for games before October 2022 (the Odds API's 10-minute snapshots).
rows_pricedintegerRows the rule took that had a price.
rows_agedintegerRows whose price carries its snapshot time and was kept.
median_snapshot_age_minutesnumbernullableMedian minutes from snapshot to target, kept rows.
max_snapshot_age_minutesnumbernullableOldest kept snapshot, in minutes before the target.
stale_priceintegerRows left out: price older than max_stale_minutes.
after_momentintegerRows left out: price seen after the target.
age_unknownintegerRows kept whose price has no snapshot time (Kelly's own frozen lines).
line_sourcesobjectKept rows by line source (sheet_line, odds_api_early, odds_api_close).
left_outobjectrequiredGames the rule matched but didn't bet.
no_lineintegerNo Kelly line at the bet moment.
both_sidesintegerThe rule matched both teams.
stale_priceintegerThe price was older than max_stale_minutes at the bet moment, or seen after it (F5).
unpairedintegerTeasers: legs left off a ticket (a week with too few, or the leftover in chronological pairing).
no_priceintegerRacing: runners with no price to settle at.
non_runnersintegerRacing: non-runners (void).
void_racesintegerRacing: abandoned or void races.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
capsobjectnullableExposure caps applied in time order (F39); the record and checks are on the capped ledger.
skippedintegerBets the caps skipped.
by_reasonobjectSkipped bets by cap.
uncappedobjectThe record without the caps.
random_priority_mean_roinumbernullableMean ROI over random tie-breaks.
ruleobjectrequiredThe rule and the options it was tested with.
rule_hashstringThe rule's identity: SHA-256 of its canonical formula (spacing, case and the order of AND and OR conditions don't change it) with the competition, market, side, bet moment, seasons, season type, stake and price. Detail and tests are not part of it.
pattern ^sha256:[0-9a-f]{64}$
translation_idstringThe rules:translate result this formula came from (F43), when the request named one.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
readsarray of objectEvery name the rule reads, once.
namestringThe attribute, with its relation (opponent.rest_days).
knownstringIts moment: static, pregame, edition, close or final (a window's input is read before the game).
windowedbooleanA backward window reads it (earlier games only).
formula_versionstringThe formula language version the rule was read with (F45).
catalogue_versionstringnullableThe stat catalogue version its names were bound against.
reportobjectA shared report's own facts (get_rule_report only, F54).
report_idstringThe report.
report_urlstringIts page.
format uri
visibilitystringunlisted or public.
One of: unlisted, public (open: new values may be added).
author_rules_triedintegerThe author's rules tried on this market when shared.
frozenbooleanAlways true.
rerunobjectThe run's request, to test it again on today's data.
report_urlstringnullableA frozen report of this run anyone with the link can open (F54): only when the request asked (share) and the call has an account; else null. share_rule_run makes or changes one later.
format uri
chartsarray of objectOnly when the request asked (charts) at detail full: equity against the no-skill band, the drawdown and the rolling ROI as Kelly Chart Specs (F47).
transferobjectnullableOther markets and sports (F33): the same rule on its other market and on other leagues, numbers mapped to the same meaning, pooled with DerSimonian-Laird random effects. Evidence beside the verdict, never part of it; these runs don't count toward rules tried. Present only when the request asks for transfer.
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
whystringnullableWhy nothing ran, when nothing did.
optionsobjectThe leagues, markets and thresholds asked for.
cellsarray of objectEach market and league in a fixed order (the other market first, then sibling leagues), never sorted by a result; up to 12.
competitionstringThe league.
marketstringThe market.
thresholdsstringliteral (the rule's own numbers) or quantile (each number moved to the same percentile of its field in this league).
One of: literal, quantile (open: new values may be added).
formulastringThe formula this cell ran (quantile cells carry the moved numbers).
seasonsarray of stringSeasons read, as this league labels them.
mappedarray of objectnullableQuantile cells only.
numberstringThe number as written.
fieldstringWhat it is compared with.
percentilenumbernullableIts percentile in the rule's own league (a fraction).
valuenumbernullableThe number at the same percentile here.
decisionsintegerWins plus losses.
betsintegerGraded bets.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% score interval.
senumbernullableROI's standard error at 1 unit risked a bet.
p_valuenumbernullableThe cell's own one-sided p against no skill at its prices.
p_holmnumbernullableHolm's adjusted p across the cells.
pooledbooleanThe cell is in the pool (another league, one a league, 10+ decisions).
pooledobjectnullableThe pool of the other leagues' cells.
roinumbernullableRandom-effects ROI.
roi_ci95array of numbernullableIts 95% interval.
senumbernullableIts standard error.
p_valuenumbernullableTwo-sided p of the pooled ROI.
i2numbernullableI²: the share of the spread across leagues beyond chance (a fraction).
tau2numbernullableτ²: the between-league variance of true ROI.
qnumbernullableCochran's Q.
cellsintegerCells pooled.
same_directionobjectnullablePooled cells whose ROI has the rule's own sign.
agreeintegerCells the same way.
cellsintegerCells.
p_valuenumbernullableTwo-sided sign-test p.
textstring"4 of 5".
not_testablearray of objectLeagues and markets left out, with why.
competitionstringThe league.
marketstringnullableThe market.
missingarray of stringFields that league lacks.
whystringWhy.
notestringWhat the panel is and isn't.
run_idstringnullableThis run, kept on the account (run_…): the same request on the same data is the same run. GET /v1/rules/runs/{run_id} reads it back. Null without an account.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
data_versionstringThe data the run read: Kelly's lake snapshot (or the data's as-of time). A new snapshot is a new run of the same rule.
manifestobjectnullableWhat this run read and ran, so the same request can be run again and checked against it (ADR-K885). No prices, lines or rows of source data.
versionintegerThe manifest's own layout version.
codeobjectThe code that graded the run.
engineintegerThe tester's revision; it rises with any fix that can move a result on the same data.
releasestringnullableThe deployed commit.
formula_versionstringnullableThe formula language's version.
catalogue_versionstringnullableThe stat catalogue's version.
attributes_sha256stringnullableA hash of the attribute registry: what each input means and when it is known.
settlement_rulesstringnullableThe settlement rules' version.
dataobjectThe data the run read.
versionstringThe same value as data_version.
snapshotsarray of stringThe lake snapshots read, in order.
one_snapshotbooleanWhether every season was read on one snapshot.
as_ofstringnullableWhen the read was made (RFC 3339).
format date-time
data_throughobjectHow far Kelly's history went, by league.
latest_resultstringnullableThe newest game read that had a result.
rows_built_atstringnullableWhen the prepared rows were built, where the run read them.
format date-time
queriesobjectEach reviewed lake query the tester reads, with a hash of its text.
requestobjectThe request.
request_hashstringThe request's hash.
rule_hashstringThe rule's hash.
seedintegerThe seed every simulation in the run used.
sourcesobjectWhere the run's inputs came from.
pricesobjectBets priced, by the source of the line.
inputsarray of objectEach input the rule read, with the moment it is known.
namestringThe input.
knownstringnullableThe moment it is known.
seasonsarray of objectEach season read.
seasonstringThe season.
rowsintegerTeam-games in scope.
left_outobjectMatched games that were not bet, by reason.
sha256stringThe hash of this block. Two runs with the same hash read the same data with the same code.
since_runobjectnullableOnly on a run read back: what has changed in Kelly's tester since the run was kept.
statusstringcurrent: nothing that can move its numbers changed. superseded: something did. no_manifest: the run was kept before manifests.
One of: current, superseded, no_manifest (open: new values may be added).
engine_changedbooleannullableWhether the tester's revision has risen since.
changedarray of objectThe parts that differ, each with its value then (from) and now (to).
partstringengine, release, formula_language, catalogue, attributes, settlement_rules or queries.
tested_atstringWhen the run was kept (RFC 3339); only on a run read back.
format date-time
rule_versionobjectnullableThe saved rule this test was kept under (save, or rule_id): its rule_id, name, version and versions_tried. Null when the test wasn't saved.
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerrequiredThis rule's version under it: the same rule again keeps its version.
min 1
new_versionbooleanThis test made a new version.
versions_triedintegerrequiredVersions saved under it, this one included.
min 1
since_savedobjectnullableThe saved version's record since it was saved (backtester T6, F50), when the test names its rule_id and repeats that version: verdict v2's since_saved check reads it. Null for a new version (its clock starts now) or an unsaved test.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
tests_on_accountobjectnullableRules this account has tried on this competition and market, this one included (the rules-tried ledger: every distinct rule tested on the site, the API or Kelly MCP, kept for good). The not-luck check adjusts for the larger of this count and the request's own tests. Null when the call has no account.
competitionstringrequiredThe competition.
marketstringrequiredThe market.
countintegerrequiredDistinct rules tried.
min 1
shown_countedintegerVariants an earlier report showed (its condition-removed forms, nearby numbers and home or away splits) that joined the count because this rule is one of them (audit F6).
stakingstringHow each bet was staked.
matchedintegerBets taken.
ungradedintegerBets not yet played.
noticestringrequiredThis tests a rule on past games; it isn't a pick.
calibrationobjectnullableoutput: probability only. The formula's chances against the market's no-vig close of the same side (moneyline: the team wins; spread: it covers the bet's line), on the newest 30% of the scored team-games (whole days), and the bets it makes at the edge threshold.
formulastringThe probability formula as sent.
bet_formulastringThe rule it bets: the chance at the bet's own price worth more than the edge threshold (the rule tested, counted and kept).
edge_thresholdnumberExpected return a bet needs, p × decimal price − 1 (0.02 by default).
basisstringnewest_30: scored on the holdout only.
One of: newest_30 (open: new values may be added).
gamesintegerGames scored.
rowsintegerTeam-games scored (both teams of a game).
in_sample_rowsintegerTeam-games before the holdout, not scored.
holdout_fromstringnullableThe holdout's first day.
left_outobjectTeam-games not scored.
pushesintegerPushes and half results.
outside_0_1integerValues below 0 or above 1.
no_valueintegerRows the formula gave no number for.
clippedintegerValues of 0 or 1 clipped to 1e-6 from the edge.
briernumbernullableBrier score of the formula.
brier_marketnumbernullableBrier score of the close.
brier_skillnumbernullable1 − formula / close (above 0: the formula scores better).
log_lossnumbernullableLog loss of the formula.
log_loss_marketnumbernullableLog loss of the close.
log_loss_skillnumbernullable1 − formula / close.
dm_p_valuenumbernullableDiebold–Mariano p on the per-game log-loss difference (two-sided).
dm_statisticnumbernullableThe HLN-corrected statistic (positive: the formula's loss is higher).
mcbnumbernullableCORP miscalibration (Brier).
dscnumbernullableCORP discrimination.
uncnumbernullableUncertainty; brier = mcb − dsc + unc.
mcb_p_valuenumbernullableThe MCB test: the chance a calibrated formula looks this miscalibrated (999 resamples).
reliabilityarray of objectnullableThe CORP reliability curve at up to 20 forecast values.
pnumbernullableThe formula's probability.
observednumbernullableHow often it happened (PAV fit).
lonumbernullable90% consistency band, low.
hinumbernullable90% consistency band, high.
ecenumbernullableExpected calibration error over ten equal-width bins (bin-dependent; for familiarity).
distinct_forecastsintegerDistinct values the formula gave.
slopenumbernullableCalibration slope b in y ~ a + b·logit p (1 is ideal; under 1 overconfident).
slope_senumbernullableIts game-clustered standard error.
interceptnumbernullableCalibration intercept a (0 is ideal).
encompassingobjectnullabley ~ a + b_q·logit q + b_p·logit p, game-clustered: does the formula know something the close doesn't?
b_modelnumbernullableb_p.
b_marketnumbernullableb_q.
se_modelnumbernullableb_p's standard error.
p_valuenumbernullableTwo-sided p of b_p.
edge_binsarray of objectnullableRealized results against the formula's edge.
edgenumbernullableMean of p − q in the bin.
realized_minus_marketnumbernullableMean of y − q in the bin.
rowsintegerTeam-games in the bin.
statusstringThe calibrated check: pass, fail or na (under 200 games, or one value).
One of: pass, fail, na (open: new values may be added).
valuestringThe check's line.
statisticnumbernullableThe encompassing p.
thresholdnumbernullable0.05.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad · snapshot_expired · upstream_timeout · idempotency_key_reused
POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=home",
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"object": "rule_test",
"staking": "flat, 1 unit risked a bet",
"matched": 1,
"ungraded": 0,
"overall": {
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"kelly_fraction": 0.5,
"crossed": false,
"roi_cs95": [
-0.999045,
0.9090450000000001
]
},
"method": "exact_binomial",
"t_stat": null,
"p_value_exact": 0.5235602094240838,
"p_value_clustered": 0.17005720935900537,
"p_value_vs_vig": null,
"null": "break_even",
"sims": null,
"seed": null,
"p_adjusted": 0.5235602094240838,
"adjust_method": "none",
"tests": 1,
"tests_effective": 1,
"reality_check_p": null,
"spa_p": null,
"clears_t3": false,
"haircut": null,
"haircut_roi": null
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F36.2@1"
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"roi_ci": {
"method": "score",
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"bca": null,
"block_bootstrap": null,
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"iid_score": [
-0.6054908095394765,
0.9099999999999999
]
},
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"deflated": {
"sharpe_per_bet": null,
"skew": null,
"kurtosis": null,
"psr": null,
"dsr": null,
"sharpe_expected_max": null,
"roi_bar": null,
"trials_effective": 1,
"sample_needed": null,
"sample_needed_single": null,
"sample_needed_power80": null,
"sample_needed_plan": null
}
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 1,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": 0,
"closes_stale": 0,
"points_coverage": 0,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"path": {
"max_drawdown_units": 0,
"longest_losing_run": 0,
"series": {
"x": "bet_index",
"index": [
1
],
"dates": [
"2025-10-21"
],
"units": [
0.91
],
"drawdown": [
0
],
"base_units": null,
"band_p025": null,
"band_p975": null,
"rolling_roi_100": [
null
],
"rolling_p025": [
null
],
"rolling_p975": [
null
],
"markers": [
{
"type": "season",
"at": 1,
"label": "2025-26"
}
]
},
"downsampled": false,
"expected_max_drawdown_zero_edge": null
},
"seasons": [
{
"season": "2025-26",
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
0.9099999999999999,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0500500000000001,
"p_value": 0.9523356030665205,
"crossed": false
}
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"clv_mean": null
}
],
"holdout": null,
"bets": [
{
"key": "gm_1hnx5nrmk4xjpz6fzyfz:tm_b4w5y9xy5f9sgg1cmkvy",
"starts_at": "2025-10-21T23:30:00Z",
"season": "2025-26",
"label": "2025-10-21 tm_b4w5y9xy5f9sgg1cmkvy vs tm_205kvyzpxcr3et677thx",
"price": {
"decimal": 1.91
},
"line": -4.5,
"result": "win",
"close_prob": null,
"units_to_date": 0.9099999999999999,
"how_graded": "Home -4.5 at the close; final 110–100: win",
"settlement": [
"final_score"
],
"units": 0.9099999999999999,
"drawdown_to_date": 0,
"side": "home",
"market": "spread",
"bet_moment": "close",
"clv_prob": null,
"sample": null,
"why": [
{
"clause": "home",
"value": true
}
]
}
],
"dependence": {
"cluster_by": [
"slate",
"team_season"
],
"slate": "date",
"clusters": 1,
"design_effect": 1,
"effective_bets": 1,
"method": "none",
"se": null,
"se_iid": null
},
"honest": {
"roi": -0.04514385341429974,
"ci80": [
-0.07077022186419332,
-0.019517484964406154
],
"prior_mean": -0.04545450206611379,
"prior_sd": 0.02,
"shrink": 0.999560632688928,
"posterior_sd": 0.019995605844174143,
"selection_adj": 0,
"decay_factor": 0.74,
"stabilizes_at_bets": 2275,
"tries": 1,
"basis_rules": 0,
"raw_roi": 0.9099999999999999,
"uses_clv": false,
"prior_basis": "seed",
"prior": {
"source": "margin",
"mean": -0.04545450206611379,
"closes": 0,
"capped": false,
"words": "shrunk toward the market's usual margin of -4.5%"
}
},
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 1,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": null,
"source": "market",
"status": "na",
"why": "the rule has no who-or-when condition to filter, so its base is the whole market"
},
"stress": {
"half_point_worse": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838
},
"flat_110": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9090909090909092,
"roi": 0.9090909090909092,
"p_value": 0.5238095238095238
},
"other_side": {
"bets": 1,
"decisions": 1,
"record": "0-1",
"units": -1,
"roi": -1,
"p_value": 1
},
"break_even": {
"price_multiplier": 0,
"american_avg": null,
"decimal_avg": null,
"line_points": 3
},
"slippage_mc": null,
"grid": [
{
"vig": -105,
"line_shift": 0,
"units": 0.9523809523809523,
"roi": 0.9523809523809523
},
{
"vig": -105,
"line_shift": -0.5,
"units": 0.9523809523809523,
"roi": 0.9523809523809523
},
{
"vig": -105,
"line_shift": -1,
"units": 0.9523809523809523,
"roi": 0.9523809523809523
},
{
"vig": -110,
"line_shift": 0,
"units": 0.9090909090909092,
"roi": 0.9090909090909092
},
{
"vig": -110,
"line_shift": -0.5,
"units": 0.9090909090909092,
"roi": 0.9090909090909092
},
{
"vig": -110,
"line_shift": -1,
"units": 0.9090909090909092,
"roi": 0.9090909090909092
},
{
"vig": -115,
"line_shift": 0,
"units": 0.8695652173913042,
"roi": 0.8695652173913042
},
{
"vig": -115,
"line_shift": -0.5,
"units": 0.8695652173913042,
"roi": 0.8695652173913042
},
{
"vig": -115,
"line_shift": -1,
"units": 0.8695652173913042,
"roi": 0.8695652173913042
},
{
"vig": -120,
"line_shift": 0,
"units": 0.8333333333333335,
"roi": 0.8333333333333335
},
{
"vig": -120,
"line_shift": -0.5,
"units": 0.8333333333333335,
"roi": 0.8333333333333335
},
{
"vig": -120,
"line_shift": -1,
"units": 0.8333333333333335,
"roi": 0.8333333333333335
}
]
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 3,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [
{
"rule": "final_score",
"bets": 1
}
],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 1,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 1,
"line_sources": {
"sheet_line": 1
}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 3
},
{
"step": "one_bet_a_game",
"rows": 3,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 1,
"out": {
"no_line": 2
}
},
{
"step": "graded",
"rows": 1,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 1,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 3,
"bets": 1,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 2,
"kept": 1,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 1,
"window": 50,
"rolling": [],
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "fewer than (100) graded bets"
},
"ablation": null,
"nearby": {
"numbers": [],
"variants": 0,
"distinct_bet_sets": 0,
"grid": null,
"roi_median": null,
"roi_p05": null,
"roi_p95": null,
"plateau_share": null,
"spike_ratio": null,
"reading": null,
"na": "the rule compares no numbers"
},
"monte_carlo": null,
"noise": {
"seed": 3397954809,
"inputs": null,
"inputs_na": "the rule compares no numbers",
"lines": {
"runs": 200,
"profitable_runs": 200,
"profitable_share": 1,
"roi_median": 0.9099999999999999,
"roi_p05": 0.9099999999999999,
"line_prob": 0.5
},
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 1,
"units": 0.9099999999999999,
"by": [
{
"group": "season",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "team",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "opponent",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "month",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "weekday",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
}
],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": true,
"seed": 769905824,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"verdict": {
"version": 2,
"headline": "Too few bets to judge",
"level": "too_few",
"passed": 0,
"applicable": 2,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": [
{
"key": "enough_bets",
"group": "real",
"label": "Enough bets",
"status": "fail",
"zone": "red",
"value": "(1) decisions are too few to size the sample",
"statistic": 1,
"threshold": null
},
{
"key": "not_luck",
"group": "real",
"label": "Not just luck",
"status": "fail",
"zone": "red",
"value": "p 0.524",
"statistic": 0.5235602094240838,
"threshold": 0.05
},
{
"key": "beats_base",
"group": "real",
"label": "Beats its own base",
"status": "na",
"zone": "na",
"value": "the rule has no who-or-when condition to filter, so its base is the whole market",
"statistic": null,
"threshold": null
},
{
"key": "half_point_worse",
"group": "pressure",
"label": "Holds ½ point worse",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"group": "pressure",
"label": "Nearby thresholds hold",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"group": "pressure",
"label": "Not one lucky stretch",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "beats_close",
"group": "still_working",
"label": "Beats the close",
"status": "na",
"zone": "na",
"value": "bets at the close can't beat it",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"group": "still_working",
"label": "Still working",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "since_saved",
"group": "still_working",
"label": "Since saved",
"status": "na",
"zone": "na",
"value": "(0) of (30) bets since it was saved",
"statistic": null,
"threshold": null
}
],
"held_back_by": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 6899137762779026,
"lose_all_ever": 0,
"lose_all_ever_allowing_error": 0.277131008418,
"ruin_ever_if_rate_exact": 0,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": [],
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": 1,
"push_rate": 0,
"full_fraction": null,
"shrink": 1,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"zero_because": "too_few",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 100.91,
"units": 0.9099999999999966,
"max_drawdown": 0,
"growth_per_bet": 0.009058844488345926,
"roi": 0.9099999999999966
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101,
"units": 1,
"max_drawdown": 0,
"growth_per_bet": 0.009950330853168092,
"roi": 0.9099999999999999
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101.82,
"units": 1.8199999999999932,
"max_drawdown": 0,
"growth_per_bet": 0.018036362486098575,
"roi": 0.9099999999999966,
"stake_fraction": 0.02
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"bets": 1,
"sharpe_per_bet": null,
"sharpe_daily": null,
"sharpe_per_season": null,
"sortino_per_bet": null,
"profit_factor": null,
"expected_units_per_bet": 0.9099999999999999,
"avg_win_units": 0.9099999999999999,
"avg_loss_units": null,
"max_drawdown_units": 0,
"return_on_drawdown": null,
"recovery_factor": null,
"ulcer_index": 0,
"martin_ratio": null,
"sqn": null,
"probabilistic_sharpe": null,
"longest_underwater_bets": 0,
"longest_underwater_days": 0,
"current_drawdown_units": 0,
"sharpe_per_bet_ci95": null,
"sortino_per_bet_ci95": null,
"profit_factor_ci95": null,
"warnings": [
"no_losing_bets",
"no_spread_in_returns",
"small_sample"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "spreads",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured spreads limits for nba at this moment yet."
},
"splits": [
{
"label": "Home",
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"p_adjusted": 1,
"shown_rows": 2
},
{
"label": "Favorite",
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"p_adjusted": 1,
"shown_rows": 2
}
],
"without": [],
"rule": {
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 2,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 1,
"shown_counted": 0
},
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:13e69ec9c84eb683918ee67e91ae21ac69cc813cfbc4758f625524596be5d3da",
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"seed": 18147943352127158000
},
"sources": {
"prices": {
"sheet_line": 1
},
"inputs": [
{
"name": "home",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 2,
"stale_price": 0
},
"sha256": "sha256:d78d99c7954d4b2367b7baa69328756b5aa4464bb52ddcb24cdf05a0484557ad"
},
"run_id": "run_mkytv75qe5a297z7y3af",
"rule_version": null,
"report_url": null,
"summary": "Too few bets to judge: honest forward ROI -4.5%, 1-0 against the spread, 2025-26, +0.9 units, +91.0% ROI past (0 of 2 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_bc8483563e5ff5a2ff264826",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=AND(home, points_for > 100)",
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"type": "https://docs.askkelly.ai/problems/rule_not_testable",
"title": "Rule can't be tested",
"status": 422,
"code": "rule_not_testable",
"detail": "points_for is this game's own result, known only after it ends. For past games use a window, such as ROLLING_AVG(points_for, 4) or LAG(points_for).",
"param": "formula",
"doc_url": "https://docs.askkelly.ai/problems/rule_not_testable",
"request_id": "req_00111c5c6d3369550070d21a",
"retryable": false,
"display_message": "That rule can't be tested on past games as written.",
"problems": [
{
"code": "look_ahead",
"path": "points_for",
"message": "points_for is this game's own result, known only after it ends. For past games use a window, such as ROLLING_AVG(points_for, 4) or LAG(points_for).",
"fixes": [
{
"label": "Use the game before: LAG(points_for)",
"formula": "=AND(home, LAG(points_for) > 100)"
},
{
"label": "Use the last 4 games: ROLLING_AVG(points_for, 4)",
"formula": "=AND(home, ROLLING_AVG(points_for, 4) > 100)"
}
]
}
]
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=AND(home, rest_days >= 1)",
"competition": "nba",
"seasons": [
"2025-26"
],
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"detail": "summary"
}
{
"object": "rule_test",
"staking": "flat, 1 unit risked a bet",
"matched": 0,
"ungraded": 0,
"overall": {
"wins": 0,
"losses": 0,
"pushes": 0,
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"win_rate": null,
"break_even": null,
"significance": null,
"few_decisions": true,
"small_sample": true,
"deflated": null
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 0,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": null,
"closes_stale": 0,
"points_coverage": null,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"path": {
"max_drawdown_units": 0,
"longest_losing_run": 0
},
"seasons": [],
"holdout": null,
"dependence": null,
"honest": null,
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 0,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": "=home",
"source": "auto",
"decisions": 1,
"status": "na",
"why": "the base is under twice the rule's bets"
},
"stress": {
"half_point_worse": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"flat_110": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"other_side": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"break_even": null,
"slippage_mc": null
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 2,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 0,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 0,
"line_sources": {}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 2
},
{
"step": "one_bet_a_game",
"rows": 2,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 0,
"out": {
"no_line": 2
}
},
{
"step": "graded",
"rows": 0,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 0,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 2,
"bets": 0,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
},
{
"attribute": "rest_days",
"rows": 6,
"rows_blank": 2,
"share": 0.3333,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 2,
"kept": 0,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 0,
"window": null,
"rolling": [],
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "no graded bets"
},
"ablation": {
"base": {
"decisions": 2,
"roi": -0.04500000000000004
},
"value": "excess_units",
"total": 0,
"rule": {
"decisions": 0,
"units": 0
},
"clauses": [
{
"condition": "home",
"shapley_units": 0.4775,
"label": "dead"
},
{
"condition": "rest_days >= 1",
"shapley_units": -0.4775,
"label": "dead"
}
],
"pairs": [],
"method": "exact",
"evaluations": 3
},
"nearby": {
"numbers": [],
"variants": 0,
"distinct_bet_sets": 0,
"grid": null,
"roi_median": null,
"roi_p05": null,
"roi_p95": null,
"plateau_share": null,
"spike_ratio": null,
"reading": null,
"na": "the rule compares no numbers"
},
"monte_carlo": null,
"noise": {
"seed": 1547856035,
"inputs": {
"runs": 200,
"profitable_runs": 0,
"profitable_share": 0,
"roi_median": null,
"roi_p05": null,
"input_scale": 0.1,
"overlap_median": 1,
"numbers": 1
},
"inputs_na": null,
"lines": null,
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 0,
"units": 0,
"by": [],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": true,
"seed": 789056186,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"verdict": {
"version": 2,
"headline": "No bets",
"level": "too_few",
"passed": 0,
"applicable": 0,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 7881387689213470,
"lose_all_ever": null,
"lose_all_ever_allowing_error": null,
"ruin_ever_if_rate_exact": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": null,
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": null,
"push_rate": 0,
"full_fraction": null,
"shrink": null,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"sharpe_per_bet": null,
"profit_factor": null,
"warnings": [
"no_bets"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "spreads",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured spreads limits for nba at this moment yet."
},
"splits": null,
"without": null,
"rule": {
"formula": "=AND(home, rest_days >= 1)",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
},
{
"name": "rest_days",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 2,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 2,
"shown_counted": 0
},
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:31b058b095b203fde86de3ba2f6cd5169d6ca0e8e27b7e910fd88cee4045e3d9",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"seed": 11684609526682410000
},
"sources": {
"prices": {},
"inputs": [
{
"name": "home",
"known": "pregame"
},
{
"name": "rest_days",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 2,
"stale_price": 0
},
"sha256": "sha256:bb85d223f5ef4254dbaebc41ec6fecc5b89a3785ff34c18ded93c58b39409306"
},
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"rule_version": {
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"version": 2,
"new_version": true,
"versions_tried": 2
},
"report_url": null,
"summary": "The rule matched no game Kelly holds a line for, 2025-26.",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_1cb31f79b11a161aa60974c0",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=IF(home, 0.6, 0.4)",
"competition": "nba",
"market": "moneyline",
"seasons": [
"2025-26"
],
"output": "probability",
"edge_threshold": 0.02
}
{
"object": "rule_test",
"staking": "flat, 1 unit risked a bet",
"matched": 0,
"ungraded": 0,
"overall": {
"wins": 0,
"losses": 0,
"pushes": 0,
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"win_rate": null,
"break_even": null,
"significance": null,
"few_decisions": true,
"small_sample": true,
"deflated": null
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 0,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": null,
"closes_stale": 0,
"points_coverage": null,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"path": {
"max_drawdown_units": 0,
"longest_losing_run": 0,
"series": null,
"downsampled": false,
"expected_max_drawdown_zero_edge": null
},
"seasons": [],
"holdout": null,
"bets": [],
"dependence": null,
"honest": null,
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 0,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": null,
"source": "market",
"status": "na",
"why": "the rule has no who-or-when condition to filter, so its base is the whole market"
},
"stress": {
"half_point_worse": null,
"flat_110": null,
"other_side": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"break_even": null,
"slippage_mc": null,
"grid": [
{
"price_multiplier": 1,
"units": 0,
"roi": null
},
{
"price_multiplier": 0.98,
"units": 0,
"roi": null
},
{
"price_multiplier": 0.96,
"units": 0,
"roi": null
},
{
"price_multiplier": 0.94,
"units": 0,
"roi": null
}
]
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 0,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "moneyline_tie_push",
"text": "A two-way moneyline on a game that ends tied (an NFL tie) pushes.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 0,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 0,
"line_sources": {}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 0
},
{
"step": "one_bet_a_game",
"rows": 0,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 0,
"out": {
"no_line": 0
}
},
{
"step": "graded",
"rows": 0,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 0,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 0,
"bets": 0,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 0,
"kept": 0,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 0,
"window": null,
"rolling": [],
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "no graded bets"
},
"ablation": null,
"nearby": {
"na": "the rule's numbers can't be moved one at a time",
"numbers": [],
"variants": 0
},
"monte_carlo": null,
"noise": {
"seed": 3363965115,
"inputs": null,
"inputs_na": "the rule's numbers can't be jittered one at a time",
"lines": null,
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 0,
"units": 0,
"by": [],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": false,
"seed": 724149288,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"calibration": {
"formula": "=IF(home, 0.6, 0.4)",
"bet_formula": "=(IF(home, 0.6, 0.4)) * TO_DECIMAL(moneyline_close) - 1 > 0.02",
"edge_threshold": 0.02,
"basis": "newest_30",
"games": 0,
"rows": 0,
"left_out": {
"pushes": 0,
"outside_0_1": 0,
"no_value": 0
},
"clipped": 0,
"in_sample_rows": 0,
"holdout_from": null,
"status": "na",
"value": "too few games held out to judge (0; 200 needed)",
"statistic": null,
"threshold": 0.05
},
"verdict": {
"version": 2,
"headline": "No bets",
"level": "too_few",
"passed": 0,
"applicable": 0,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 86435566430832,
"lose_all_ever": null,
"lose_all_ever_allowing_error": null,
"ruin_ever_if_rate_exact": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": [],
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": null,
"push_rate": 0,
"full_fraction": null,
"shrink": null,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"bets": 0,
"warnings": [
"no_bets"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "h2h",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured h2h limits for nba at this moment yet."
},
"splits": [],
"without": [],
"rule": {
"formula": "=(IF(home, 0.6, 0.4)) * TO_DECIMAL(moneyline_close) - 1 > 0.02",
"competition": "nba",
"market": "moneyline",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:cc8c716e7628b75866e3e8ec2057957dbfb69aeff28266f9b6dc4feefacea17a",
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"left_out": {
"no_line": 0,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 0,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (0).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": {
"competition": "nba",
"market": "moneyline",
"count": 1,
"shown_counted": 0
},
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:0502a9b168695c1ecd9975e4036d8096ecf5edb09bb05ca4661781ac73b628c0",
"rule_hash": "sha256:cc8c716e7628b75866e3e8ec2057957dbfb69aeff28266f9b6dc4feefacea17a",
"seed": 4199034409599712000
},
"sources": {
"prices": {},
"inputs": [
{
"name": "home",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 0,
"stale_price": 0
},
"sha256": "sha256:837903c84f3848f6b9367acf695f7f303b14225bd1f17ac2a662ab539fe7c2e7"
},
"run_id": "run_y7vwhyx3err5m5j5grdx",
"rule_version": null,
"report_url": null,
"summary": "The rule matched no game Kelly holds a line for, 2025-26.",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_d5bec7bcae67a2c0b10ec009",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=0.5",
"competition": "nba",
"market": "over",
"seasons": [
"2025-26"
],
"output": "probability"
}
{
"type": "https://docs.askkelly.ai/problems/rule_not_testable",
"title": "Rule can't be tested",
"status": 422,
"code": "rule_not_testable",
"detail": "A probability formula bets a moneyline or a spread, not over.",
"param": "market",
"doc_url": "https://docs.askkelly.ai/problems/rule_not_testable",
"request_id": "req_1af92e34745ad20f1b131054",
"retryable": false,
"display_message": "That rule can't be tested on past games as written.",
"problems": [
{
"code": "probability_market",
"path": null,
"message": "A probability formula bets a moneyline or a spread, not over.",
"fixes": [
"Use market: moneyline (the chance the team wins) or spread (the chance it covers)."
]
}
]
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=home",
"competition": "nba",
"seasons": [
"2025-26"
],
"save": true,
"name": "Home teams",
"detail": "summary"
}
{
"object": "rule_test",
"staking": "flat, 1 unit risked a bet",
"matched": 1,
"ungraded": 0,
"overall": {
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"kelly_fraction": 0.5,
"crossed": false,
"roi_cs95": [
-0.999045,
0.9090450000000001
]
},
"method": "exact_binomial",
"t_stat": null,
"p_value_exact": 0.5235602094240838,
"p_value_clustered": 0.17005720935900537,
"p_value_vs_vig": null,
"null": "break_even",
"sims": null,
"seed": null,
"p_adjusted": 0.5235602094240838,
"adjust_method": "none",
"tests": 1,
"tests_effective": 1,
"reality_check_p": null,
"spa_p": null,
"clears_t3": false,
"haircut": null,
"haircut_roi": null
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F36.2@1"
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"roi_ci": {
"method": "score",
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"bca": null,
"block_bootstrap": null,
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"iid_score": [
-0.6054908095394765,
0.9099999999999999
]
},
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"deflated": {
"sharpe_per_bet": null,
"skew": null,
"kurtosis": null,
"psr": null,
"dsr": null,
"sharpe_expected_max": null,
"roi_bar": null,
"trials_effective": 1,
"sample_needed": null,
"sample_needed_single": null,
"sample_needed_power80": null,
"sample_needed_plan": null
}
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 1,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": 0,
"closes_stale": 0,
"points_coverage": 0,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"path": {
"max_drawdown_units": 0,
"longest_losing_run": 0
},
"seasons": [
{
"season": "2025-26",
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
0.9099999999999999,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0500500000000001,
"p_value": 0.9523356030665205,
"crossed": false
}
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"clv_mean": null
}
],
"holdout": null,
"dependence": {
"cluster_by": [
"slate",
"team_season"
],
"slate": "date",
"clusters": 1,
"design_effect": 1,
"effective_bets": 1,
"method": "none",
"se": null,
"se_iid": null
},
"honest": {
"roi": -0.04514385341429974,
"ci80": [
-0.07077022186419332,
-0.019517484964406154
],
"prior_mean": -0.04545450206611379,
"prior_sd": 0.02,
"shrink": 0.999560632688928,
"posterior_sd": 0.019995605844174143,
"selection_adj": 0,
"decay_factor": 0.74,
"stabilizes_at_bets": 2275,
"tries": 1,
"basis_rules": 0,
"raw_roi": 0.9099999999999999,
"uses_clv": false,
"prior_basis": "seed",
"prior": {
"source": "margin",
"mean": -0.04545450206611379,
"closes": 0,
"capped": false,
"words": "shrunk toward the market's usual margin of -4.5%"
}
},
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 1,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": null,
"source": "market",
"status": "na",
"why": "the rule has no who-or-when condition to filter, so its base is the whole market"
},
"stress": {
"half_point_worse": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838
},
"flat_110": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9090909090909092,
"roi": 0.9090909090909092,
"p_value": 0.5238095238095238
},
"other_side": {
"bets": 1,
"decisions": 1,
"record": "0-1",
"units": -1,
"roi": -1,
"p_value": 1
},
"break_even": {
"price_multiplier": 0,
"american_avg": null,
"decimal_avg": null,
"line_points": 3
},
"slippage_mc": null
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 3,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [
{
"rule": "final_score",
"bets": 1
}
],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 1,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 1,
"line_sources": {
"sheet_line": 1
}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 3
},
{
"step": "one_bet_a_game",
"rows": 3,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 1,
"out": {
"no_line": 2
}
},
{
"step": "graded",
"rows": 1,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 1,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 3,
"bets": 1,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 2,
"kept": 1,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 1,
"window": 50,
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "fewer than (100) graded bets"
},
"ablation": null,
"nearby": {
"numbers": [],
"variants": 0,
"distinct_bet_sets": 0,
"grid": null,
"roi_median": null,
"roi_p05": null,
"roi_p95": null,
"plateau_share": null,
"spike_ratio": null,
"reading": null,
"na": "the rule compares no numbers"
},
"monte_carlo": null,
"noise": {
"seed": 3397954809,
"inputs": null,
"inputs_na": "the rule compares no numbers",
"lines": {
"runs": 200,
"profitable_runs": 200,
"profitable_share": 1,
"roi_median": 0.9099999999999999,
"roi_p05": 0.9099999999999999,
"line_prob": 0.5
},
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 1,
"units": 0.9099999999999999,
"by": [
{
"group": "season",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "team",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
}
],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": true,
"seed": 769905824,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"verdict": {
"version": 2,
"headline": "Too few bets to judge",
"level": "too_few",
"passed": 0,
"applicable": 2,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": [
{
"key": "enough_bets",
"group": "real",
"label": "Enough bets",
"status": "fail",
"zone": "red",
"value": "(1) decisions are too few to size the sample",
"statistic": 1,
"threshold": null
},
{
"key": "not_luck",
"group": "real",
"label": "Not just luck",
"status": "fail",
"zone": "red",
"value": "p 0.524",
"statistic": 0.5235602094240838,
"threshold": 0.05
},
{
"key": "beats_base",
"group": "real",
"label": "Beats its own base",
"status": "na",
"zone": "na",
"value": "the rule has no who-or-when condition to filter, so its base is the whole market",
"statistic": null,
"threshold": null
},
{
"key": "half_point_worse",
"group": "pressure",
"label": "Holds ½ point worse",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"group": "pressure",
"label": "Nearby thresholds hold",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"group": "pressure",
"label": "Not one lucky stretch",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "beats_close",
"group": "still_working",
"label": "Beats the close",
"status": "na",
"zone": "na",
"value": "bets at the close can't beat it",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"group": "still_working",
"label": "Still working",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "since_saved",
"group": "still_working",
"label": "Since saved",
"status": "na",
"zone": "na",
"value": "(0) of (30) bets since it was saved",
"statistic": null,
"threshold": null
}
],
"held_back_by": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 6899137762779026,
"lose_all_ever": 0,
"lose_all_ever_allowing_error": 0.277131008418,
"ruin_ever_if_rate_exact": 0,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": null,
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": 1,
"push_rate": 0,
"full_fraction": null,
"shrink": 1,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"zero_because": "too_few",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 100.91,
"units": 0.9099999999999966,
"max_drawdown": 0,
"growth_per_bet": 0.009058844488345926,
"roi": 0.9099999999999966
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101,
"units": 1,
"max_drawdown": 0,
"growth_per_bet": 0.009950330853168092,
"roi": 0.9099999999999999
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101.82,
"units": 1.8199999999999932,
"max_drawdown": 0,
"growth_per_bet": 0.018036362486098575,
"roi": 0.9099999999999966,
"stake_fraction": 0.02
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"sharpe_per_bet": null,
"profit_factor": null,
"warnings": [
"no_losing_bets",
"no_spread_in_returns",
"small_sample"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "spreads",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured spreads limits for nba at this moment yet."
},
"splits": null,
"without": null,
"rule": {
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 2,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 1,
"shown_counted": 0
},
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:cce3946d8c33952c9cdbbfee66781fa4c6308af5f93c465426093a4f102255e8",
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"seed": 18147943352127158000
},
"sources": {
"prices": {
"sheet_line": 1
},
"inputs": [
{
"name": "home",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 2,
"stale_price": 0
},
"sha256": "sha256:5cb4599cac45cd1b6bc1127162da0ff9c2214e7bb3bc3a9a351625c56bc99e51"
},
"run_id": "run_rc69h189x95wy5rhfet2",
"rule_version": {
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"version": 1,
"new_version": true,
"versions_tried": 1
},
"report_url": null,
"summary": "Too few bets to judge: honest forward ROI -4.5%, 1-0 against the spread, 2025-26, +0.9 units, +91.0% ROI past (0 of 2 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_a8e97cdf6c3247af334f13d4",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:test
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=home",
"competition": "nba",
"seasons": [
"2025-26"
],
"detail": "summary",
"transfer": true,
"transfer_leagues": [
"ncaab",
"nfl"
]
}
{
"object": "rule_test",
"staking": "flat, 1 unit risked a bet",
"matched": 1,
"ungraded": 0,
"overall": {
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": {
"tests": 2,
"bonferroni_threshold": 0.025,
"bonferroni_survives": false,
"bonferroni_adjusted_p": 1,
"benjamini_hochberg_survives_alone": false,
"familywise_error_uncorrected": 0.09750000000000003
},
"anytime_valid": {
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"kelly_fraction": 0.5,
"crossed": false,
"roi_cs95": [
-0.999045,
0.9090450000000001
]
},
"method": "exact_binomial",
"t_stat": null,
"p_value_exact": 0.5235602094240838,
"p_value_clustered": 0.17005720935900537,
"p_value_vs_vig": null,
"null": "break_even",
"sims": null,
"seed": null,
"p_adjusted": 1,
"adjust_method": "holm",
"tests": 2,
"tests_effective": 2,
"reality_check_p": null,
"spa_p": null,
"clears_t3": false,
"haircut": 0.567948481856096,
"haircut_roi": 0.3931668815109526
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F12.8@1",
"F36.2@1"
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"roi_ci": {
"method": "score",
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"bca": null,
"block_bootstrap": null,
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"iid_score": [
-0.6054908095394765,
0.9099999999999999
]
},
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"deflated": {
"sharpe_per_bet": null,
"skew": null,
"kurtosis": null,
"psr": null,
"dsr": null,
"sharpe_expected_max": null,
"roi_bar": null,
"trials_effective": 2,
"sample_needed": null,
"sample_needed_single": null,
"sample_needed_power80": null,
"sample_needed_plan": null
}
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 1,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": 0,
"closes_stale": 0,
"points_coverage": 0,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"path": {
"max_drawdown_units": 0,
"longest_losing_run": 0
},
"seasons": [
{
"season": "2025-26",
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
0.9099999999999999,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0500500000000001,
"p_value": 0.9523356030665205,
"crossed": false
}
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"clv_mean": null
}
],
"holdout": null,
"dependence": {
"cluster_by": [
"slate",
"team_season"
],
"slate": "date",
"clusters": 1,
"design_effect": 1,
"effective_bets": 1,
"method": "none",
"se": null,
"se_iid": null
},
"honest": {
"roi": -0.04532667665457775,
"ci80": [
-0.07095304510447134,
-0.019700308204684165
],
"prior_mean": -0.04545450206611379,
"prior_sd": 0.02,
"shrink": 0.999560632688928,
"posterior_sd": 0.019995605844174143,
"selection_adj": -0.00018282324027801353,
"decay_factor": 0.74,
"stabilizes_at_bets": 2275,
"tries": 2,
"basis_rules": 0,
"raw_roi": 0.9099999999999999,
"uses_clv": false,
"prior_basis": "seed",
"prior": {
"source": "margin",
"mean": -0.04545450206611379,
"closes": 0,
"capped": false,
"words": "shrunk toward the market's usual margin of -4.5%"
}
},
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 1,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": null,
"source": "market",
"status": "na",
"why": "the rule has no who-or-when condition to filter, so its base is the whole market"
},
"stress": {
"half_point_worse": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838
},
"flat_110": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9090909090909092,
"roi": 0.9090909090909092,
"p_value": 0.5238095238095238
},
"other_side": {
"bets": 1,
"decisions": 1,
"record": "0-1",
"units": -1,
"roi": -1,
"p_value": 1
},
"break_even": {
"price_multiplier": 0,
"american_avg": null,
"decimal_avg": null,
"line_points": 3
},
"slippage_mc": null
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 3,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [
{
"rule": "final_score",
"bets": 1
}
],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 1,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 1,
"line_sources": {
"sheet_line": 1
}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 3
},
{
"step": "one_bet_a_game",
"rows": 3,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 1,
"out": {
"no_line": 2
}
},
{
"step": "graded",
"rows": 1,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 1,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 3,
"bets": 1,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 2,
"kept": 1,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 1,
"window": 50,
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "fewer than (100) graded bets"
},
"ablation": null,
"nearby": {
"numbers": [],
"variants": 0,
"distinct_bet_sets": 0,
"grid": null,
"roi_median": null,
"roi_p05": null,
"roi_p95": null,
"plateau_share": null,
"spike_ratio": null,
"reading": null,
"na": "the rule compares no numbers"
},
"monte_carlo": null,
"noise": {
"seed": 3397954809,
"inputs": null,
"inputs_na": "the rule compares no numbers",
"lines": {
"runs": 200,
"profitable_runs": 200,
"profitable_share": 1,
"roi_median": 0.9099999999999999,
"roi_p05": 0.9099999999999999,
"line_prob": 0.5
},
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 1,
"units": 0.9099999999999999,
"by": [
{
"group": "season",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
},
{
"group": "team",
"groups": 1,
"best": null,
"without_best": null,
"without_best_two": null,
"effective_groups": 1,
"p_more_concentrated": null,
"jackknife_roi_se": null
}
],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": true,
"seed": 769905824,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"verdict": {
"version": 2,
"headline": "Too few bets to judge",
"level": "too_few",
"passed": 0,
"applicable": 2,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": [
{
"key": "enough_bets",
"group": "real",
"label": "Enough bets",
"status": "fail",
"zone": "red",
"value": "(1) decisions are too few to size the sample",
"statistic": 1,
"threshold": null
},
{
"key": "not_luck",
"group": "real",
"label": "Not just luck",
"status": "fail",
"zone": "red",
"value": "p 1.000 after (2) rules tried",
"statistic": 1,
"threshold": 0.05
},
{
"key": "beats_base",
"group": "real",
"label": "Beats its own base",
"status": "na",
"zone": "na",
"value": "the rule has no who-or-when condition to filter, so its base is the whole market",
"statistic": null,
"threshold": null
},
{
"key": "half_point_worse",
"group": "pressure",
"label": "Holds ½ point worse",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"group": "pressure",
"label": "Nearby thresholds hold",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"group": "pressure",
"label": "Not one lucky stretch",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "beats_close",
"group": "still_working",
"label": "Beats the close",
"status": "na",
"zone": "na",
"value": "bets at the close can't beat it",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"group": "still_working",
"label": "Still working",
"status": "na",
"zone": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "since_saved",
"group": "still_working",
"label": "Since saved",
"status": "na",
"zone": "na",
"value": "(0) of (30) bets since it was saved",
"statistic": null,
"threshold": null
}
],
"held_back_by": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 6899137762779026,
"lose_all_ever": 0,
"lose_all_ever_allowing_error": 0.277131008418,
"ruin_ever_if_rate_exact": 0,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": null,
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": 1,
"push_rate": 0,
"full_fraction": null,
"shrink": 1,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"zero_because": "too_few",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 100.91,
"units": 0.9099999999999966,
"max_drawdown": 0,
"growth_per_bet": 0.009058844488345926,
"roi": 0.9099999999999966
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101,
"units": 1,
"max_drawdown": 0,
"growth_per_bet": 0.009950330853168092,
"roi": 0.9099999999999999
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 1,
"end_bankroll_units": 101.82,
"units": 1.8199999999999932,
"max_drawdown": 0,
"growth_per_bet": 0.018036362486098575,
"roi": 0.9099999999999966,
"stake_fraction": 0.02
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"sharpe_per_bet": null,
"profit_factor": null,
"warnings": [
"no_losing_bets",
"no_spread_in_returns",
"small_sample"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "spreads",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured spreads limits for nba at this moment yet."
},
"transfer": {
"options": {
"leagues": [
"ncaab",
"nfl"
],
"markets": "auto",
"thresholds": "both"
},
"status": "measured",
"why": null,
"cells": [
{
"competition": "nba",
"market": "moneyline",
"thresholds": "literal",
"formula": "=home",
"seasons": [
"2025-26"
],
"pooled": false,
"decisions": 1,
"bets": 1,
"record": "1-0",
"units": 0.6,
"roi": 0.6,
"roi_ci95": null,
"se": null,
"p_value": 0.625,
"p_holm": 0.625
},
{
"competition": "nfl",
"market": "spread",
"thresholds": "literal",
"formula": "=home",
"seasons": [
"2025"
],
"pooled": false,
"decisions": 0,
"bets": 0,
"record": "0-0",
"units": 0,
"roi": null,
"roi_ci95": null,
"se": null,
"p_value": null,
"p_holm": null
}
],
"pooled": null,
"same_direction": null,
"not_testable": [
{
"competition": "ncaab",
"market": "spread",
"missing": [],
"why": "Kelly holds no games for these seasons"
}
],
"note": "Other leagues answer a different question: this is evidence beside the verdict, never part of it, and these runs don't count toward your rules tried."
},
"splits": null,
"without": null,
"rule": {
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 2,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 2,
"shown_counted": 0
},
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:e34109992aadb32980f9ac83c6cc89e10bd44348bc5d2f112d605e274426126b",
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"seed": 18147943352127158000
},
"sources": {
"prices": {
"sheet_line": 1
},
"inputs": [
{
"name": "home",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 2,
"stale_price": 0
},
"sha256": "sha256:e4dee7803798bc56f8675cb900f10ae66d38314a8ec9f7ed827cbe55896fde52"
},
"run_id": "run_43g3m0cfryzcxkcb7sfv",
"rule_version": null,
"report_url": null,
"summary": "Too few bets to judge: honest forward ROI -4.5%, 1-0 against the spread, 2025-26, +0.9 units, +91.0% ROI past (0 of 2 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_379ea34707419be658325b02",
"live_topics": [],
"rg_line": null
}POST /v1/charts:export
public, no key needed · 0 credits, not charged during the beta · MCP tool export_chart · status live
A Kelly chart spec, as a tool returned it under charts, written as Vega-Lite (rows inline), CSV, Python (pandas and Altair), R (ggplot2) or the spec itself: the file's text, media type and name. Every number is the engine's as the spec carried it; columns the spec's rights block (book prices) are left out, and a chart whose rows can't leave askkelly.ai is refused.
Body, application/json (required):
chartstringThe chart: one spec from a tool result's charts list (kcs 1), unchanged, as JSON text.
min length 2 · max length 2000000
formatstringvega_lite, csv, python, r or spec.
One of: vega_lite, csv, python, r, spec (open: new values may be added).
200 The chart export facts.
objectconstrequiredAlways chart_export.
Always "chart_export".
formatstringrequiredspec, vega_lite, csv, python or r.
One of: spec, vega_lite, csv, python, r (open: new values may be added).
media_typestringrequiredThe file's media type (application/json, text/csv, text/x-python, text/x-r).
filenamestringrequiredA file name for it, from the chart's title.
textstringrequiredThe file itself: JSON, CSV or a code block to paste. Every number is the engine's, as the spec carried it.
kindstringrequiredThe chart's catalogue form (equity, funnel, forest, ...).
titlestringrequiredThe chart's title.
rowsintegerrequiredRows written.
min 0
columns_removedarray of stringrequiredColumns left out under the data's licence.
provenanceobjectrequiredWhere the numbers came from: the tool, the engine version, as of and the sources.
toolstringrequiredThe engine tool that computed them.
enginestringrequiredThe engine and its version.
as_ofstringrequiredWhen the numbers stand.
format date-time
sourcesarray of stringrequiredTheir sources.
argsobjectThe tool's public arguments, to ask for the chart again.
chartstringnullableWhich of the tool's charts it is.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large
POST https://api.askkelly.ai/v1/charts:export
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"chart": "{"kcs": 1, "kind": "bars", "title": "ROI by sport", "data": {"key": "sport", "columns": [{"id": "sport", "type": "str", "label": "Sport"}, {"id": "roi", "type": "float", "label": "ROI", "unit": "roi", "format": "+.1%"}, {"id": "lo95", "type": "float", "unit": "roi"}, {"id": "hi95", "type": "float", "unit": "roi"}, {"id": "n_bets", "type": "int", "label": "Bets"}], "rows": [{"sport": "NFL", "roi": 0.041, "lo95": -0.012, "hi95": 0.094, "n_bets": 410}, {"sport": "NBA", "roi": -0.022, "lo95": -0.081, "hi95": 0.037, "n_bets": 330}]}, "x": {"field": "roi", "type": "quantitative", "title": "ROI"}, "y": {"field": "sport", "type": "nominal"}, "series": [{"field": "roi", "role": "market", "label": "ROI"}], "intervals": [{"low": "lo95", "high": "hi95", "level": 0.95}], "rules": [{"x": 0, "role": "break_even", "label": "Break-even"}], "count": "n_bets", "summary": "NFL +4.1% over 410 bets and NBA -2.2% over 330; neither can be told from luck yet.", "provenance": {"tool": "get_my_results", "engine": "results 2", "as_of": "2026-10-04T17:30:00Z", "sources": ["bet log"]}}",
"format": "csv"
}
{
"object": "chart_export",
"format": "csv",
"media_type": "text/csv",
"filename": "roi-by-sport.csv",
"text": "sport,roi,lo95,hi95,n_bets\nNFL,0.041,-0.012,0.094,410\nNBA,-0.022,-0.081,0.037,330\n",
"kind": "bars",
"title": "ROI by sport",
"rows": 2,
"columns_removed": [],
"provenance": {
"tool": "get_my_results",
"engine": "results 2",
"as_of": "2026-10-04T17:30:00Z",
"sources": [
"bet log"
]
},
"summary": "ROI by sport as CSV: 2 rows.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_518717293aa676623e5f7334",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/charts:export
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"chart": "{"kcs": 1, "kind": "bars", "title": "ROI by sport", "data": {"key": "sport", "columns": [{"id": "sport", "type": "str", "label": "Sport"}, {"id": "roi", "type": "float", "label": "ROI", "unit": "roi", "format": "+.1%"}, {"id": "lo95", "type": "float", "unit": "roi"}, {"id": "hi95", "type": "float", "unit": "roi"}, {"id": "n_bets", "type": "int", "label": "Bets"}], "rows": [{"sport": "NFL", "roi": 0.041, "lo95": -0.012, "hi95": 0.094, "n_bets": 410}, {"sport": "NBA", "roi": -0.022, "lo95": -0.081, "hi95": 0.037, "n_bets": 330}]}, "x": {"field": "roi", "type": "quantitative", "title": "ROI"}, "y": {"field": "sport", "type": "nominal"}, "series": [{"field": "roi", "role": "market", "label": "ROI"}], "intervals": [{"low": "lo95", "high": "hi95", "level": 0.95}], "rules": [{"x": 0, "role": "break_even", "label": "Break-even"}], "count": "n_bets", "summary": "NFL +4.1% over 410 bets and NBA -2.2% over 330; neither can be told from luck yet.", "provenance": {"tool": "get_my_results", "engine": "results 2", "as_of": "2026-10-04T17:30:00Z", "sources": ["bet log"]}}",
"format": "pdf"
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "format: 'pdf' is not one of ['vega_lite', 'csv', 'python', 'r', 'spec']",
"param": "format",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_521f14c947cee75e1599fd24",
"retryable": false,
"display_message": "Something in that request was not understood."
}POST https://api.askkelly.ai/v1/charts:export
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"chart": "{"kcs": 1, "kind": "bars", "title": "ROI by sport", "data": {"key": "sport", "columns": [{"id": "sport", "type": "str", "label": "Sport"}, {"id": "roi", "type": "float", "label": "ROI", "unit": "roi", "format": "+.1%"}, {"id": "lo95", "type": "float", "unit": "roi"}, {"id": "hi95", "type": "float", "unit": "roi"}, {"id": "n_bets", "type": "int", "label": "Bets"}], "rows": [{"sport": "NFL", "roi": 0.041, "lo95": -0.012, "hi95": 0.094, "n_bets": 410}, {"sport": "NBA", "roi": -0.022, "lo95": -0.081, "hi95": 0.037, "n_bets": 330}]}, "x": {"field": "roi", "type": "quantitative", "title": "ROI"}, "y": {"field": "sport", "type": "nominal"}, "series": [{"field": "roi", "role": "market", "label": "ROI"}], "intervals": [{"low": "lo95", "high": "hi95", "level": 0.95}], "rules": [{"x": 0, "role": "break_even", "label": "Break-even"}], "count": "n_bets", "summary": "NFL +4.1% over 410 bets and NBA -2.2% over 330; neither can be told from luck yet.", "provenance": {"tool": "get_my_results", "engine": "results 2", "as_of": "2026-10-04T17:30:00Z", "sources": ["bet log"]}}",
"format": "python"
}
{
"object": "chart_export",
"format": "python",
"media_type": "text/x-python",
"filename": "roi-by-sport.py",
"text": "# ROI by sport\n# Kelly chart (bars, KCS 1) from get_my_results, engine results 2, as of 2026-10-04T17:30:00Z\n# Sources: bet log\n# NFL +4.1% over 410 bets and NBA -2.2% over 330; neither can be told from luck yet.\nimport json\n\nimport altair as alt\nimport pandas as pd\n\nrows = json.loads('[{\"sport\": \"NFL\", \"roi\": 0.041, \"lo95\": -0.012, \"hi95\": 0.094, \"n_bets\": 410}, {\"sport\": \"NBA\", \"roi\": -0.022, \"lo95\": -0.081, \"hi95\": 0.037, \"n_bets\": 330}]')\ndf = pd.DataFrame.from_records(rows)\nspec = json.loads('{\\n \"$schema\": \"https://vega.github.io/schema/vega-lite/v5.json\",\\n \"title\": {\\n \"text\": \"ROI by sport\"\\n },\\n \"description\": \"NFL +4.1% over 410 bets and NBA -2.2% over 330; neither can be told from luck yet.\",\\n \"usermeta\": {\\n \"kelly\": {\\n \"kcs\": 1,\\n \"kind\": \"bars\",\\n \"provenance\": {\\n \"tool\": \"get_my_results\",\\n \"engine\": \"results 2\",\\n \"as_of\": \"2026-10-04T17:30:00Z\",\\n \"sources\": [\\n \"bet log\"\\n ]\\n },\\n \"rights\": {\\n \"export_rows\": true,\\n \"export_columns_blocked\": []\\n }\\n }\\n },\\n \"layer\": [\\n {\\n \"mark\": {\\n \"type\": \"rule\",\\n \"color\": \"#2f62b3\",\\n \"strokeWidth\": 2\\n },\\n \"encoding\": {\\n \"y\": {\\n \"field\": \"sport\",\\n \"type\": \"nominal\",\\n \"sort\": null\\n },\\n \"x\": {\\n \"field\": \"lo95\",\\n \"type\": \"quantitative\",\\n \"title\": \"ROI\",\\n \"scale\": {\\n \"zero\": false\\n }\\n },\\n \"x2\": {\\n \"field\": \"hi95\"\\n }\\n }\\n },\\n {\\n \"mark\": {\\n \"type\": \"rule\",\\n \"color\": \"#14171c\",\\n \"strokeDash\": [\\n 4,\\n 3\\n ],\\n \"strokeWidth\": 1\\n },\\n \"encoding\": {\\n \"x\": {\\n \"datum\": 0.0,\\n \"type\": \"quantitative\"\\n }\\n }\\n },\\n {\\n \"transform\": [\\n {\\n \"filter\": {\\n \"field\": \"roi\",\\n \"valid\": true\\n }\\n },\\n {\\n \"fold\": [\\n \"roi\"\\n ],\\n \"as\": [\\n \"_series\",\\n \"_value\"\\n ]\\n }\\n ],\\n \"mark\": {\\n \"type\": \"bar\"\\n },\\n \"encoding\": {\\n \"x\": {\\n \"field\": \"_value\",\\n \"type\": \"quantitative\",\\n \"title\": \"ROI\",\\n \"scale\": {\\n \"zero\": false\\n }\\n },\\n \"y\": {\\n \"field\": \"sport\",\\n \"type\": \"nominal\",\\n \"sort\": null\\n },\\n \"color\": {\\n \"field\": \"_series\",\\n \"type\": \"nominal\",\\n \"scale\": {\\n \"domain\": [\\n \"roi\"\\n ],\\n \"range\": [\\n \"#2f62b3\"\\n ]\\n },\\n \"legend\": {\\n \"title\": null,\\n \"labelExpr\": \"{\\\\\"roi\\\\\": \\\\\"ROI\\\\\"}[datum.label]\"\\n }\\n }\\n }\\n }\\n ]\\n}')\nchart = alt.Chart.from_dict({**spec, \"data\": {\"values\": rows}})\nchart\n",
"kind": "bars",
"title": "ROI by sport",
"rows": 2,
"columns_removed": [],
"provenance": {
"tool": "get_my_results",
"engine": "results 2",
"as_of": "2026-10-04T17:30:00Z",
"sources": [
"bet log"
]
},
"summary": "ROI by sport as Python (pandas + Altair): 2 rows.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_9984d73cccd12e8e049a4a5e",
"live_topics": [],
"rg_line": null
}POST /v1/rules:compare_source
public, no key needed · 1 credit, not charged during the beta · MCP tool compare_rule_to_source · status live
A betting rule imported from another tool, tested as test_rule tests it, beside the record the source reported (a pasted results table, a list of games or a stated record), game by game: a status (reproduced, close or differs), the records side by side, the source's own games regraded by Kelly, the record since the import date, each reason a game differs with its count, and every differing game with both lines (backtester T9b). The source's figures are as reported; never a pick or a ranking.
Body, application/json (required):
formulastringThe rule: a formula true or false on each team-game (or each runner in a race, for horse_racing, greyhounds and harness), starting with =.
min length 2 · max length 2000
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...; racing: horse_racing, greyhounds or harness (seasons are calendar years).
pattern ^[a-z0-9_]{2,24}$
seasonsarray of stringSeasons to test, oldest first (1 to 25), e.g. ["2021", "2022"]; or use from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
sidestringrule (default) bets the team the rule matched; fade bets the other side.
One of: rule, fade (open: new values may be added).
bet_momentstringclose (default), edition (football: the weekly sheet's Monday 10 AM ET line) or day_before (other sports: 24 hours before the start).
One of: close, edition, day_before (open: new values may be added).
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
stakestringrisk (default): 1 unit risked a bet; to_win: each bet sized to win 1 unit.
One of: risk, to_win (open: new values may be added).
pricestringconsensus (default): Kelly's price at the bet moment; flat_110: every bet at -110.
One of: consensus, flat_110 (open: new values may be added).
warm_upstringas_written (default) grades every bet; require_full grades only bets whose windows had all the earlier games they read (ROLLING_AVG(margin, 5) needs 5).
One of: as_written, require_full (open: new values may be added).
marketstringspread (default), moneyline, over or under.
One of: spread, moneyline, over, under (open: new values may be added).
source_textstringThe source's results table as pasted (CSV, tab-separated, or JSON: a list of games or an import draft's source_record), with headings such as Date, Team, Opponent, Site, Line, Result, Units; or its record alone (42-30-3, +9.4 units, 11.2% ROI).
min length 1 · max length 5000000
sincestringThe import date (YYYY-MM-DD): the record from it on is the rule's clean test.
format date
200 The rule proof facts.
objectconstrequiredAlways rule_proof.
Always "rule_proof".
statusstringrequiredreproduced (nothing differs), close (every difference has a known cause) or differs (at least one has none).
One of: reproduced, close, differs (open: new values may be added).
status_reasonstringrequiredThe status in words.
recordsobjectrequiredThe records side by side.
sourceobjectrequiredThe source's record, as reported.
winsintegernullableWins.
lossesintegernullableLosses.
pushesintegernullablePushes.
recordstringnullableWins-losses(-pushes).
unitsnumbernullableUnits won or lost, 1 unit risked a bet unless the test said to_win.
roinumbernullableUnits over bets.
as_reportedbooleanAlways true: the source's own figures, not Kelly's.
kellyobjectrequiredKelly's test of the rule over the seasons asked.
winsintegernullableWins.
lossesintegernullableLosses.
pushesintegernullablePushes.
recordstringnullableWins-losses(-pushes).
unitsnumbernullableUnits won or lost, 1 unit risked a bet unless the test said to_win.
roinumbernullableUnits over bets.
kelly_in_source_windowobjectnullableKelly's test between the source's first and last games.
winsintegernullableWins.
lossesintegernullableLosses.
pushesintegernullablePushes.
recordstringnullableWins-losses(-pushes).
unitsnumbernullableUnits won or lost, 1 unit risked a bet unless the test said to_win.
roinumbernullableUnits over bets.
fromstringnullableFirst date.
tostringnullableLast date.
source_games_regradedobjectnullableThe source's own games graded by Kelly at the bet moment, whatever the rule says: a check of the source's record itself.
winsintegernullableWins.
lossesintegernullableLosses.
pushesintegernullablePushes.
recordstringnullableWins-losses(-pushes).
unitsnumbernullableUnits won or lost, 1 unit risked a bet unless the test said to_win.
roinumbernullableUnits over bets.
gamesintegerGames matched to Kelly's.
no_lineintegerOf them, games with no Kelly line.
momentstringThe bet moment graded.
since_importobjectnullableKelly's test from the import date on: the rule's clean record (I1).
winsintegernullableWins.
lossesintegernullableLosses.
pushesintegernullablePushes.
recordstringnullableWins-losses(-pushes).
unitsnumbernullableUnits won or lost, 1 unit risked a bet unless the test said to_win.
roinumbernullableUnits over bets.
sincestringThe import date.
gamesobjectnullableCounts: source, kelly, both, same, source_only, kelly_only, result_differs, outside.
reasonsarray of objectrequiredEach reason found, in a fixed order (never sorted by count).
codestringrequiredThe reason.
One of: other_line, push_rule, no_line, missing_stat, no_game, both_sides, short_window, void, outside_seasons, outside_source_window, condition, not_in_source, other_side, result_differs, team_unread (open: new values may be added).
labelstringrequiredThe reason in words.
explainedbooleanrequiredA known cause (another line, a push rule, a data gap, a window); false when the import itself may be wrong.
countintegerrequiredGames.
rowsarray of objectrequiredGames that differ, by date (first 200).
datestringrequirednullableThe game's date (YYYY-MM-DD).
instringrequiredsource (only the source bet it), kelly (only Kelly's test did) or both (results differ).
One of: source, kelly, both (open: new values may be added).
reasonstringrequiredThe reason's code (reasons[].code).
detailstringrequiredWhy, in words, with both lines when a line is the cause.
sourceobjectnullableThe source's game as it listed it.
teamstringnullableThe team, the source's words.
opponentstringnullableThe opponent, the source's words.
homebooleannullableHome (true), away (false), neutral or not said (null).
linenumbernullableThe source's line.
resultstringnullablewin, loss or push, as reported.
unitsnumbernullableThe source's units for the game, as reported.
textstringnullableThe row as the source wrote it.
kellyobjectnullableKelly's bet or game.
rows_countintegerrequiredGames that differ, all of them.
deltasobjectnullableA record-only comparison: Kelly minus the source, for wins, losses, pushes and units.
testobjectKelly's test of the rule: overall and the verdict's headline (test_rule has the full report).
source_readobjectnullableHow the source's list was read: rows, rows left out and why, the headings used, the stated record against the games' own count.
ruleobjectThe rule and options tested.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
noticestringrequiredThis compares an import with its source's reported record; private, never a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/rules:compare_source
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=home",
"competition": "nba",
"seasons": [
"2025-26"
],
"source_text": "1-0, +0.9 units"
}
{
"object": "rule_proof",
"rule": {
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
]
},
"records": {
"source": {
"wins": 1,
"losses": 0,
"pushes": 0,
"units": 0.9,
"roi": 0.9,
"as_reported": true
},
"kelly": {
"wins": 1,
"losses": 0,
"pushes": 0,
"record": "1-0",
"units": 0.91,
"roi": 0.91
}
},
"test": {
"overall": {
"wins": 1,
"losses": 0,
"pushes": 0,
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"win_rate": 1,
"break_even": 0.5235602094240838,
"over_break_even": 0.47643979057591623,
"significance": {
"p_value": 0.5235602094240838,
"no_skill_match_per_100": 52.35602094240838,
"coin_match_per_100": 50,
"z": 0.9539392014169455,
"significant": false,
"alpha": 0.05,
"multiple_comparisons": null,
"anytime_valid": {
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"kelly_fraction": 0.5,
"crossed": false,
"roi_cs95": [
-0.999045,
0.9090450000000001
]
},
"method": "exact_binomial",
"t_stat": null,
"p_value_exact": 0.5235602094240838,
"p_value_clustered": 0.17005720935900537,
"p_value_vs_vig": null,
"null": "break_even",
"sims": null,
"seed": null,
"p_adjusted": 0.5235602094240838,
"adjust_method": "none",
"tests": 1,
"tests_effective": 1,
"reality_check_p": null,
"spa_p": null,
"clears_t3": false,
"haircut": null,
"haircut_roi": null
},
"sample_needed": null,
"few_decisions": true,
"small_sample": true,
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F36.2@1"
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"roi_ci": {
"method": "score",
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"bca": null,
"block_bootstrap": null,
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"iid_score": [
-0.6054908095394765,
0.9099999999999999
]
},
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"deflated": {
"sharpe_per_bet": null,
"skew": null,
"kurtosis": null,
"psr": null,
"dsr": null,
"sharpe_expected_max": null,
"roi_bar": null,
"trials_effective": 1,
"sample_needed": null,
"sample_needed_single": null,
"sample_needed_power80": null,
"sample_needed_plan": null
}
},
"verdict": "Too few bets to judge"
},
"as_of": "2026-11-06T00:00:00Z",
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"notice": "This compares your import with the record its source reported. It is private to you; it isn't a pick and doesn't rank anyone.",
"status": "reproduced",
"deltas": {
"wins": 0,
"losses": 0,
"pushes": 0,
"units": 0.01
},
"games": null,
"reasons": [],
"rows": [],
"rows_count": 0,
"status_reason": "Kelly's record matches the source's.",
"source_read": {
"games": 0,
"skipped_count": 0,
"skipped": [],
"columns": {},
"stated": {
"wins": 1,
"losses": 0,
"pushes": 0,
"units": 0.9
},
"from_games": null,
"list_and_record_agree": null
},
"summary": "Reproduced: Kelly's test 1-0, +0.9 units; the source reported 1-0, +0.9 units.",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_7679266df624114290c90502",
"live_topics": [],
"rg_line": null
}POST /v1/rules:test_portfolio
public, no key needed · 1 credit, not charged during the beta · MCP tool test_portfolio · status live
Up to ten betting rules tested on the same seasons as one book (backtester T4, F38): for every two rules the bets they share, the games where they take opposite sides, and the correlation of their daily units; one combined ledger with shared bets staked once (or stacked) and opposite sides netted (or both bet), weighted equally, by inverse volatility or for equal risk; the combined record with its own verdict and p-value counting every member's rules tried, ruin, staking plans and risk-adjusted stats. Never a search over combinations.
Body, application/json (required):
formulasarray of stringThe rules, 1 to 10: each a formula true or false on a team-game, starting with =.
min items 1 · max items 10
marketsarray of stringEach rule's market (spread, moneyline, over, under), in the order of formulas; one value applies to all (default spread).
min items 1 · max items 10
sidesarray of stringEach rule's side (rule or fade), in order; one value applies to all (default rule).
min items 1 · max items 10
bet_momentsarray of stringEach rule's bet moment (close or edition), in order; one value applies to all.
min items 1 · max items 10
labelsarray of stringA short name for each rule, in order (default r1, r2, ...).
min items 1 · max items 10
rules_triedarray of integerRules tried for each rule, in order (default 1 each); the combined p-value counts them all.
min items 1 · max items 10
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
seasonsarray of stringSeasons to test, oldest first (1 to 25); or use from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
stakestringrisk (default): 1 unit risked a bet; to_win: each bet sized to win 1 unit.
One of: risk, to_win (open: new values may be added).
pricestringconsensus (default) or flat_110.
One of: consensus, flat_110 (open: new values may be added).
weightingstringequal (default), inverse_vol or equal_risk.
One of: equal, inverse_vol, equal_risk (open: new values may be added).
duplicatesstringdedupe (default): a bet two rules share is staked once; stack: once a rule.
One of: dedupe, stack (open: new values may be added).
opposingstringnet (default): opposite sides of a game cancel; both: both are bet.
One of: net, both (open: new values may be added).
detailstringfull (default) or summary (no bets list or bankroll curve).
One of: full, summary (open: new values may be added).
bankroll_unitsnumberBankroll in units for the ruin chance and the staking replays (default 100).
max 1000000
horizon_betsintegerBets ahead the ruin chance looks (default 500).
min 1 · max 1000
pathsintegerSimulated futures for the ruin chance (default 5000).
min 1000 · max 10000
staking_plansarray of stringStaking plans to replay: flat, to_win, percent, kelly (default all four).
min items 1 · max items 4
percentnumberShare of the bankroll the percent plan bets (default 0.02).
max 0.5
kelly_fractionnumberMultiple of the Kelly criterion the kelly plan bets; omitted, the honest multiple from the record's sample size.
max 1
kelly_capnumberMost of the bankroll the kelly plan puts on one bet (default 0.05).
max 1
200 The rule portfolio facts.
objectconstrequiredAlways rule_portfolio.
Always "rule_portfolio".
verdictobjectrequiredVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
combinedobjectrequiredThe combined ledger: record, units, ROI, the record test with tests = every member's rules tried, max drawdown, and how shared and opposite bets were handled.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits at the weights.
roinumbernullableUnits over units risked.
testsintegerEvery member's rules tried.
max_drawdown_unitsnumberWorst peak-to-trough drop, units.
netted_gamesintegerGames where opposite sides met.
membersarray of objectrequiredEach rule, in the order given.
rulestringIts label.
formulastringIts formula.
marketstringIts market.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits.
roinumbernullableROI.
levelstringIts own verdict level.
testsintegerRules tried for it.
pairsarray of objectrequiredEvery two rules, in the order given.
astringThe first rule's label.
bstringThe second rule's label.
jaccardnumbernullableShared bets over the bets either takes.
opposing_gamesintegerGames where they take opposite sides of the same market.
daily_corrnumbernullableCorrelation of their daily units, days either bets (zero-filled).
weightsarray of objectWeights, mean 1.
risk_contributionsarray of objectnullableShares of risk, summing to 1.
diversification_rationumbernullableΣ w·σ over the book's σ, daily units.
effective_rulesnumbernullable(Σλ)²/Σλ² over the correlation matrix's eigenvalues.
weightingstringequal, inverse_vol or equal_risk.
duplicatesstringdedupe or stack.
opposingstringnet or both.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
seasonsarray of objectSeasons in calendar order.
betsarray of objectThe combined book in date order (left out at detail summary).
ruleobjectThe options the rules were tested with.
noticestringrequiredThis tests rules on past games; it isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/rules:test_portfolio
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formulas": [
"=home",
"=home"
],
"sides": [
"rule",
"fade"
],
"competition": "nba",
"seasons": [
"2025-26"
],
"detail": "summary"
}
{
"object": "rule_portfolio",
"verdict": {
"version": 2,
"headline": "No bets",
"level": "too_few",
"passed": 0,
"applicable": 0,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment"
},
"checks": []
},
"combined": {
"record": "0-0",
"wins": 0,
"losses": 0,
"pushes": 0,
"bets": 0,
"decisions": 0,
"units": 0,
"roi": null,
"win_rate": null,
"break_even": null,
"significance": null,
"max_drawdown_units": 0,
"longest_losing_run": 0,
"tests": 5,
"deduped": 0,
"stacked": 0,
"netted_games": 1
},
"members": [
{
"rule": "r1",
"formula": "=home",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"record": "1-0",
"decisions": 1,
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"level": "too_few",
"tests": 2
},
{
"rule": "r2",
"formula": "=home",
"market": "spread",
"side": "fade",
"bet_moment": "close",
"record": "0-1",
"decisions": 1,
"units": -1,
"roi": -1,
"level": "too_few",
"tests": 3
}
],
"pairs": [
{
"a": "r1",
"b": "r2",
"shared_bets": 0,
"jaccard": 0,
"opposing_games": 1,
"daily_corr": null,
"daily_corr_shared_days": null,
"shared_days": 1
}
],
"weighting": "equal",
"duplicates": "dedupe",
"opposing": "net",
"weights": [
{
"rule": "r1",
"weight": 1
},
{
"rule": "r2",
"weight": 1
}
],
"risk_contributions": null,
"diversification_ratio": null,
"effective_rules": null,
"seasons": [],
"rule": {
"competition": "nba",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus"
},
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rules together on past games. It isn't a pick and doesn't rank selections.",
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 6437125516759677,
"lose_all_ever": null,
"lose_all_ever_allowing_error": null,
"ruin_ever_if_rate_exact": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": null,
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": null,
"push_rate": 0,
"full_fraction": null,
"shrink": null,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"sharpe_per_bet": null,
"profit_factor": null,
"warnings": [
"no_bets"
]
},
"summary": "The (2) rules matched no game Kelly holds a line for.",
"assumptions": [
"Bets shared by rules are staked once; opposite sides of a game are netted.",
"Weights: equal.",
"The p-value counts every member's rules tried (5)."
],
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_a914a84cdf6b903fd164f76d",
"live_topics": [],
"rg_line": null
}GET /v1/rules/templates
public, no key needed · 0 credits, not charged during the beta · MCP tool list_rule_templates · status live
Kelly's system templates: well-known public betting angles as rule formulas, each with its competition, market, side, what it does, the idea behind it and the inputs a person may change. In the library's order, never ranked; a template's record comes from POST /v1/rules:test.
competitionstring · queryThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
marketstring · queryOnly templates on this market.
One of: spread, moneyline, over, under, teaser (open: new values may be added).
tagstring · queryOnly templates with this tag (rest, underdog, playoffs, teaser...).
min length 1 · max length 40
200 The rule templates facts.
objectconstrequiredAlways rule_templates.
Always "rule_templates".
templatesarray of objectrequiredThe templates that match the filters.
idstringStable ID, e.g. nfl_home_underdog_after_bye.
namestringWhat it does, in a few words.
competitionstringnfl, ncaaf, nba, ncaab, nhl...
marketstringspread, moneyline, over, under or teaser.
sidestringrule bets the matched team; fade bets the other side.
season_typestringregular, postseason or both.
formula_templatestringThe formula with each input as {name}.
formulastringThe formula with the values in values.
inputsarray of objectThe inputs, in order.
namestringIts name, as in formula_template's {name}.
labelstringWhat it is.
defaultnumberThe template's value.
minnumberSmallest allowed.
maxnumberLargest allowed.
stepnumberValues move in these steps from min.
valuesobjectEach input's value in formula.
what_it_doesstringThe bet, in one sentence.
ideastringThe idea behind it; a public name or author is cited by name.
notesstringnullableHow the rule reads the data, when it matters.
tagsarray of stringTags.
sourceobjectWhat an import of it carries: kind template, tool kelly_template:<id>.
totalintegerrequiredHow many.
competitionsarray of stringCompetitions with templates.
tagsarray of stringEvery tag in the library.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled
GET https://api.askkelly.ai/v1/rules/templates?competition=nba
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_templates",
"templates": [
{
"id": "nba_playoff_zig_zag",
"name": "Playoff teams after a loss (zig-zag)",
"competition": "nba",
"market": "spread",
"side": "rule",
"season_type": "postseason",
"teaser": null,
"formula_template": "=LAG(result) = \"L\"",
"formula": "=LAG(result) = \"L\"",
"conditions": [
{
"label": "Lost the last playoff game",
"formula": "LAG(result) = \"L\""
}
],
"inputs": [],
"values": {},
"what_it_does": "Bets a team against the spread in the playoff game after it lost a playoff game.",
"idea": "Known in public betting writing as the zig-zag theory, an idea that a team adjusts after a loss within a series.",
"notes": "Only playoff games are read, so a team's first playoff game never matches.",
"tags": [
"playoffs",
"previous game"
],
"source": {
"kind": "template",
"tool": "kelly_template:nba_playoff_zig_zag"
}
},
{
"id": "nba_rested_against_back_to_back",
"name": "Rested teams against a team on a back-to-back",
"competition": "nba",
"market": "spread",
"side": "rule",
"season_type": "regular",
"teaser": null,
"formula_template": "=AND(NOT(back_to_back), opponent.back_to_back)",
"formula": "=AND(NOT(back_to_back), opponent.back_to_back)",
"conditions": [
{
"label": "Did not play the night before",
"formula": "NOT(back_to_back)"
},
{
"label": "The opponent played the night before",
"formula": "opponent.back_to_back"
}
],
"inputs": [],
"values": {},
"what_it_does": "Bets a team against the spread when its opponent played the night before and it did not.",
"idea": "The second night of a back-to-back is the most discussed fatigue spot in the NBA.",
"notes": null,
"tags": [
"rest",
"back-to-back"
],
"source": {
"kind": "template",
"tool": "kelly_template:nba_rested_against_back_to_back"
}
},
{
"id": "nba_fourth_game_in_seven_days",
"name": "Against teams on their fourth game in seven days",
"competition": "nba",
"market": "spread",
"side": "fade",
"season_type": "regular",
"teaser": null,
"formula_template": "=games_last_7_days >= {min_games}",
"formula": "=games_last_7_days >= 4",
"conditions": [
{
"label": "4 or more games in seven days, this one included",
"formula": "games_last_7_days >= 4"
}
],
"inputs": [
{
"name": "min_games",
"label": "Games in the last seven days counting this one",
"default": 4,
"min": 3,
"max": 5,
"step": 1
}
],
"values": {
"min_games": 4
},
"what_it_does": "Bets against a team, against the spread, when this is its fourth game in seven days.",
"idea": "A schedule-density angle; the count includes the game itself.",
"notes": null,
"tags": [
"rest",
"schedule",
"fade"
],
"source": {
"kind": "template",
"tool": "kelly_template:nba_fourth_game_in_seven_days"
}
},
{
"id": "nba_home_underdogs",
"name": "NBA home underdogs",
"competition": "nba",
"market": "spread",
"side": "rule",
"season_type": "regular",
"teaser": null,
"formula_template": "=AND(home, spread_close >= {min_points})",
"formula": "=AND(home, spread_close >= 0.5)",
"conditions": [
{
"label": "At home",
"formula": "home"
},
{
"label": "Underdog by 0.5 or more",
"formula": "spread_close >= 0.5"
}
],
"inputs": [
{
"name": "min_points",
"label": "Smallest underdog line",
"default": 0.5,
"min": 0.5,
"max": 20,
"step": 0.5
}
],
"values": {
"min_points": 0.5
},
"what_it_does": "Bets NBA home teams getting points against the spread.",
"idea": "A long-discussed public angle; this tests it on Kelly's closing lines.",
"notes": null,
"tags": [
"underdog",
"home"
],
"source": {
"kind": "template",
"tool": "kelly_template:nba_home_underdogs"
}
}
],
"total": 4,
"competitions": [
"nba",
"ncaab",
"ncaaf",
"nfl",
"nhl"
],
"tags": [
"back-to-back",
"blowout",
"fade",
"favourite",
"home",
"key numbers",
"playoffs",
"previous game",
"rest",
"road",
"schedule",
"streak",
"teaser",
"totals",
"under",
"underdog"
],
"summary": "4 system templates for NBA, in the library's order (not ranked). Each is a starting point to test or import; its record comes from test_rule.",
"assumptions": [
"Templates are listed in the library's order, never by any record."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_29b1e40e81ce21a6c4979c70",
"live_topics": [],
"rg_line": null
}POST /v1/rules/templates:fill
public, no key needed · 0 credits, not charged during the beta · MCP tool fill_rule_template · status live
One system template filled in with the inputs given (its defaults otherwise), optionally ANDed with the caller's own conditions, with the exact body for POST /v1/rules:test and the import arguments (source kind template).
Body, application/json (required):
template_idstringThe template's ID from list_rule_templates, e.g. nfl_home_underdog_after_bye.
pattern ^[a-z][a-z0-9_]{2,63}$
inputsarray of stringValues for the template's inputs, each name=number, e.g. ["min_points=3"]; each within its min and max, in its steps; the rest keep the template's defaults.
max items 10
and_formulastringThe person's own conditions ANDed with the template's, e.g. =NOT(home).
min length 1 · max length 1000
seasonsarray of stringSeasons to put in test_rule_arguments, oldest first.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
200 The rule template facts.
objectconstrequiredAlways rule_template.
Always "rule_template".
templateobjectrequiredOne system template: a well-known public betting angle as a rule, named by what it does (I3).
idstringStable ID, e.g. nfl_home_underdog_after_bye.
namestringWhat it does, in a few words.
competitionstringnfl, ncaaf, nba, ncaab, nhl...
marketstringspread, moneyline, over, under or teaser.
sidestringrule bets the matched team; fade bets the other side.
season_typestringregular, postseason or both.
formula_templatestringThe formula with each input as {name}.
formulastringThe formula with the values in values.
inputsarray of objectThe inputs, in order.
namestringIts name, as in formula_template's {name}.
labelstringWhat it is.
defaultnumberThe template's value.
minnumberSmallest allowed.
maxnumberLargest allowed.
stepnumberValues move in these steps from min.
valuesobjectEach input's value in formula.
what_it_doesstringThe bet, in one sentence.
ideastringThe idea behind it; a public name or author is cited by name.
notesstringnullableHow the rule reads the data, when it matters.
tagsarray of stringTags.
sourceobjectWhat an import of it carries: kind template, tool kelly_template:<id>.
test_rule_argumentsobjectrequiredThe body for POST /v1/rules:test (Kelly MCP's test_rule) as it is.
import_argumentsobjectWhat to send to the import (source kind template).
next_callobjectThe call that gives the template's record: test_rule.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · not_found
POST https://api.askkelly.ai/v1/rules/templates:fill
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"template_id": "nfl_home_underdog_after_bye",
"inputs": [
"min_points=3"
],
"from_season": "2021",
"to_season": "2024"
}
{
"object": "rule_template",
"template": {
"id": "nfl_home_underdog_after_bye",
"name": "Home underdogs after a bye week",
"competition": "nfl",
"market": "spread",
"side": "rule",
"season_type": "regular",
"teaser": null,
"formula_template": "=AND(home, spread_close >= {min_points}, rest_days >= {min_rest_days})",
"formula": "=AND(home, spread_close >= 3, rest_days >= 13)",
"conditions": [
{
"label": "At home",
"formula": "home"
},
{
"label": "Underdog by 3 or more",
"formula": "spread_close >= 3"
},
{
"label": "13 or more days since the last game",
"formula": "rest_days >= 13"
}
],
"inputs": [
{
"name": "min_points",
"label": "Smallest underdog line",
"default": 0.5,
"min": 0.5,
"max": 14,
"step": 0.5
},
{
"name": "min_rest_days",
"label": "Days since the team's last game",
"default": 13,
"min": 10,
"max": 21,
"step": 1
}
],
"values": {
"min_points": 3,
"min_rest_days": 13
},
"what_it_does": "Bets the home team against the spread when it is the underdog and had a week off before the game.",
"idea": "An extra week to prepare, at home, with the market pricing the team as the underdog. A bye week is 13 or more days between games.",
"notes": null,
"tags": [
"rest",
"underdog",
"home"
],
"source": {
"kind": "template",
"tool": "kelly_template:nfl_home_underdog_after_bye"
}
},
"test_rule_arguments": {
"formula": "=AND(home, spread_close >= 3, rest_days >= 13)",
"competition": "nfl",
"market": "spread",
"side": "rule",
"season_type": "regular",
"from_season": "2021",
"to_season": "2024"
},
"import_arguments": {
"source": {
"kind": "template",
"tool": "kelly_template:nfl_home_underdog_after_bye"
},
"text": "=AND(home, spread_close >= 3, rest_days >= 13)",
"competition": "nfl",
"market": "spread",
"side": "rule"
},
"next_call": {
"tool": "test_rule",
"arguments": {
"formula": "=AND(home, spread_close >= 3, rest_days >= 13)",
"competition": "nfl",
"market": "spread",
"side": "rule",
"season_type": "regular",
"from_season": "2021",
"to_season": "2024"
}
},
"summary": "Home underdogs after a bye week: =AND(home, spread_close >= 3, rest_days >= 13). Test it with test_rule to see its record.",
"assumptions": [
"Smallest underdog line: 3",
"Days since the team's last game: 13 (the template's default)"
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_ec3db33575ad6ef23eeedbd6",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules/templates:fill
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"template_id": "nfl_home_dogs"
}
{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No template is called nfl_home_dogs.",
"param": "template_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_8333b63a22d4a84d8c2ecf17",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules:interview
public, no key needed · 1 credit, not charged during the beta · MCP tool build_rule · status live
The interview wizard: the answers so far (league, seasons, market, side, home or away, favourite or underdog, rest, back-to-backs, last game, total, season type) as labelled conditions and one rule formula, the next question with its choices, and the team-games the rule matches so far, read point in time as the rule tester reads them. Never a record or a pick: the record comes from POST /v1/rules:test.
Body, application/json (required):
competitionstringnfl, ncaaf, nba, ncaab, wnba, nhl or mlb.
One of: nfl, ncaaf, nba, ncaab, wnba, nhl, mlb (open: new values may be added).
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
marketstringspread, moneyline, over or under.
One of: spread, moneyline, over, under (open: new values may be added).
sidestringon (bet the team the conditions describe) or against (bet the other side).
One of: on, against (open: new values may be added).
venuestringhome, away or any.
One of: home, away, any (open: new values may be added).
rolestringfavourite, underdog or any (spread sports only).
One of: favourite, underdog, any (open: new values may be added).
line_minnumberWith role: the smallest line, in points (laying or getting).
min 0 · max 100
line_maxnumberWith role: the largest line, in points.
min 0 · max 100
rest_min_daysnumberFewest days since the team's last game.
min 0 · max 100
rest_edge_minnumberFewest extra days of rest over the opponent.
min 0 · max 100
back_to_backstringWho played the night before (basketball, hockey, baseball): team, opponent, team_not_opponent, opponent_not_team, neither or any.
One of: team, opponent, team_not_opponent, opponent_not_team, neither, any (open: new values may be added).
last_resultstringThe team's last game: won, lost or any.
One of: won, lost, any (open: new values may be added).
last_margin_minnumberWith last_result: by at least this many points.
min 0 · max 100
last_atsstringLast game against the spread: covered, failed or any (spread sports only).
One of: covered, failed, any (open: new values may be added).
total_minnumberSmallest closing total.
min 0 · max 400
total_maxnumberLargest closing total.
min 0 · max 400
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The rule interview facts.
objectconstrequiredAlways rule_interview.
Always "rule_interview".
questionsarray of objectrequiredEvery question for this league, in order.
keystringIts argument name.
askstringThe question.
choicesarray of stringIts choices; empty for a number.
numbersarray of stringThe number arguments it takes.
answeredbooleanAnswered in this request.
next_questionobjectnullableThe first unanswered question.
skipped_questionsarray of stringQuestions that don't apply to this league.
conditionsarray of objectrequiredEach answer's condition, in question order.
questionstringThe question it answers.
labelstringThe condition in words.
formulastringThe condition as a formula, without =.
formulastringrequirednullableThe conditions ANDed; null before any.
test_rule_argumentsobjectnullableThe body for POST /v1/rules:test.
import_argumentsobjectnullableWhat to send to the import (source kind chat, tool kelly_interview).
completebooleanEvery question is answered.
matchedintegerrequirednullableTeam-games the rule matches over the seasons (each game has two); null until league and seasons are set.
games_in_scopeintegernullableTeam-games read.
by_seasonarray of objectCounts by season.
problemsarray of objectThe rule tester's own problems.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/rules:interview
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"competition": "nba",
"from_season": "2025-26",
"to_season": "2025-26",
"venue": "home"
}
{
"object": "rule_interview",
"questions": [
{
"key": "competition",
"ask": "Which league is the system for?",
"choices": [
"nfl",
"ncaaf",
"nba",
"ncaab",
"wnba",
"nhl",
"mlb"
],
"numbers": [],
"answered": true
},
{
"key": "seasons",
"ask": "Which seasons should Kelly test it on?",
"choices": [],
"numbers": [
"from_season",
"to_season"
],
"answered": true
},
{
"key": "market",
"ask": "What do you bet: the spread, the moneyline, the over or the under?",
"choices": [
"spread",
"moneyline",
"over",
"under"
],
"numbers": [],
"answered": false
},
{
"key": "side",
"ask": "Do you bet on the team your conditions describe, or against it?",
"choices": [
"on",
"against"
],
"numbers": [],
"answered": false
},
{
"key": "venue",
"ask": "Home or away?",
"choices": [
"home",
"away",
"any"
],
"numbers": [],
"answered": true
},
{
"key": "role",
"ask": "Favourite or underdog?",
"choices": [
"favourite",
"underdog",
"any"
],
"numbers": [
"line_min",
"line_max"
],
"answered": false
},
{
"key": "rest",
"ask": "How rested is the team? Fewest days since its last game, or most extra days over the opponent.",
"choices": [],
"numbers": [
"rest_min_days",
"rest_edge_min"
],
"answered": false
},
{
"key": "back_to_back",
"ask": "Back-to-backs: who played the night before?",
"choices": [
"team",
"opponent",
"team_not_opponent",
"opponent_not_team",
"neither",
"any"
],
"numbers": [],
"answered": false
},
{
"key": "last_result",
"ask": "How did the team's last game go?",
"choices": [
"won",
"lost",
"any"
],
"numbers": [
"last_margin_min"
],
"answered": false
},
{
"key": "last_ats",
"ask": "Did the team cover the spread last game?",
"choices": [
"covered",
"failed",
"any"
],
"numbers": [],
"answered": false
},
{
"key": "total",
"ask": "Any limits on the game's closing total?",
"choices": [],
"numbers": [
"total_min",
"total_max"
],
"answered": false
},
{
"key": "season_type",
"ask": "Regular season, playoffs or both?",
"choices": [
"regular",
"postseason",
"both"
],
"numbers": [],
"answered": false
}
],
"next_question": {
"key": "market",
"ask": "What do you bet: the spread, the moneyline, the over or the under?",
"choices": [
"spread",
"moneyline",
"over",
"under"
],
"numbers": []
},
"skipped_questions": [],
"conditions": [
{
"question": "venue",
"label": "At home",
"formula": "home"
}
],
"formula": "=home",
"test_rule_arguments": {
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"season_type": "regular",
"from_season": "2025-26",
"to_season": "2025-26"
},
"import_arguments": {
"source": {
"kind": "chat",
"tool": "kelly_interview"
},
"text": "=home",
"competition": "nba",
"market": "spread",
"side": "rule"
},
"complete": false,
"matched": 3,
"games_in_scope": 6,
"by_season": [
{
"season": "2025-26",
"matched": 3,
"team_games": 6
}
],
"problems": [],
"summary": "1 condition so far: =home matches 3 of 6 team-games.",
"assumptions": [
"Each answer adds one condition; an answer of any adds none.",
"matched counts team-games (each game has two), before lines are checked; the record comes from test_rule."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_10cfe25bbb94869626132398",
"live_topics": [],
"rg_line": null
}POST /v1/data-requests:draft
public, no key needed · 0 credits, not charged during the beta · MCP tool create_data_request · status live
A letter asking a betting tool for the caller's own data (saved systems, filters, alerts, bets) in a machine-readable file under the GDPR, the UK GDPR or the CCPA, with the legal deadline and next steps. Written from a fixed template; Kelly sends nothing and stores nothing.
Body, application/json (required):
platformstringThe service holding the person's systems, by name.
min length 1 · max length 80
lawstringgdpr (default), uk_gdpr or ccpa.
One of: gdpr, uk_gdpr, ccpa (open: new values may be added).
includearray of stringWhat to ask for: systems, alerts, bets, results, account (default systems, alerts, bets).
min items 1 · max items 5
requester_namestringThe account holder's name for the letter; omitted, a placeholder.
max length 80
account_loginstringThe email or username on the account; omitted, a placeholder.
max length 120
200 The data request facts.
objectconstrequiredAlways data_request.
Always "data_request".
platformstringThe service it is addressed to.
lawstringgdpr, uk_gdpr or ccpa.
law_namestringThe law, in words.
subjectstringrequiredThe email's subject.
bodystringrequiredThe letter.
includearray of stringWhat it asks for.
deadlinestringrequiredWhen the reply is due, as the law sets it.
deadline_daysintegerDays the law allows before any extension.
who_can_use_itstringWho the law covers.
send_tostringWhere to send it.
next_stepsarray of stringWhat to do next.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large
POST https://api.askkelly.ai/v1/data-requests:draft
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"platform": "Example Tool",
"law": "gdpr",
"include": [
"systems",
"bets"
]
}
{
"object": "data_request",
"platform": "Example Tool",
"law": "gdpr",
"law_name": "the EU General Data Protection Regulation",
"subject": "Data access and portability request (GDPR): [your full name]",
"body": "To the Example Tool privacy team,\n\nThis is a request from the holder of the Example Tool account below, under Articles 15 (right of access) and 20 (right to data portability) of the EU General Data Protection Regulation.\n\nAccount: [the email address or username on your account]\nName: [your full name]\n\nPlease send a copy of the following personal data you hold about this account holder:\n\n1. Every betting system, query, filter set, angle or trend created or saved on the account, with each condition, setting and parameter exactly as stored, its name and the dates it was created and last changed.\n2. Every bet, pick or play logged or tracked on the account, with its date, event, market, side, line, price, stake and result.\n\nPlease provide it in a structured, commonly used and machine-readable format (Article 20(1)), such as CSV or JSON, so that it can be moved to another service. Where you hold a system or filter set as settings rather than text, please export the settings themselves (each field and its value), not a screenshot.\n\nIf you need to confirm the account holder's identity, please reply to this email with what you need. If you will not provide any of the data above, please say which part and the reason.\n\nThank you,\n[your full name]\n",
"include": [
"systems",
"bets"
],
"deadline": "one month from receipt, which the controller may extend by two further months for complex or numerous requests if it tells you why within the first month (Article 12(3)).",
"deadline_days": 30,
"who_can_use_it": "people in the EU or EEA, or any customer of a company established there",
"send_to": "The privacy or data-protection contact named in Example Tool's privacy policy (often a privacy@ address or a request form). Send it from the email address on your account.",
"next_steps": [
"Send the letter from the email address on your account and keep a copy with the date.",
"When Example Tool sends the file, bring it to Kelly's import (a file or a paste): Kelly maps each saved system to a rule and lists any condition it can't map.",
"If no reply comes by the deadline, write again citing your first request's date."
],
"summary": "A GDPR data request to Example Tool for systems, bets; the reply is due within 30 days.",
"assumptions": [
"Kelly writes the letter; you send it from your own email. Nothing you enter here is stored."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_6dff1262472cb88612da987c",
"live_topics": [],
"rg_line": null
}POST /v1/rules:test_exchange
public, no key needed · 1 credit, not charged during the beta · MCP tool test_exchange_strategy · status live
An exchange trading strategy run tick by tick through recorded exchange markets (ADR-K681): enter when a formula on the runner's order book is true (back or lay; take the best price, join the queue or offer a price), then green up at a take-profit, stop-loss, exit rule or time before the off, or hold to the result. Waiting orders queue behind the money shown at their price; unmatched orders lapse at the off unless kept; removals cut earlier odds; commission is on each market's net winnings. Returns the verdict and its checks, orders, fills, how trades ended, gross, commission, net, ROI on stake matched, a t-test, stress tests (a tick worse, more commission, the other fill model), each market and every trade. Until Betfair Historical Data is loaded it runs on made-up sample markets and says so.
Body, application/json (required):
entry_whenstringWhen to enter: a formula true or false on each tick of a runner (=AND(seconds_to_off <= 300, price_rank = 1, wom_back > 0.6)). The first tick it is true places an entry order.
min length 2 · max length 2000
entry_sidestringback (bet it to win) or lay (bet it to lose). Default back.
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price. Default take.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
min 1.01 · max 1000
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one. Default 0.
min -50 · max 50
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken. Default 0.
min 0 · max 50
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability). Default 10.0.
max 100000
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose. Default stake.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed. Default 1.
min 1 · max 50
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
min 0 · max 86400
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse). Default false.
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
min 1 · max 300
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
min 1 · max 300
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind). Default false.
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
min length 2 · max length 2000
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
min -86400 · max 86400
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result. Default true.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%). Default 0.05.
min 0 · max 0.2
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result. Default 1.0.
max 100
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic). Default queue.
One of: queue, touch (open: new values may be added).
datastringsample (default; made-up markets until Betfair Historical Data is loaded) or betfair.
One of: sample, betfair (open: new values may be added).
event_typestringOnly this sport's markets.
One of: horse_racing, greyhounds, harness, soccer, tennis, cricket, basketball, american_football, ice_hockey, baseball, golf, rugby_union, rugby_league, other (open: new values may be added).
market_typestringOnly this market type (WIN, PLACE, MATCH_ODDS...).
pattern ^[A-Za-z0-9_]{2,40}$
marketsarray of stringOnly these market IDs (1.234567890).
min items 1 · max items 500
from_datestringFirst off date (YYYY-MM-DD, UTC).
format date
to_datestringLast off date (YYYY-MM-DD, UTC).
format date
detailstringfull (default) or summary (no markets or trades lists).
One of: full, summary (open: new values may be added).
200 The exchange test facts.
objectconstrequiredAlways exchange_test.
Always "exchange_test".
strategyobjectrequiredThe strategy as run: every action (exchange.contract.ACTIONS) with its value.
dataobjectrequiredThe markets it ran on.
sourcestring | arraysample (made-up markets) or betfair_historical; a list when markets came from both.
samplebooleanTRUE when every market is a made-up sample: the figures test the simulator, not the strategy.
marketsintegerMarkets run.
framesintegerMoments stepped through.
ticksintegerRunner rows read.
event_typesarray of stringSports.
fromstringnullableThe first market's off.
tostringnullableThe last market's off.
verdictobjectrequiredThe headline and its checks.
headlinestringrequiredThe verdict in words.
checksarray of objectrequiredEach check.
keystringrequiredThe check.
One of: enough_trades, made_money, not_luck, tick_worse, queue_fills, newest_30 (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
samplebooleanSample markets: no verdict is passed.
overallobjectrequiredEvery trade together.
orders_placedintegerEntry orders placed.
tradesintegerTrades with a matched entry.
never_matchedintegerEntries that matched nothing.
stake_askednumberEntry stake asked for.
stake_matchednumberEntry stake matched.
fill_ratenumbernullablestake_matched over stake_asked.
greenedintegerTrades closed level on every outcome.
held_to_resultintegerTrades held to the result.
exitsobjectGreened trades by how they got out.
take_profitintegertake_profit
stop_lossintegerstop_loss
trailing_stopintegertrailing_stop
exit_whenintegerexit_when
before_offintegerbefore_off
exit_blockedintegerTrades whose exit couldn't be placed or lapsed.
wonintegerTrades with net above 0.
lostintegerTrades with net below 0.
levelintegerTrades with net exactly 0.
grossnumberProfit before commission.
commissionnumberCommission paid.
netnumberProfit after commission.
roinumbernullablenet over stake_matched.
mean_netnumbernullableAverage net a trade.
sd_netnumbernullableStandard deviation of net a trade.
tnumbernullablet statistic of the mean net against 0.
p_valuenumbernullableOne-sided p-value of the t-test (trades treated as independent).
avg_ticksnumbernullableAverage ticks captured on greened trades.
max_drawdownnumberLargest fall in cumulative net, trade by trade.
small_samplebooleanUnder 30 matched trades.
stressobjectrequiredThe same strategy under worse conditions.
one_tick_worseobjectEvery take matched one tick worse.
tradesintegerTrades with a matched entry.
stake_matchednumberEntry stake matched.
netnumberProfit after commission.
roinumbernullablenet over stake_matched.
commission_plus_3ptsobjectCommission three points higher.
tradesintegerTrades with a matched entry.
stake_matchednumberEntry stake matched.
netnumberProfit after commission.
roinumbernullablenet over stake_matched.
other_fill_modelobjectThe other fill model (queue or touch).
tradesintegerTrades with a matched entry.
stake_matchednumberEntry stake matched.
netnumberProfit after commission.
roinumbernullablenet over stake_matched.
newest_30objectnullableThe newest 30% of markets (10 or more markets).
marketsintegerMarkets.
tradesintegerTrades.
netnumberNet.
roinumbernullableROI.
by_event_typearray of objectEach sport, by name.
event_typestringThe sport.
tradesintegerTrades.
stake_matchednumberStake matched.
wonintegerWon.
lostintegerLost.
netnumberNet.
roinumbernullableROI.
left_outobjectrequiredWhat didn't become a trade, or settled void.
entry_signalsintegerTicks where the entry rule fired and an order could go in.
no_priceintegerSignals with no price to take or join.
lapsed_in_playintegerUnmatched orders lapsed at the turn in play.
below_min_stakeintegerGreen-ups below the exchange's smallest stake (held instead).
removed_voidintegerTrades on removed runners (void).
unsettledintegerTrades in markets with no result yet.
marketsarray of objectEach market, in order of the off (detail full).
market_idstringThe exchange's market ID.
event_namestringnullableThe event.
event_typestringnullableThe sport.
market_typestringnullableWIN, MATCH_ODDS and so on.
off_timestringnullableThe scheduled off (RFC 3339).
sourcestringnullablesample or betfair_historical.
tradesintegerTrades with a matched bet.
grossnumberProfit before commission.
commissionnumberCommission on the market's net winnings.
netnumberProfit after commission.
tradesarray of objectEvery trade, in order (detail full).
keystring{market_id}:{selection_id}:{entered_at}.
market_idstringThe exchange's market ID.
event_namestringThe event.
event_typestringThe sport.
selection_idstringThe runner's ID in the market.
runnerstringThe runner.
sidestringback or lay.
One of: back, lay (open: new values may be added).
entered_atstringWhen the entry order went in (RFC 3339).
format date-time
seconds_to_offnumbernullableSeconds to the scheduled off then.
in_playbooleanEntered in play.
entry_pricenumbernullableThe matched entry's average price, decimal.
stake_askednumberThe entry stake asked for (a lay's backer's stake).
stake_matchednumberThe entry stake matched.
shownnumberMoney shown at the best price on the side taken or joined, when entered.
unmatched_cancellednumberEntry stake cancelled unmatched (fill or kill, the off, an exit).
exitstringHow it ended: take_profit, stop_loss, trailing_stop, exit_when, before_off (greened), held (to the result) or never_matched.
exit_notestringnullableWhy an exit didn't happen (below min_stake, lapsed at the off, never matched).
exited_atstringnullableWhen the green-up finished matching (RFC 3339).
format date-time
exit_pricenumbernullableThe green-up's average price, decimal.
ticksintegernullableTicks from entry to exit, positive in profit (greened trades).
betsintegerMatched bets the trade made (each partial fill is one).
resultstringnullableThe runner's result: winner, loser, placed or removed.
grossnumberProfit before commission.
commissionnumberThe trade's share of its market's commission.
netnumberProfit after commission.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
enginestringThe simulator's revision (results move only when it changes).
contract_versionstringThe tick contract's version.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad
POST https://api.askkelly.ai/v1/rules:test_exchange
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"entry_when": "=result = 1"
}
{
"type": "https://docs.askkelly.ai/problems/rule_not_testable",
"title": "Rule can't be tested",
"status": 422,
"code": "rule_not_testable",
"detail": "No attribute named result on tick rows.",
"param": "entry_when",
"doc_url": "https://docs.askkelly.ai/problems/rule_not_testable",
"request_id": "req_5ad8a2412ec1d129ea117878",
"retryable": false,
"display_message": "That rule can't be tested on past games as written.",
"problems": [
{
"code": "unknown_name",
"path": "entry_when",
"message": "No attribute named result on tick rows.",
"fixes": []
}
]
}POST https://api.askkelly.ai/v1/rules:test_exchange
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"entry_when": "=AND(seconds_to_off <= 300, price_rank = 1)",
"take_profit_ticks": 2,
"stop_loss_ticks": 3,
"exit_before_off_s": 10,
"detail": "summary"
}
{
"object": "exchange_test",
"strategy": {
"entry_when": "=AND(seconds_to_off <= 300, price_rank = 1)",
"entry_side": "back",
"entry_order": "take",
"entry_price": null,
"entry_offset_ticks": 0,
"max_slippage_ticks": 0,
"stake": 10,
"stake_type": "stake",
"max_entries": 1,
"fill_or_kill_s": null,
"keep_in_play": false,
"take_profit_ticks": 2,
"stop_loss_ticks": 3,
"trailing_stop": false,
"exit_when": null,
"exit_before_off_s": 10,
"green_up": true,
"commission": 0.05,
"min_stake": 1,
"fill_model": "queue"
},
"data": {
"source": "sample",
"sample": true,
"markets": 3,
"frames": 1004,
"ticks": 6308,
"event_types": [
"greyhounds",
"horse_racing",
"soccer"
],
"from": "2026-03-08T20:00:00.000Z",
"to": "2026-03-08T20:45:00.000Z"
},
"verdict": {
"headline": "Sample markets only: made-up data that tests the simulator, not the strategy",
"checks": [
{
"key": "enough_trades",
"label": "At least 30 matched trades",
"status": "fail",
"value": "3 matched trades"
},
{
"key": "made_money",
"label": "Made money after commission",
"status": "pass",
"value": "net +0.07"
},
{
"key": "not_luck",
"label": "Not just luck (one-sided t-test, 5%)",
"status": "fail",
"value": "p = 0.458"
},
{
"key": "tick_worse",
"label": "Still makes money with every take a tick worse",
"status": "fail",
"value": "net -2.24"
},
{
"key": "newest_30",
"label": "The newest 30% of markets made money",
"status": "na",
"value": "too few markets (under 10)"
}
],
"passed": 1,
"applicable": 4,
"sample": true
},
"overall": {
"orders_placed": 3,
"trades": 3,
"never_matched": 0,
"stake_asked": 30,
"stake_matched": 30,
"fill_rate": 1,
"greened": 3,
"held_to_result": 0,
"exits": {
"take_profit": 2,
"stop_loss": 1,
"trailing_stop": 0,
"exit_when": 0,
"before_off": 0
},
"exit_blocked": 0,
"won": 2,
"lost": 1,
"level": 0,
"gross": 0.1,
"commission": 0.03,
"net": 0.07,
"roi": 0.0023,
"mean_net": 0.0233,
"sd_net": 0.3356,
"t": 0.12,
"p_value": 0.4576,
"avg_ticks": 0,
"max_drawdown": 0.35,
"small_sample": true
},
"stress": {
"one_tick_worse": {
"trades": 3,
"stake_matched": 30,
"net": -2.24,
"roi": -0.0747
},
"commission_plus_3pts": {
"trades": 3,
"stake_matched": 30,
"net": 0.06,
"roi": 0.002,
"commission": 0.08
},
"other_fill_model": {
"trades": 3,
"stake_matched": 30,
"net": 0.07,
"roi": 0.0023,
"fill_model": "touch"
}
},
"newest_30": null,
"by_event_type": [
{
"event_type": "greyhounds",
"trades": 1,
"stake_matched": 10,
"won": 1,
"lost": 0,
"net": 0.3,
"roi": 0.03
},
{
"event_type": "horse_racing",
"trades": 1,
"stake_matched": 10,
"won": 1,
"lost": 0,
"net": 0.12,
"roi": 0.012
},
{
"event_type": "soccer",
"trades": 1,
"stake_matched": 10,
"won": 0,
"lost": 1,
"net": -0.35,
"roi": -0.035
}
],
"left_out": {
"entry_signals": 4,
"no_price": 1,
"lapsed_in_play": 0,
"below_min_stake": 0,
"removed_void": 0,
"unsettled": 0
},
"engine": "1",
"contract_version": "1.0.0",
"assumptions": [
"These are made-up sample markets (Kelly holds no real exchange ticks yet): the figures show how the simulator works, not whether the strategy makes money.",
"Your orders don't move later prices: the recorded book is what it was, so a stake large against the money shown matches more easily here than it would have (each trade lists the money shown).",
"A waiting order starts behind the money already shown at its price and matches only once that much has traded there (behind the third level's money when its price is further out).",
"Commission is 5% of each market's net winnings; discounts, premium charges and the exchange's own fees on high-volume accounts aren't applied.",
"Bets matched on other runners before a removal settle at odds cut by the removed runner's adjustment factor; bets on a removed runner are void.",
"Unmatched orders lapse when the market turns in play; a position whose exit lapsed is held to the result."
],
"as_of": "2026-10-04T17:30:00Z",
"summary": "Sample markets only: made-up data that tests the simulator, not the strategy: 3 trades backing in 3 sample markets, 3 greened and 0 held to the result, net +0.07 after 0.03 commission, +0.2% on stake matched (1 of 4 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_eb036ecfb9db90e9513b93c4",
"live_topics": [],
"rg_line": null
}POST /v1/rules:test_batch
public, no key needed · 1 credit, not charged during the beta · MCP tool test_rules_batch · status live
Up to 100 betting rules tested on the same seasons and options in one call: a list of formulas, or a template and a grid of values (every combination is a rule). One row each, in the order asked: the record, units, ROI, p-value and the p-value adjusted for every rule tried, this batch's included (tests_charged). A rule the tester refuses keeps its row with its problems and isn't counted. Rows are never sorted by a result.
Body, application/json (required):
formulasarray of stringRules to test, each a formula true or false on each team-game, starting with = (up to 100 with the grid's).
max items 100
formula_templatestringA formula with {name} slots the grid fills, e.g. =AND(home, rest_days >= {k}).
min length 2 · max length 2000
gridarray of stringEach slot's values as name=numbers, e.g. ["k=1,2,3"]; every combination is a rule.
min items 1 · max items 4
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...; racing: horse_racing, greyhounds or harness (seasons are calendar years).
pattern ^[a-z0-9_]{2,24}$
seasonsarray of stringSeasons to test, oldest first (1 to 25), e.g. ["2021", "2022"]; or use from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
marketstringspread (default), moneyline, over, under or draw, as test_rule reads them (a teaser rule is tested with test_rule).
One of: spread, moneyline, over, under, draw (open: new values may be added).
sidestringrule (default) bets the team the rule matched; fade bets the other side.
One of: rule, fade (open: new values may be added).
bet_momentstringclose (default), edition (football: the weekly sheet's Monday 10 AM ET line) or day_before (other sports: 24 hours before the start).
One of: close, edition, day_before (open: new values may be added).
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
stakestringrisk (default): 1 unit risked a bet; to_win: each bet sized to win 1 unit.
One of: risk, to_win (open: new values may be added).
pricestringconsensus (default): Kelly's price at the bet moment; flat_110: every bet at -110.
One of: consensus, flat_110 (open: new values may be added).
testsintegerRules tried on this data outside Kelly, this one included (default 1). Kelly counts the account's own; this may only raise that count.
min 1 · max 100000
200 The rule batch facts.
objectconstrequiredAlways rule_batch.
Always "rule_batch".
ruleobjectThe options every rule shared: competition, market, side, bet moment, seasons, season type, stake, price, and the template and grid when given.
resultsarray of objectrequiredOne row a rule, in the order asked: the formulas, then the grid's combinations.
formulastringrequiredThe formula tested (a template filled with its values).
valuesobjectnullableThe grid's values for this rule; null for a formula given as is.
rule_hashstringnullableThe rule's identity (see test_rule's rule.rule_hash); null when refused.
testedbooleanrequiredFalse when the tester refused the rule (see problems).
problemsarray of objectrequiredWhy it was refused, each with fixes; empty when tested.
codestringlook_ahead, edge_rule, syntax, ...
messagestringIn words.
left_outobjectGames left out: no line at the bet moment, or the rule matched both teams.
betsintegerGraded bets.
decisionsintegerWins plus losses.
recordstringW-L-P.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableThe p-value adjusted for every rule tried (tests): Bonferroni.
small_samplebooleanUnder 30 decisions.
variantsintegerrequiredRules in the batch.
testedintegerrequiredRules tested.
refusedintegerrequiredRules the tester refused (not counted).
tests_chargedintegerrequiredRules new to the account's ledger from this batch.
testsintegerrequiredThe rules-tried count every p_adjusted uses.
tests_on_accountobjectnullableThe account's ledger after this batch: competition, market, count; null without an account.
data_versionstringThe data every rule read.
noticestringIt isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/rules:test_batch
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula_template": "=AND(home, rest_days >= {k})",
"grid": [
"k=0,1,2"
],
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"object": "rule_batch",
"rule": {
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"formula_template": "=AND(home, rest_days >= {k})",
"grid": [
"k=0,1,2"
]
},
"results": [
{
"formula": "=AND(home, rest_days >= 0)",
"values": {
"k": 0
},
"rule_hash": "sha256:ef4d674ec75c877a5b37cfc1e0b4116e4014ff75d00a93d9139071c3ff450616",
"tested": true,
"problems": [],
"left_out": {
"no_line": 2,
"both_sides": 0
},
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"p_adjusted": 1,
"small_sample": true
},
{
"formula": "=AND(home, rest_days >= 1)",
"values": {
"k": 1
},
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"tested": true,
"problems": [],
"left_out": {
"no_line": 2,
"both_sides": 0
},
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"p_adjusted": null,
"small_sample": true
},
{
"formula": "=AND(home, rest_days >= 2)",
"values": {
"k": 2
},
"rule_hash": "sha256:1e9e3c96193481744619102207b3b7c22de187b64f1fdd3c062d04c31cc8d451",
"tested": true,
"problems": [],
"left_out": {
"no_line": 1,
"both_sides": 0
},
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"p_adjusted": null,
"small_sample": true
}
],
"variants": 3,
"tested": 3,
"refused": 0,
"tests_charged": 2,
"tests": 5,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 5
},
"data_version": "snap_52",
"notice": "This tests your rules on past games. It isn't a pick and doesn't rank selections.",
"summary": "(3) rules tested together, (3) run and (0) refused; each p-value is adjusted for (5) rules tried. Test any one with test_rule for its verdict.",
"assumptions": [
"Each rule's p_adjusted counts every rule tried on NBA spread (5), this batch's included.",
"Rows keep the order asked for; they are never sorted by a result."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_ea7a1a4a7699228aeb77d682",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:test_batch
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula_template": "=AND(home, rest_days >= {k})",
"grid": [
"days=1,2"
],
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "The template's slots (k) and the grid's keys (days) must match.",
"param": "grid",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_6ca470b3bef4bd2066792cba",
"retryable": false,
"display_message": "Something in that request was not understood."
}GET /v1/rules/runs/{run_id}
public, no key needed · 0 credits, not charged during the beta · MCP tool get_rule_run · status live
One of your own kept rule tests by its run_id: the report it returned, on the data it read (data_version), without the bets list. Runs are private: another account's run is not_found.
run_idstring · pathrequiredThe run_id a test_rule result gave (run_…).
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
200 The rule test facts.
objectconstrequiredAlways rule_test.
Always "rule_test".
verdictobjectrequiredVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
overallobjectrequiredEvery graded bet.
recordstringWins-losses(-pushes).
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
betsintegerGraded bets.
decisionsintegerWins plus losses.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_intervalarray of numbernullableROI's 95% interval at 1 unit risked a bet.
win_ratenumbernullableWins over decisions.
break_evennumbernullableThe win rate that breaks even at these prices.
significanceobjectnullableThe record test (kmath.records.record_test) and verdict v2's statistics (kmath.significance): the p-value, adjusted for the rules tried, and the anytime-valid test.
p_valuenumbernullableThe larger of the exact and the clustered one-sided p against no skill at these prices (F12, F23).
methodstringnullableexact_binomial, exact_simulation, saddlepoint or z_score.
znumbernullableThe score z: units over the square root of Σ(d − 1).
t_statnumbernullableThe t statistic with the sample variance.
p_value_exactnumbernullableThe exact (or z) tail alone.
p_value_clusterednumbernullableThe tail with bets clustered by slate and team-season (F23).
p_value_vs_vignumbernullableAgainst a random bet at the no-vig close (needs closes).
p_adjustednumbernullableThe p-value adjusted for the rules tried (Romano-Wolf on the ledger's rules, Bonferroni on the count otherwise; F14).
adjust_methodstringnullableromano_wolf, holm or none.
testsintegernullableRules tried in this family, this one included.
tests_effectivenumbernullableEffective tests (Li-Ji): near-duplicate rules count as fewer.
reality_check_pnumbernullableWhite's Reality Check p (when the ledger holds the rules' units).
spa_pnumbernullableHansen's SPA p (when the ledger holds the rules' units).
clears_t3booleannullablez at or above 3 (Harvey-Liu-Zhu's bar).
haircutnumbernullableHarvey-Liu haircut: the share of the ROI the rules tried take away.
haircut_roinumbernullableROI after the haircut, per decision.
anytime_validobjectnullableThe record as a bet against no skill (F17): e-value, anytime p, ROI range.
e_valuenumbernullableThe e-value now.
p_valuenumbernullable1 over the largest e-value so far.
max_enumbernullableThe largest e-value along the path.
roi_cs95array of numbernullableAn ROI range valid at every bet.
roi_ci95array of numbernullableROI's 95% score interval, clustered (F13, F23).
units_ci95array of numbernullableUnits' 95% interval: decisions times the ROI interval.
roi_ciobjectnullableROI intervals at ci_level: score (the one shown), BCa, block bootstrap over slates and anytime-valid (F13).
deflatedobjectnullablePer-bet Sharpe, PSR and DSR at the effective tests, and the bets needed: deflated (sample_needed), single test and 80% power, each times the design effect (F15).
sample_neededintegernullableDecisions needed, deflated for the rules tried; null with no edge to prove.
sample_needed_singleintegernullableDecisions needed for one rule tried.
sample_needed_power80integernullableDecisions for 80% power at this ROI.
sample_needed_planintegernullableDecisions for 80% power at plan_roi.
dsrnumbernullableDeflated Sharpe ratio.
psrnumbernullableProbabilistic Sharpe ratio.
sample_neededobjectnullableBets needed to tell this win rate from break-even (one-sided, with 80% power, ...).
small_samplebooleanUnder 30 decisions.
few_decisionsbooleanUnder 5 decisions.
honestobjectnullableThe honest forward ROI (F22, the headline ROI): empirical Bayes toward a prior mean (the rule's own closing-line value when at least half its bets have a close, else the usual two-way margin), less the winner's curse for the rules tried and the out-of-sample decay.
roinumbernullableHonest forward ROI.
ci80array of numbernullable80% range.
raw_roinumbernullableThe past ROI per decision, beside it.
prior_meannumbernullableThe prior mean the ROI is shrunk toward (prior.source says which).
prior_sdnumbernullableThe spread of true ROIs (0.02 until fitted on the ledger).
shrinknumbernullableWeight on the prior, se²/(se² + τ²).
selection_adjnumbernullableWinner's-curse correction for the rules tried.
decay_factornumbernullableOut-of-sample decay on the excess (0.74).
stabilizes_at_betsintegernullableDecisions at which the record and the prior weigh the same.
uses_clvbooleanThe prior mean is the rule's own closing-line value.
prior_basisstringfitted (prior_sd from the ledger's fit) or seed.
priorobjectWhere the prior mean comes from.
sourcestringclv (the rule's mean closing-line value), margin (the two-way margin, -4.55%) or fitted.
meannumbernullableThe prior mean used.
closesintegerDecided bets with a no-vig close.
cappedbooleanA closing-line value above 0 was held at 0: under 50 bets have a close.
wordsstringThe prior, said plainly.
dependenceobjectnullableBets clustered by slate (the date; the week in the NFL and college football) and team-season (F23).
clustersintegerSlates.
design_effectnumberClustered over iid variance.
effective_betsintegerDecisions over the design effect (at least 1).
methodstringcr1_twoway, cr1_slate, cr1_larger_oneway, wild_bootstrap or wild_bootstrap_webb.
baseobjectnullableThe rule against the rest of its own base (F19): the who-or-when conditions alone, or base_formula.
formulastringnullableThe base, or null for the whole market.
sourcestringauto, user or market.
One of: auto, user, market (open: new values may be added).
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
roinumbernullableThe base's ROI per decision.
liftnumbernullableRule ROI less base ROI.
p_value_vs_basenumbernullableSeason-stratified permutation p of the rule within its base.
units_from_basenumbernullableUnits the base alone would have made on the rule's stakes.
units_from_filternumbernullableThe rule's units less units_from_base.
clvobjectrequiredClosing-line value as evidence (F21): each bet's no-vig CLV, q_close x price - 1, on bets Kelly holds a close for. Empty when betting at the close.
bets_with_closeintegerBets with a close.
bets_without_closeintegerBets without one.
referencestringThe close used: consensus (the no-vig average across books).
devigstringHow the vig was removed: multiplicative.
lines_convertedintegerBets whose close was carried from the closing line to the bet's line (NFL spreads and totals).
coveragenumbernullableShare of bets with a close probability (null with no bets).
closes_staleintegerCloses left out as older than 60 minutes at the start, or seen after it (audit F9): those bets keep their result but have no CLV.
points_coveragenumbernullableShare of bets with points beaten (spreads and totals).
points_testobjectnullableThe clustered test on points beaten (audit F3).
betsintegerBets with points beaten.
meannumbernullableMean points the bet's line beat the close by.
std_error_clusterednumbernullableStandard error clustered by slate and team-season.
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean points beaten > 0.
cluster_methodstringHow the standard error was clustered.
basisstringWhat Beats the close is judged on: probability, or points when under 80% of bets have a close probability and more have points beaten (audit F3).
One of: probability, points (open: new values may be added).
meannumbernullableMean CLV (fraction): the bets' expected ROI if the close is right.
ci95array of numbernullableClustered 95% interval on the mean.
min items 2 · max items 2
p_valuenumbernullableOne-sided clustered p of mean CLV > 0.
beat_ratenumbernullableShare of bets that beat the close, ties left out.
beat_rate_ci95array of numbernullableWilson 95% interval on beat_rate.
min items 2 · max items 2
sign_test_pnumbernullableExact one-sided sign test against half.
mean_probnumbernullableMean of the close's no-vig probability less the bet moment's.
mean_pointsnumbernullableMean points the bet's line beat the closing line by (spreads and totals).
points_betsintegerBets mean_points is over.
profit_slopenumbernullablePer-bet units regressed on CLV: near 1 when CLV turned into profit.
profit_slope_senumbernullableIts slate-clustered standard error.
sample_neededintegernullableBets CLV needs for a 95% interval to clear 0 at this mean and spread (null unless the mean is above 0).
sample_needed_roiintegernullableBets profit would need at the same edge.
significanceobjectnullableThe CLV mean's tests (null under two bets).
tnumbernullableMean over its iid standard error.
ci95array of numbernullableStudent 95% interval (iid).
min items 2 · max items 2
std_errornumbernullableiid standard error.
bets_needednumbernullableBets a t of 2 needs at this mean and spread.
std_error_clusterednumbernullableStandard error clustered by slate and team-season (F23).
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean CLV > 0 (wild cluster bootstrap under 42 slates, else Student t); beats_close reads it.
cluster_methodstringHow the standard error was clustered.
design_effectnumbernullableClustered variance over iid variance.
notestringWhy the block is empty, when it is.
luckobjectThe luck-adjusted result (F20): the units the market's fair probabilities expected the decided bets to make against what they made. Pushes count zero both ways.
readingstringluck (bets before the close: the gap is outcome noise) or versus_the_market (bets at the close: the gap is the rule's edge on the closing market).
One of: luck, versus_the_market (open: new values may be added).
close_sourcestringconsensus.
devigstringmultiplicative.
betsintegerDecided bets with a fair probability.
bets_without_probintegerDecided bets without one.
left_outobjectBets left out.
pushintegerPushes.
halfintegerHalf results (quarter lines).
lines_convertedintegerBets whose close was carried to the bet's line.
expected_unitsnumbernullableSum of stake x (q x (price - 1) - (1 - q)).
actual_unitsnumbernullableUnits the bets made.
luck_unitsnumbernullableactual_units less expected_units.
luck_se_unitsnumbernullableluck's standard error given the probabilities.
expected_winsnumbernullableSum of q.
actual_winsintegernullableWins.
luck_percentilenumbernullableWhere the wins sit among the market's (Poisson-binomial, mid-p): 0.95 is luckier than (95) of (100) identical paths.
expected_roinumbernullableexpected_units over units risked.
actual_roinumbernullableactual_units over units risked.
thetanumbernullablePer-bet units regressed on expected units (30 bets or more): near 1 when the close is calibrated on these bets.
by_seasonarray of objectEach season, in calendar order; luck_units add up to the total.
seasonstringSeason.
betsintegerDecided bets with a probability.
expected_unitsnumberUnits the market expected.
actual_unitsnumberUnits made.
luck_unitsnumberactual_units less expected_units.
capacityobjectnullableF8: the limits Kelly measured at the bet moment (ADR-K783); evidence beside the verdict, never in it. No book publishes its limits, so only measured ones are shown.
availablebooleanWhether measured limits are shown for this market and moment.
marketstringnullableThe market as line history names it (spreads, totals, h2h).
before_startstringnullableThe time before the start the bet moment falls in.
One of: 0_1h, 1_12h, 12_36h, 36_72h, 72h_plus (open: new values may be added).
booksarray of objectPer book, highest median first.
bookstringnullableThe book as the vendor names it.
vendorstringnullableWho measured it.
currencystringnullableThe stake currency, null when the vendor doesn't say.
mediannumbernullableThe median limit.
p25numbernullableIts 25th percentile.
p75numbernullableIts 75th percentile.
minnumbernullableThe smallest.
maxnumbernullableThe largest.
snapshotsintegerLimit readings behind it.
gamesintegerGames behind it.
first_seenstringnullableThe first reading's date.
last_seenstringnullableThe last reading's date.
max_stake_per_betobjectnullableThe highest book median: the most one book typically took on one bet.
bookstringnullableThe book.
amountnumbernullableThe median limit.
currencystringnullableIts currency.
per_season_at_limitnumbernullablemax_stake_per_bet times the rule's bets a season.
bets_per_seasonnumbernullableThe rule's bets a season.
limits_heldintegerLimit readings Kelly holds for this market and moment.
reasonstringnullableWhy none are shown, when available is false.
stressobjectrequiredThe same bets graded harder.
half_point_worseobjectnullableEvery line half a point worse (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
flat_110objectnullableEvery bet at -110 (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
other_sideobjectThe other side of every bet.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
break_evenobjectnullableF7: how much worse the price or line can get before the profit is gone.
price_multipliernumbernullableThe share of each payout (d - 1) that breaks even (under 1: the room to spare).
american_avgnumbernullableThe flat American price that breaks even.
decimal_avgnumbernullableThe same, decimal.
line_pointsnumbernullableThe most points worse (in halves, up to 3) the line could be with a profit left; 0 when none; null for moneylines.
gridarray of objectnullableF7: the bets at -105 to -120 with the line 0 to 1 point worse (moneylines: payouts cut 0-6%); left out at detail summary.
vigintegernullableEvery bet at this American price.
line_shiftnumbernullablePoints the line moved (negative: worse).
price_multipliernumbernullableMoneylines: the share of each payout kept.
unitsnumbernullableUnits.
roinumbernullableROI.
slippage_mcobjectnullableF7: each bet's line worsened by a draw from the market's own Monday-to-close moves, 1,000 seeded runs; null without such moves.
runsintegerRuns.
roi_p05numbernullable5th percentile ROI.
roi_p50numbernullableMedian ROI.
moves_fromintegerLine moves the draws come from.
move_mediannumbernullableTheir median size, points.
price_worseobjectnullableTeasers: every ticket ten cents worse (price_american). Racing: every bet's winnings 5% smaller.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
legsobjectnullableTeasers: every teased leg, judged against the per-leg break-even (1/D)^(1/n).
recordstringLegs won-lost(-pushed).
decisionsintegerLegs won plus lost.
cover_ratenumbernullableLegs won over legs decided.
break_even_per_legnumberThe cover rate a leg needs at the ticket's price.
significanceobjectnullableThe legs' record test against the per-leg break-even.
sample_neededobjectnullableLegs needed to tell this cover rate from the break-even.
random_baselineobjectnullableF18: the rule against random bets of the same count on the same games, matched by season and price bucket.
simsintegerRandom rules drawn.
betsintegerThe rule's bets.
poolintegerBets the market offered.
percentilenumbernullableShare of random rules with a lower ROI.
beats_per_100integerRandom rules out of 100 the rule beats.
p_valuenumbernullable(1 + random rules at or above) / (sims + 1).
analytic_pnumbernullableThe finite-population normal cross-check.
random_roi_mediannumbernullableMedian random ROI (about minus the vig).
random_roi_p05numbernullable5th percentile random ROI.
random_roi_p95numbernullable95th percentile.
best_of_testsobjectWould the best of M random rules (M = rules tried) do this well?
testsintegerM.
p_valuenumbernullable1 - F(ROI)^M.
histogramobjectnullableRandom ROIs in 30 bins (edges, counts); full detail only.
decayobjectnullableF24: is the rule still making money? Rolling ROI and CLV, one unknown break (sup-F, block permutation p), a CUSUM alarm, a known-date test, the trend and the decay ratio.
betsintegerGraded bets.
windowintegernullableRolling window, bets.
rollingarray of objectnullableUp to 200 points; full detail only.
breakobjectnullableThe most likely date the edge changed.
atstringDate.
bet_indexintegerBets before it.
sup_fnumbernullablesup-F.
p_valuenumbernullableBlock permutation p-value.
at_edgebooleanWithin the first or last 15%: not a confirmed break.
roi_beforenumbernullableROI before.
roi_afternumbernullableROI after.
bets_afterintegerBets after.
cusumobjectnullablePage's lower CUSUM alarm.
known_breakobjectnullableA Chow test at break_date.
trend_per_100_betsnumbernullableProfit per bet's change per 100 bets (1 unit risked).
decay_rationumbernullableROI after the break (or the newest 30%) over ROI before.
clv_breakobjectnullableThe break test on CLV.
roi_newest_30numbernullableROI of the newest 30% of bets.
ablationobjectnullableF27: each top-level AND condition's Shapley share of the units beyond the base's ROI.
baseobjectEvery game with a line: decisions and ROI.
totalnumbernullableThe rule's units beyond the base's ROI (the Shapley values add up to it).
clausesarray of objectOne a condition, in the order written.
conditionstringThe condition.
shapley_unitsnumbernullableIts share of the extra units.
labelstringdead, harmful, load_bearing or helps.
One of: dead, harmful, load_bearing, helps (open: new values may be added).
fixobjectnullableRemove it: the rule without it.
pairsarray of objectPairs (up to 6 conditions).
methodstringexact or sampled.
One of: exact, sampled (open: new values may be added).
nearbyobjectnullableF28: every number in the rule moved two steps each way (steps fixed by the field).
numbersarray of objectThe rule's numbers.
variantsintegerVariants in the grid.
distinct_bet_setsintegerVariants with their own bets.
roi_mediannumbernullableMedian ROI of the variants.
roi_p05numbernullable5th percentile.
roi_p95numbernullable95th percentile.
spike_rationumbernullable(ROI - median) / IQR of the variants.
readingstringnullableplateau, slope or spike.
nastringnullableWhy there are no numbers to move.
monte_carloobjectnullableF31: drawdowns and losing runs the same bets give in other orders.
simsintegerSimulations.
modestringblock, iid or permutation.
max_drawdown_unitsobjectMax drawdown, units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
longest_losing_runobjectLongest losing run, bets.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
final_unitsobjectFinal units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
prob_below_zeronumbernullableShare of runs finishing below zero.
worst_100_bets_units_p05numbernullable5th percentile of the worst 100-bet stretch.
noiseobjectnullableF32: jittered inputs, lines moved half a point either way, and an earlier bet moment.
inputsobjectnullableThe rule's compared numbers jittered.
runsintegerRuns.
profitable_runsintegerRuns with a profit.
roi_mediannumbernullableMedian ROI.
roi_p05numbernullable5th percentile ROI.
overlap_mediannumbernullableMedian Jaccard overlap of the bets with the rule's.
linesobjectnullableLines (moneylines: payouts) moved at random.
delayobjectnullableThe rule bet at the earlier moment.
concentrationobjectnullableF34: the profit without the best season, team, opponent, month and weekday.
byarray of objectGroupings in a fixed order.
groupstringseason, team, opponent, month or weekday.
bestobjectnullableThe best group (3+ bets).
without_bestobjectnullableThe rule without it.
effective_groupsnumbernullable1 / Herfindahl of the profitable groups' shares.
p_more_concentratednumbernullablePermutation p-value (teams and opponents).
trim_top_0_05objectnullableWithout the top 5% of bets (moneylines) or bet-days.
split_dimensionsarray of objectnullableF35: the rule by fixed dimensions with shrunk ROI and intervals.
dimensionstringThe dimension.
heterogeneity_pnumbernullableChi-square p-value of win rates across its levels.
heterogeneity_qnumbernullableBenjamini-Hochberg q across the dimensions.
real_differencebooleanq < 0.10.
where_profitstringnullableThe level with the most shrunk extra units, when real.
cellsarray of objectnullableLevels in a fixed order; full detail only.
robustnessobjectThe suite's run: the verdict v2 checks it feeds and its seed.
checksarray of objecthalf_point_worse, nearby_holds, not_concentrated and still_working.
keystringrequiredThe check.
One of: half_point_worse, nearby_holds, not_concentrated, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na.
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumbernullableThe figure the status compares.
thresholdnumbernullableWhat it is compared with.
fast_pathbooleanThe rule's numbers were re-bet from its compiled columns.
seedintegerThe seed (a hash of the rule and its options).
evaluationsintegerFormula evaluations the suite ran.
offarray of stringBlocks switched off.
pathobjectThe units path in date order: drawdown and longest losing run; at detail full its series (F47).
max_drawdown_unitsnumberThe deepest fall from a peak, units.
longest_losing_runintegerMost losses in a row (pushes don't break it).
seriesobjectnullableBet-indexed lines of the graded bets (detail full): index the bet numbers kept (past 600 bets thinned by LTTB, the deepest drawdown always kept), dates, units, drawdown, base_units (the market base's ROI at the same stakes, null unless measured), band_p025 and band_p975 (the no-skill range: 1,000 seeded runs of the same bets at their fair chances; null under 20 bets or for racing), rolling_roi_100 with its no-skill range rolling_p025 and rolling_p975, and markers (season starts, the holdout).
xstringAlways bet_index.
One of: bet_index (open: new values may be added).
indexarray of integerBet numbers (1 is the first graded bet).
datesarray of stringEach kept bet's date (YYYY-MM-DD, UTC).
unitsarray of numberUnits to date.
drawdownarray of numberUnits below the running peak (0 or below).
base_unitsarray of numbernullableThe market base at the same stakes.
band_p025array of numbernullableNo-skill 2.5th percentile.
band_p975array of numbernullableNo-skill 97.5th percentile.
rolling_roi_100array of numberROI over the last 100 bets (null before).
rolling_p025array of numberIts no-skill 2.5th percentile.
rolling_p975array of numberIts no-skill 97.5th percentile.
markersarray of objectWhere each season and the holdout start.
typestringseason or holdout.
One of: season, holdout (open: new values may be added).
atintegerThe bet number it starts at.
labelstringnullableThe season, or newest 30%.
downsampledbooleanThe series were thinned to 600 points.
expected_max_drawdown_zero_edgenumbernullableThe median deepest drawdown of the no-skill runs, units.
holdoutobjectnullableThe latest 30% of bets against the earlier 70%.
seasonsarray of objectSeasons in calendar order.
splitsarray of objectnullableHome, away, favorite, underdog and line bands, in a fixed order; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
withoutarray of objectnullableOne row a top-level AND condition: the rule's record without it; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
p_value_vs_rulenumbernullableThe rule against the bets the condition removes (F19).
p_value_vs_rule_adjustednumbernullablep_value_vs_rule, Holm-adjusted over the conditions shown.
betsarray of objectEvery bet in date order (left out at detail summary).
keystring{game_id}:{team_id} (racing: {race_id}:{runner_id}).
starts_atstringKickoff (RFC 3339).
format date-time
seasonstringThe season.
labelstringDate, team and opponent.
priceobjectnullableThe price at the bet moment, decimal.
linenumbernullableThe bet's line as graded (the team's spread, the total).
resultstringnullablewin, loss or push (soccer quarter lines: half_win or half_loss); null until played.
close_probnumbernullableThe fair closing probability of the same side, for CLV.
units_to_datenumberUnits after this bet, in date order.
how_gradedstringnullableHow the bet was graded, in words.
unitsnumbernullableUnits the bet won or lost at the run's staking (racing: its exact profit or loss with dead heats, each way, Rule 4 and commission); null until played.
drawdown_to_datenumbernullableUnits to date less their running peak (0 or below); null until played.
sidestringnullablehome or away (spread, moneyline), over, under or draw.
marketstringThe market graded (the other total when fading a total).
bet_momentstringWhen the bet was taken: edition, day_before or close.
clv_probnumbernullableThe fair closing chance less the fair chance at the bet moment, when Kelly holds both.
samplestringnullableholdout for the latest 30% of graded bets by time, else in_sample; null with no holdout or until played.
One of: in_sample, holdout (open: new values may be added).
whyarray of objectWhy it matched: up to six conditions (F46).
clausestringThe condition as written.
valuenumber | boolean | stringnullableThe compared expression's number for a comparison (rest_days - opponent.rest_days for rest_days - opponent.rest_days >= 3), else true or false.
settlementarray of stringThe settlement rules that decided the bet (IDs in settlement.rules).
leak_testobjectThe leak test (F3): the rule and each part of it evaluated again on a sample of rows with the data cut at the bet moment. A leak refuses the test (422 rule_not_testable, problem leak_found), so a report always says clean or not_run.
statusstringrequiredclean, leak_found or not_run.
One of: clean, leak_found, not_run (open: new values may be added).
rows_checkedintegerrequiredRows rebuilt at the bet moment.
matched_checkedintegerOf them, rows the rule matched.
matches_flippedintegerRows whose match changed with the data cut.
partsarray of objectrequiredParts that changed, in the formula's order (empty when clean).
namestringThe part, as written (rule is the whole rule).
rows_changedintegerRows where it changed.
max_changenumbernullableThe largest change in a number.
examplestringOne changed row, {game_id}:{team_id}.
warm_upobjectThe warm-up check (F4): bets taken before the rule's windows had the earlier games they read (one season at a time; windows never reach into last season).
policystringThe policy graded in the report.
One of: as_written, require_full (open: new values may be added).
games_neededintegerEarlier games every window needs (0: the rule has no window).
windowsarray of objectEach window in the formula's order.
exprstringThe window, as written.
games_neededintegernullableEarlier games it reads (null: season to date).
short_rowsintegerRows in scope with fewer earlier games this season.
short_betsintegerBets the rule took with fewer.
short_betsintegerBets taken with a window short.
by_policyarray of objectnullableas_written, then require_full; null when the rule has no window.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
warningbooleanShort bets are over 10% of the bets, or leaving them out moves ROI by more than half its 95% interval's half-width.
settlementobjectHow bets were settled (F9): Kelly's own grading rules as data, and each rule's bets.
profilestringThe rules profile (kelly_standard).
rules_versionstringThe rules table's date.
appliedarray of objectRules that decided at least one bet, in table order.
rulestringRule ID.
betsintegerBets.
rulesarray of objectThe rules for this league and market.
idstringRule ID.
textstringThe rule in words.
effective_fromstringThe date Kelly's grader took it up.
coverageobjectCoverage (F10): from every game in scope to the bets graded, each season's priced share, blank inputs and whether the rows left out for no line differ from the bets kept.
flowarray of objectThe steps in order; rule_true rows = decisions + every row out after it.
stepstringin_scope, rule_true, one_bet_a_game, full_windows, priced, graded or decisions.
gamesintegerGames (in_scope only).
rowsintegerRows.
outobjectRows that left at this step, by reason.
by_seasonarray of objectEvery season asked for, in order.
seasonstringSeason.
gamesintegerGames in scope.
rowsintegerRows.
rule_trueintegerRows the rule matched.
betsintegerBets graded.
inputs_missingarray of objectInputs read bare on the row, with blanks.
attributestringThe input.
rows_blankintegerRows with no value.
left_out_differsobjectRows left out for no line against the bets kept: standardized mean difference over home, neutral site and rest days; flagged over 0.1 with 10 or more left out.
flagbooleanThey differ.
max_smdnumbernullableThe largest difference.
covariatestringnullableWhere it is.
left_outintegerRows.
keptintegerRows.
by_covariatearray of objectEach covariate's difference.
momentobjectThe bet moment's clock (F5): its target time, the prices' snapshot age at it and where the lines came from.
bet_momentstringedition, day_before or close.
targetstringnullableThe moment's time for a game, in words.
max_stale_minutesintegerOldest price kept, in minutes before the target.
floor_minutes_before_oct_2022integerThe limit's floor for games before October 2022 (the Odds API's 10-minute snapshots).
rows_pricedintegerRows the rule took that had a price.
rows_agedintegerRows whose price carries its snapshot time and was kept.
median_snapshot_age_minutesnumbernullableMedian minutes from snapshot to target, kept rows.
max_snapshot_age_minutesnumbernullableOldest kept snapshot, in minutes before the target.
stale_priceintegerRows left out: price older than max_stale_minutes.
after_momentintegerRows left out: price seen after the target.
age_unknownintegerRows kept whose price has no snapshot time (Kelly's own frozen lines).
line_sourcesobjectKept rows by line source (sheet_line, odds_api_early, odds_api_close).
left_outobjectrequiredGames the rule matched but didn't bet.
no_lineintegerNo Kelly line at the bet moment.
both_sidesintegerThe rule matched both teams.
stale_priceintegerThe price was older than max_stale_minutes at the bet moment, or seen after it (F5).
unpairedintegerTeasers: legs left off a ticket (a week with too few, or the leftover in chronological pairing).
no_priceintegerRacing: runners with no price to settle at.
non_runnersintegerRacing: non-runners (void).
void_racesintegerRacing: abandoned or void races.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
capsobjectnullableExposure caps applied in time order (F39); the record and checks are on the capped ledger.
skippedintegerBets the caps skipped.
by_reasonobjectSkipped bets by cap.
uncappedobjectThe record without the caps.
random_priority_mean_roinumbernullableMean ROI over random tie-breaks.
ruleobjectrequiredThe rule and the options it was tested with.
rule_hashstringThe rule's identity: SHA-256 of its canonical formula (spacing, case and the order of AND and OR conditions don't change it) with the competition, market, side, bet moment, seasons, season type, stake and price. Detail and tests are not part of it.
pattern ^sha256:[0-9a-f]{64}$
translation_idstringThe rules:translate result this formula came from (F43), when the request named one.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
readsarray of objectEvery name the rule reads, once.
namestringThe attribute, with its relation (opponent.rest_days).
knownstringIts moment: static, pregame, edition, close or final (a window's input is read before the game).
windowedbooleanA backward window reads it (earlier games only).
formula_versionstringThe formula language version the rule was read with (F45).
catalogue_versionstringnullableThe stat catalogue version its names were bound against.
reportobjectA shared report's own facts (get_rule_report only, F54).
report_idstringThe report.
report_urlstringIts page.
format uri
visibilitystringunlisted or public.
One of: unlisted, public (open: new values may be added).
author_rules_triedintegerThe author's rules tried on this market when shared.
frozenbooleanAlways true.
rerunobjectThe run's request, to test it again on today's data.
report_urlstringnullableA frozen report of this run anyone with the link can open (F54): only when the request asked (share) and the call has an account; else null. share_rule_run makes or changes one later.
format uri
chartsarray of objectOnly when the request asked (charts) at detail full: equity against the no-skill band, the drawdown and the rolling ROI as Kelly Chart Specs (F47).
transferobjectnullableOther markets and sports (F33): the same rule on its other market and on other leagues, numbers mapped to the same meaning, pooled with DerSimonian-Laird random effects. Evidence beside the verdict, never part of it; these runs don't count toward rules tried. Present only when the request asks for transfer.
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
whystringnullableWhy nothing ran, when nothing did.
optionsobjectThe leagues, markets and thresholds asked for.
cellsarray of objectEach market and league in a fixed order (the other market first, then sibling leagues), never sorted by a result; up to 12.
competitionstringThe league.
marketstringThe market.
thresholdsstringliteral (the rule's own numbers) or quantile (each number moved to the same percentile of its field in this league).
One of: literal, quantile (open: new values may be added).
formulastringThe formula this cell ran (quantile cells carry the moved numbers).
seasonsarray of stringSeasons read, as this league labels them.
mappedarray of objectnullableQuantile cells only.
numberstringThe number as written.
fieldstringWhat it is compared with.
percentilenumbernullableIts percentile in the rule's own league (a fraction).
valuenumbernullableThe number at the same percentile here.
decisionsintegerWins plus losses.
betsintegerGraded bets.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% score interval.
senumbernullableROI's standard error at 1 unit risked a bet.
p_valuenumbernullableThe cell's own one-sided p against no skill at its prices.
p_holmnumbernullableHolm's adjusted p across the cells.
pooledbooleanThe cell is in the pool (another league, one a league, 10+ decisions).
pooledobjectnullableThe pool of the other leagues' cells.
roinumbernullableRandom-effects ROI.
roi_ci95array of numbernullableIts 95% interval.
senumbernullableIts standard error.
p_valuenumbernullableTwo-sided p of the pooled ROI.
i2numbernullableI²: the share of the spread across leagues beyond chance (a fraction).
tau2numbernullableτ²: the between-league variance of true ROI.
qnumbernullableCochran's Q.
cellsintegerCells pooled.
same_directionobjectnullablePooled cells whose ROI has the rule's own sign.
agreeintegerCells the same way.
cellsintegerCells.
p_valuenumbernullableTwo-sided sign-test p.
textstring"4 of 5".
not_testablearray of objectLeagues and markets left out, with why.
competitionstringThe league.
marketstringnullableThe market.
missingarray of stringFields that league lacks.
whystringWhy.
notestringWhat the panel is and isn't.
run_idstringnullableThis run, kept on the account (run_…): the same request on the same data is the same run. GET /v1/rules/runs/{run_id} reads it back. Null without an account.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
data_versionstringThe data the run read: Kelly's lake snapshot (or the data's as-of time). A new snapshot is a new run of the same rule.
manifestobjectnullableWhat this run read and ran, so the same request can be run again and checked against it (ADR-K885). No prices, lines or rows of source data.
versionintegerThe manifest's own layout version.
codeobjectThe code that graded the run.
engineintegerThe tester's revision; it rises with any fix that can move a result on the same data.
releasestringnullableThe deployed commit.
formula_versionstringnullableThe formula language's version.
catalogue_versionstringnullableThe stat catalogue's version.
attributes_sha256stringnullableA hash of the attribute registry: what each input means and when it is known.
settlement_rulesstringnullableThe settlement rules' version.
dataobjectThe data the run read.
versionstringThe same value as data_version.
snapshotsarray of stringThe lake snapshots read, in order.
one_snapshotbooleanWhether every season was read on one snapshot.
as_ofstringnullableWhen the read was made (RFC 3339).
format date-time
data_throughobjectHow far Kelly's history went, by league.
latest_resultstringnullableThe newest game read that had a result.
rows_built_atstringnullableWhen the prepared rows were built, where the run read them.
format date-time
queriesobjectEach reviewed lake query the tester reads, with a hash of its text.
requestobjectThe request.
request_hashstringThe request's hash.
rule_hashstringThe rule's hash.
seedintegerThe seed every simulation in the run used.
sourcesobjectWhere the run's inputs came from.
pricesobjectBets priced, by the source of the line.
inputsarray of objectEach input the rule read, with the moment it is known.
namestringThe input.
knownstringnullableThe moment it is known.
seasonsarray of objectEach season read.
seasonstringThe season.
rowsintegerTeam-games in scope.
left_outobjectMatched games that were not bet, by reason.
sha256stringThe hash of this block. Two runs with the same hash read the same data with the same code.
since_runobjectnullableOnly on a run read back: what has changed in Kelly's tester since the run was kept.
statusstringcurrent: nothing that can move its numbers changed. superseded: something did. no_manifest: the run was kept before manifests.
One of: current, superseded, no_manifest (open: new values may be added).
engine_changedbooleannullableWhether the tester's revision has risen since.
changedarray of objectThe parts that differ, each with its value then (from) and now (to).
partstringengine, release, formula_language, catalogue, attributes, settlement_rules or queries.
tested_atstringWhen the run was kept (RFC 3339); only on a run read back.
format date-time
rule_versionobjectnullableThe saved rule this test was kept under (save, or rule_id): its rule_id, name, version and versions_tried. Null when the test wasn't saved.
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerrequiredThis rule's version under it: the same rule again keeps its version.
min 1
new_versionbooleanThis test made a new version.
versions_triedintegerrequiredVersions saved under it, this one included.
min 1
since_savedobjectnullableThe saved version's record since it was saved (backtester T6, F50), when the test names its rule_id and repeats that version: verdict v2's since_saved check reads it. Null for a new version (its clock starts now) or an unsaved test.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
tests_on_accountobjectnullableRules this account has tried on this competition and market, this one included (the rules-tried ledger: every distinct rule tested on the site, the API or Kelly MCP, kept for good). The not-luck check adjusts for the larger of this count and the request's own tests. Null when the call has no account.
competitionstringrequiredThe competition.
marketstringrequiredThe market.
countintegerrequiredDistinct rules tried.
min 1
shown_countedintegerVariants an earlier report showed (its condition-removed forms, nearby numbers and home or away splits) that joined the count because this rule is one of them (audit F6).
stakingstringHow each bet was staked.
matchedintegerBets taken.
ungradedintegerBets not yet played.
noticestringrequiredThis tests a rule on past games; it isn't a pick.
calibrationobjectnullableoutput: probability only. The formula's chances against the market's no-vig close of the same side (moneyline: the team wins; spread: it covers the bet's line), on the newest 30% of the scored team-games (whole days), and the bets it makes at the edge threshold.
formulastringThe probability formula as sent.
bet_formulastringThe rule it bets: the chance at the bet's own price worth more than the edge threshold (the rule tested, counted and kept).
edge_thresholdnumberExpected return a bet needs, p × decimal price − 1 (0.02 by default).
basisstringnewest_30: scored on the holdout only.
One of: newest_30 (open: new values may be added).
gamesintegerGames scored.
rowsintegerTeam-games scored (both teams of a game).
in_sample_rowsintegerTeam-games before the holdout, not scored.
holdout_fromstringnullableThe holdout's first day.
left_outobjectTeam-games not scored.
pushesintegerPushes and half results.
outside_0_1integerValues below 0 or above 1.
no_valueintegerRows the formula gave no number for.
clippedintegerValues of 0 or 1 clipped to 1e-6 from the edge.
briernumbernullableBrier score of the formula.
brier_marketnumbernullableBrier score of the close.
brier_skillnumbernullable1 − formula / close (above 0: the formula scores better).
log_lossnumbernullableLog loss of the formula.
log_loss_marketnumbernullableLog loss of the close.
log_loss_skillnumbernullable1 − formula / close.
dm_p_valuenumbernullableDiebold–Mariano p on the per-game log-loss difference (two-sided).
dm_statisticnumbernullableThe HLN-corrected statistic (positive: the formula's loss is higher).
mcbnumbernullableCORP miscalibration (Brier).
dscnumbernullableCORP discrimination.
uncnumbernullableUncertainty; brier = mcb − dsc + unc.
mcb_p_valuenumbernullableThe MCB test: the chance a calibrated formula looks this miscalibrated (999 resamples).
reliabilityarray of objectnullableThe CORP reliability curve at up to 20 forecast values.
pnumbernullableThe formula's probability.
observednumbernullableHow often it happened (PAV fit).
lonumbernullable90% consistency band, low.
hinumbernullable90% consistency band, high.
ecenumbernullableExpected calibration error over ten equal-width bins (bin-dependent; for familiarity).
distinct_forecastsintegerDistinct values the formula gave.
slopenumbernullableCalibration slope b in y ~ a + b·logit p (1 is ideal; under 1 overconfident).
slope_senumbernullableIts game-clustered standard error.
interceptnumbernullableCalibration intercept a (0 is ideal).
encompassingobjectnullabley ~ a + b_q·logit q + b_p·logit p, game-clustered: does the formula know something the close doesn't?
b_modelnumbernullableb_p.
b_marketnumbernullableb_q.
se_modelnumbernullableb_p's standard error.
p_valuenumbernullableTwo-sided p of b_p.
edge_binsarray of objectnullableRealized results against the formula's edge.
edgenumbernullableMean of p − q in the bin.
realized_minus_marketnumbernullableMean of y − q in the bin.
rowsintegerTeam-games in the bin.
statusstringThe calibrated check: pass, fail or na (under 200 games, or one value).
One of: pass, fail, na (open: new values may be added).
valuestringThe check's line.
statisticnumbernullableThe encompassing p.
thresholdnumbernullable0.05.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/runs/run_mkytv75qe5a297z7y3af
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_test",
"clv": {
"ci95": null,
"mean": null,
"note": "Bets at the close have no closing-line value by definition.",
"basis": "probability",
"devig": "multiplicative",
"p_value": null,
"coverage": 0,
"beat_rate": null,
"mean_prob": null,
"reference": "consensus",
"mean_points": null,
"points_bets": 0,
"points_test": null,
"sign_test_p": null,
"closes_stale": 0,
"profit_slope": null,
"significance": null,
"sample_needed": null,
"beat_rate_ci95": null,
"bets_with_close": 0,
"lines_converted": 0,
"points_coverage": 0,
"profit_slope_se": null,
"beat_close_share": null,
"sample_needed_roi": null,
"bets_without_close": 1
},
"base": {
"why": "the rule has no who-or-when condition to filter, so its base is the whole market",
"source": "market",
"status": "na",
"formula": null
},
"caps": null,
"luck": {
"bets": 0,
"devig": "multiplicative",
"theta": null,
"reading": "versus_the_market",
"left_out": {
"half": 0,
"push": 0
},
"by_season": [],
"actual_roi": null,
"luck_units": null,
"actual_wins": null,
"actual_units": null,
"close_source": "consensus",
"expected_roi": null,
"expected_wins": null,
"luck_se_units": null,
"expected_units": null,
"lines_converted": 0,
"luck_percentile": null,
"bets_without_prob": 1
},
"path": {
"series": {
"x": "bet_index",
"dates": [
"2025-10-21"
],
"index": [
1
],
"units": [
0.91
],
"markers": [
{
"at": 1,
"type": "season",
"label": "2025-26"
}
],
"drawdown": [
0
],
"band_p025": null,
"band_p975": null,
"base_units": null,
"rolling_p025": [
null
],
"rolling_p975": [
null
],
"rolling_roi_100": [
null
]
},
"downsampled": false,
"longest_losing_run": 0,
"max_drawdown_units": 0,
"expected_max_drawdown_zero_edge": null
},
"risk": {
"seed": 6899137762779026,
"curve": [],
"paths": 5000,
"method": "lundberg+posterior_monte_carlo",
"status": "too_few",
"percent": {
"status": "too_few",
"drawdown_p95": null,
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null
},
"staking": "flat",
"warnings": [
"too_few_decisions"
],
"ruin_share": 0.5,
"drawdown_p95": null,
"horizon_bets": 500,
"lose_all_ever": 0,
"bankroll_units": 100,
"stake_fraction": null,
"down_after_horizon": null,
"drawdown_p95_units": null,
"estimation_premium": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"ruin_ever_if_rate_exact": 0,
"bankroll_for_ruin_005_units": null,
"lose_all_ever_allowing_error": 0.277131008418
},
"rule": {
"side": "rule",
"price": "consensus",
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"stake": "risk",
"market": "spread",
"formula": "=home",
"seasons": [
"2025-26"
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"detail": "full"
},
"as_of": "2026-11-06T00:00:00Z",
"decay": {
"bets": 1,
"break": null,
"cusum": null,
"window": 50,
"rolling": [],
"clv_break": null,
"decay_ratio": null,
"known_break": null,
"roi_newest_30": null,
"status_reason": "fewer than (100) graded bets",
"trend_per_100_bets": null
},
"noise": {
"seed": 3397954809,
"delay": {
"na": "day-before lines for this league are still loading",
"bet_moment": "day_before"
},
"lines": {
"runs": 200,
"roi_p05": 0.9099999999999999,
"line_prob": 0.5,
"roi_median": 0.9099999999999999,
"profitable_runs": 200,
"profitable_share": 1
},
"inputs": null,
"inputs_na": "the rule compares no numbers",
"low_power": true
},
"stale": false,
"honest": {
"roi": -0.04514385341429974,
"ci80": [
-0.07077022186419332,
-0.019517484964406154
],
"prior": {
"mean": -0.04545450206611379,
"words": "shrunk toward the market's usual margin of -4.5%",
"capped": false,
"closes": 0,
"source": "margin"
},
"tries": 1,
"shrink": 0.999560632688928,
"raw_roi": 0.9099999999999999,
"prior_sd": 0.02,
"uses_clv": false,
"prior_mean": -0.04545450206611379,
"basis_rules": 0,
"prior_basis": "seed",
"decay_factor": 0.74,
"posterior_sd": 0.019995605844174143,
"selection_adj": 0,
"stabilizes_at_bets": 2275
},
"moment": {
"target": "the start",
"rows_aged": 0,
"bet_moment": "close",
"age_unknown": 1,
"rows_priced": 1,
"stale_price": 0,
"after_moment": 0,
"line_sources": {
"sheet_line": 1
},
"max_stale_minutes": 60,
"max_snapshot_age_minutes": null,
"median_snapshot_age_minutes": null,
"floor_minutes_before_oct_2022": 10
},
"nearby": {
"na": "the rule compares no numbers",
"grid": null,
"numbers": [],
"reading": null,
"roi_p05": null,
"roi_p95": null,
"variants": 0,
"roi_median": null,
"spike_ratio": null,
"plateau_share": null,
"distinct_bet_sets": 0
},
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"run_id": "run_mkytv75qe5a297z7y3af",
"splits": [
{
"roi": 0.9099999999999999,
"bets": 1,
"label": "Home",
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1,
"p_adjusted": 1,
"shown_rows": 2
},
{
"roi": 0.9099999999999999,
"bets": 1,
"label": "Favorite",
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1,
"p_adjusted": 1,
"shown_rows": 2
}
],
"stress": {
"grid": [
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": 0
},
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": -0.5
},
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": -1
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": 0
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": -0.5
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": -1
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": 0
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": -0.5
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": -1
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": 0
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": -0.5
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": -1
}
],
"flat_110": {
"roi": 0.9090909090909092,
"bets": 1,
"units": 0.9090909090909092,
"record": "1-0",
"p_value": 0.5238095238095238,
"decisions": 1
},
"break_even": {
"decimal_avg": null,
"line_points": 3,
"american_avg": null,
"price_multiplier": 0
},
"other_side": {
"roi": -1,
"bets": 1,
"units": -1,
"record": "0-1",
"p_value": 1,
"decisions": 1
},
"slippage_mc": null,
"half_point_worse": {
"roi": 0.9099999999999999,
"bets": 1,
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1
}
},
"holdout": null,
"matched": 1,
"overall": {
"roi": 0.9099999999999999,
"bets": 1,
"wins": 1,
"units": 0.9099999999999999,
"losses": 0,
"pushes": 0,
"record": "1-0",
"roi_ci": {
"bca": null,
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"method": "score",
"iid_score": [
-0.6054908095394765,
0.9099999999999999
],
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"block_bootstrap": null
},
"deflated": {
"dsr": null,
"psr": null,
"skew": null,
"roi_bar": null,
"kurtosis": null,
"sample_needed": null,
"sharpe_per_bet": null,
"trials_effective": 1,
"sample_needed_plan": null,
"sharpe_expected_max": null,
"sample_needed_single": null,
"sample_needed_power80": null
},
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F36.2@1"
],
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"win_rate": 1,
"decisions": 1,
"break_even": 0.5235602094240838,
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"significance": {
"z": 0.9539392014169455,
"null": "break_even",
"seed": null,
"sims": null,
"alpha": 0.05,
"spa_p": null,
"tests": 1,
"method": "exact_binomial",
"t_stat": null,
"haircut": null,
"p_value": 0.5235602094240838,
"clears_t3": false,
"p_adjusted": 0.5235602094240838,
"haircut_roi": null,
"significant": false,
"adjust_method": "none",
"anytime_valid": {
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"crossed": false,
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"roi_cs95": [
-0.999045,
0.9090450000000001
],
"kelly_fraction": 0.5
},
"p_value_exact": 0.5235602094240838,
"p_value_vs_vig": null,
"reality_check_p": null,
"tests_effective": 1,
"p_value_clustered": 0.17005720935900537,
"coin_match_per_100": 50,
"multiple_comparisons": null,
"no_skill_match_per_100": 52.35602094240838
},
"small_sample": true,
"few_decisions": true,
"sample_needed": null,
"over_break_even": 0.47643979057591623
},
"seasons": [
{
"roi": 0.9099999999999999,
"bets": 1,
"wins": 1,
"units": 0.9099999999999999,
"losses": 0,
"pushes": 0,
"record": "1-0",
"season": "2025-26",
"clv_mean": null,
"win_rate": 1,
"decisions": 1,
"break_even": 0.5235602094240838,
"roi_interval": [
0.9099999999999999,
0.9099999999999999
],
"significance": {
"z": 0.9539392014169455,
"alpha": 0.05,
"p_value": 0.5235602094240838,
"significant": false,
"anytime_valid": {
"crossed": false,
"e_value": 1.0500500000000001,
"p_value": 0.9523356030665205
},
"coin_match_per_100": 50,
"multiple_comparisons": null,
"no_skill_match_per_100": 52.35602094240838
},
"small_sample": true,
"few_decisions": true,
"sample_needed": null,
"over_break_even": 0.47643979057591623
}
],
"staking": "flat, 1 unit risked a bet",
"verdict": {
"gate": {
"key": "no_look_ahead",
"value": "every value the rule reads is known at the bet moment",
"status": "pass",
"leak_test": "clean"
},
"level": "too_few",
"checks": [
{
"key": "enough_bets",
"zone": "red",
"group": "real",
"label": "Enough bets",
"value": "(1) decisions are too few to size the sample",
"status": "fail",
"statistic": 1,
"threshold": null
},
{
"key": "not_luck",
"zone": "red",
"group": "real",
"label": "Not just luck",
"value": "p 0.524",
"status": "fail",
"statistic": 0.5235602094240838,
"threshold": 0.05
},
{
"key": "beats_base",
"zone": "na",
"group": "real",
"label": "Beats its own base",
"value": "the rule has no who-or-when condition to filter, so its base is the whole market",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "half_point_worse",
"zone": "na",
"group": "pressure",
"label": "Holds ½ point worse",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"zone": "na",
"group": "pressure",
"label": "Nearby thresholds hold",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"zone": "na",
"group": "pressure",
"label": "Not one lucky stretch",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "beats_close",
"zone": "na",
"group": "still_working",
"label": "Beats the close",
"value": "bets at the close can't beat it",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"zone": "na",
"group": "still_working",
"label": "Still working",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "since_saved",
"zone": "na",
"group": "still_working",
"label": "Since saved",
"value": "(0) of (30) bets since it was saved",
"status": "na",
"statistic": null,
"threshold": null
}
],
"passed": 0,
"version": 2,
"headline": "Too few bets to judge",
"applicable": 2,
"held_back_by": []
},
"warm_up": {
"policy": "as_written",
"warning": false,
"windows": [],
"by_policy": null,
"short_bets": 0,
"games_needed": 0
},
"without": [],
"ablation": null,
"capacity": {
"books": [],
"market": "spreads",
"reason": "Kelly holds no measured spreads limits for nba at this moment yet.",
"available": false,
"limits_held": 0,
"before_start": "0_1h",
"bets_per_season": null,
"max_stake_per_bet": null,
"per_season_at_limit": null
},
"coverage": {
"flow": [
{
"rows": 6,
"step": "in_scope",
"games": 3
},
{
"rows": 3,
"step": "rule_true"
},
{
"out": {
"both_sides": 0,
"same_game_total": 0
},
"rows": 3,
"step": "one_bet_a_game"
},
{
"out": {
"no_line": 2
},
"rows": 1,
"step": "priced"
},
{
"out": {
"void": 0,
"not_played_yet": 0
},
"rows": 1,
"step": "graded"
},
{
"out": {
"push": 0
},
"rows": 1,
"step": "decisions"
}
],
"by_season": [
{
"bets": 1,
"rows": 6,
"games": 3,
"season": "2025-26",
"rule_true": 3,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"rows": 6,
"share": 0,
"attribute": "home",
"rows_blank": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"flag": false,
"kept": 1,
"max_smd": null,
"left_out": 2,
"covariate": null,
"by_covariate": []
}
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"manifest": {
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"settlement_rules": "2026-11-08",
"attributes_sha256": "0000000000000000",
"catalogue_version": "1:0"
},
"data": {
"as_of": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
},
"version": "snap_52",
"snapshots": [
"snap_52"
],
"data_through": {},
"one_snapshot": true,
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z"
},
"sha256": "sha256:d78d99c7954d4b2367b7baa69328756b5aa4464bb52ddcb24cdf05a0484557ad",
"request": {
"seed": 18147943352127158000,
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"request_hash": "sha256:13e69ec9c84eb683918ee67e91ae21ac69cc813cfbc4758f625524596be5d3da"
},
"seasons": [
{
"rows": 6,
"season": "2025-26",
"priced_share": 0.3333
}
],
"sources": {
"inputs": [
{
"name": "home",
"known": "pregame"
}
],
"prices": {
"sheet_line": 1
}
},
"version": 1,
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
}
},
"ungraded": 0,
"leak_test": {
"parts": [],
"status": "clean",
"rows_checked": 6,
"matched_checked": 3,
"matches_flipped": 0
},
"dependence": {
"se": null,
"slate": "date",
"method": "none",
"se_iid": null,
"clusters": 1,
"cluster_by": [
"slate",
"team_season"
],
"design_effect": 1,
"effective_bets": 1
},
"robustness": {
"off": [],
"seed": 769905824,
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
}
],
"fast_path": true,
"evaluations": 1
},
"settlement": {
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
],
"applied": [
{
"bets": 1,
"rule": "final_score"
}
],
"profile": "kelly_standard",
"rules_version": "2026-11-08"
},
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"monte_carlo": null,
"since_saved": null,
"data_through": {},
"data_version": "snap_52",
"rule_version": null,
"concentration": {
"by": [
{
"best": null,
"group": "season",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "team",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "opponent",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "month",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "weekday",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
}
],
"bets": 1,
"units": 0.9099999999999999,
"trim_top_0_05": null
},
"latest_result": "2025-10-25T23:00:00Z",
"risk_adjusted": {
"sqn": null,
"bets": 1,
"warnings": [
"no_losing_bets",
"no_spread_in_returns",
"small_sample"
],
"ulcer_index": 0,
"martin_ratio": null,
"sharpe_daily": null,
"avg_win_units": 0.9099999999999999,
"profit_factor": null,
"avg_loss_units": null,
"sharpe_per_bet": null,
"recovery_factor": null,
"sortino_per_bet": null,
"sharpe_per_season": null,
"max_drawdown_units": 0,
"profit_factor_ci95": null,
"return_on_drawdown": null,
"sharpe_per_bet_ci95": null,
"probabilistic_sharpe": null,
"sortino_per_bet_ci95": null,
"current_drawdown_units": 0,
"expected_units_per_bet": 0.9099999999999999,
"longest_underwater_bets": 0,
"longest_underwater_days": 0
},
"rows_built_at": "2026-11-06T00:00:00Z",
"staking_plans": {
"kelly": {
"basis": "record",
"capped": null,
"shrink": 1,
"win_rate": 1,
"withheld": "stake_first_party_only",
"push_rate": 0,
"multiplier": null,
"simultaneous": false,
"zero_because": "too_few",
"full_fraction": null,
"honest_fraction": null,
"multiplier_rule": "honest"
},
"plans": [
{
"roi": 0.9099999999999966,
"bets": 1,
"plan": "flat",
"units": 0.9099999999999966,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.009058844488345926,
"stopped_at_bet": null,
"end_bankroll_units": 100.91
},
{
"roi": 0.9099999999999999,
"bets": 1,
"plan": "to_win",
"units": 1,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.009950330853168092,
"stopped_at_bet": null,
"end_bankroll_units": 101
},
{
"roi": 0.9099999999999966,
"bets": 1,
"plan": "percent",
"units": 1.8199999999999932,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.018036362486098575,
"stake_fraction": 0.02,
"stopped_at_bet": null,
"end_bankroll_units": 101.82
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"percent": 0.02,
"warnings": [
"small_sample"
],
"kelly_cap": 0.05,
"bankroll_units": 100
},
"random_baseline": null,
"split_dimensions": [],
"tests_on_account": {
"count": 1,
"market": "spread",
"competition": "nba",
"shown_counted": 0
},
"matched_both_sides": 0,
"matched_without_line": 2,
"tested_at": "2026-10-04T17:30:00Z",
"since_run": {
"status": "current",
"engine_changed": false,
"changed": []
},
"summary": "Too few bets to judge: honest forward ROI -4.5%, 1-0 against the spread, 2025-26, +0.9 units, +91.0% ROI past (0 of 2 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_5bc2152f1f186610ad0d7ecd",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/runs/run_zzzzzzzzzzzzzzzzzzzz
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No run with that ID on this account.",
"param": "run_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_0e159e15ddc94f8dc2cb4f43",
"retryable": false,
"display_message": "That could not be found."
}GET /v1/rules/{rule_id}/versions
public, no key needed · 0 credits, not charged during the beta · MCP tool list_rule_versions · status live
A rule you saved (test_rule with save or rule_id) and every version tested under it, oldest first: each formula and its options, and its latest run's record, units, ROI, p-value and verdict. Versions are never deleted, so versions_tried is honest.
rule_idstring · pathrequiredA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
200 The rule versions facts.
objectconstrequiredAlways rule_versions.
Always "rule_versions".
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringrequiredIts name.
competitionstringThe competition every version is tested on.
marketstringThe market every version is tested on.
created_atstringWhen it was first saved (RFC 3339).
format date-time
versions_triedintegerrequiredVersions saved under it, hidden ones included.
min 1
versionsarray of objectrequiredEvery version, oldest first.
versionintegerrequired1, 2, ... in the order they were saved.
min 1
formulastringrequiredThe formula as typed.
rule_hashstringrequiredThe rule's identity (see rule.rule_hash).
pattern ^sha256:[0-9a-f]{64}$
paramsobjectThe options it was tested with: competition, market, side, bet moment, seasons, season type, stake, price.
created_atstringrequiredWhen this version was saved (RFC 3339).
format date-time
run_idstringnullableIts latest kept run.
data_versionstringnullableThe data that run read.
betsintegernullableBets taken.
decisionsintegernullableWins plus losses.
recordstringnullableW-L-P.
unitsnumbernullableUnits won or lost.
roinumbernullableReturn on units risked, a fraction.
p_valuenumbernullableThe record's own p-value.
levelstringnullableThe verdict's level.
headlinestringnullableThe verdict's headline.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2/versions
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_versions",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"competition": "nba",
"market": "spread",
"created_at": "2026-10-04T17:30:00Z",
"versions_tried": 2,
"versions": [
{
"version": 1,
"formula": "=home",
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"created_at": "2026-10-04T17:30:00Z",
"hidden": false,
"run_id": "run_rc69h189x95wy5rhfet2",
"data_version": "snap_52",
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"level": "too_few",
"headline": "Too few bets to judge",
"params": {
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"robust": {
"off": [],
"line_prob": 0.5,
"skip_prob": 0,
"break_date": null,
"noise_runs": 200,
"input_scale": 0.1,
"trim_winners": 0,
"monte_carlo_mode": "block",
"monte_carlo_sims": null,
"random_baseline_sims": 2000
},
"sizing": {
"caps": {
"priority": "earliest",
"max_open_stake": null,
"max_bets_per_day": null,
"max_bets_per_game": null,
"daily_stop_loss_units": null,
"max_bets_per_team_week": null
},
"paths": 5000,
"percent": 0.02,
"kelly_cap": 0.05,
"ruin_share": 0.5,
"horizon_bets": 500,
"staking_plans": [
"flat",
"to_win",
"percent",
"kelly"
],
"bankroll_units": 100,
"kelly_fraction": null
},
"teaser": null,
"seasons": [
"2025-26"
],
"warm_up": "as_written",
"ci_level": 0.95,
"plan_roi": null,
"stake_ok": false,
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null
}
},
{
"version": 2,
"formula": "=AND(home, rest_days >= 1)",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"created_at": "2026-10-04T17:30:00Z",
"hidden": false,
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"data_version": "snap_52",
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"level": "too_few",
"headline": "No bets",
"params": {
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"robust": {
"off": [],
"line_prob": 0.5,
"skip_prob": 0,
"break_date": null,
"noise_runs": 200,
"input_scale": 0.1,
"trim_winners": 0,
"monte_carlo_mode": "block",
"monte_carlo_sims": null,
"random_baseline_sims": 2000
},
"sizing": {
"caps": {
"priority": "earliest",
"max_open_stake": null,
"max_bets_per_day": null,
"max_bets_per_game": null,
"daily_stop_loss_units": null,
"max_bets_per_team_week": null
},
"paths": 5000,
"percent": 0.02,
"kelly_cap": 0.05,
"ruin_share": 0.5,
"horizon_bets": 500,
"staking_plans": [
"flat",
"to_win",
"percent",
"kelly"
],
"bankroll_units": 100,
"kelly_fraction": null
},
"teaser": null,
"seasons": [
"2025-26"
],
"warm_up": "as_written",
"ci_level": 0.95,
"plan_roi": null,
"stake_ok": false,
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null
}
}
],
"summary": "Home teams: (2) versions tried.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_9d856ff26b3c4955513933ee",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/rl_zzzzzzzzzzzzzzzzzzzz/versions
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No saved rule with that rule_id on this account.",
"param": "rule_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_56678091b4df5f34b027177f",
"retryable": false,
"display_message": "That could not be found."
}GET /v1/rules/{rule_id}/diff
public, no key needed · 0 credits, not charged during the beta · MCP tool compare_rule_versions · status live
What changed between two versions of a rule you saved, bet by bet: the games the newer version bets that the older didn't and the reverse (by key, with result and units), and the change in units and ROI, each on its latest kept run.
rule_idstring · pathrequiredA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
frominteger · queryrequiredThe older version's number.
min 1
tointeger · queryrequiredThe newer version's number.
min 1
200 The rule diff facts.
objectconstrequiredAlways rule_diff.
Always "rule_diff".
rule_idstringrequiredThe saved rule.
namestringIts name.
fromobjectrequiredOne version of a saved rule, with its latest kept run's numbers (null when it has none).
versionintegerrequired1, 2, ... in the order they were saved.
min 1
formulastringrequiredThe formula as typed.
rule_hashstringrequiredThe rule's identity (see rule.rule_hash).
pattern ^sha256:[0-9a-f]{64}$
paramsobjectThe options it was tested with: competition, market, side, bet moment, seasons, season type, stake, price.
created_atstringrequiredWhen this version was saved (RFC 3339).
format date-time
run_idstringnullableIts latest kept run.
data_versionstringnullableThe data that run read.
betsintegernullableBets taken.
decisionsintegernullableWins plus losses.
recordstringnullableW-L-P.
unitsnumbernullableUnits won or lost.
roinumbernullableReturn on units risked, a fraction.
p_valuenumbernullableThe record's own p-value.
levelstringnullableThe verdict's level.
headlinestringnullableThe verdict's headline.
toobjectrequiredOne version of a saved rule, with its latest kept run's numbers (null when it has none).
versionintegerrequired1, 2, ... in the order they were saved.
min 1
formulastringrequiredThe formula as typed.
rule_hashstringrequiredThe rule's identity (see rule.rule_hash).
pattern ^sha256:[0-9a-f]{64}$
paramsobjectThe options it was tested with: competition, market, side, bet moment, seasons, season type, stake, price.
created_atstringrequiredWhen this version was saved (RFC 3339).
format date-time
run_idstringnullableIts latest kept run.
data_versionstringnullableThe data that run read.
betsintegernullableBets taken.
decisionsintegernullableWins plus losses.
recordstringnullableW-L-P.
unitsnumbernullableUnits won or lost.
roinumbernullableReturn on units risked, a fraction.
p_valuenumbernullableThe record's own p-value.
levelstringnullableThe verdict's level.
headlinestringnullableThe verdict's headline.
bets_addedintegerrequiredBets the newer version takes that the older didn't.
bets_removedintegerrequiredBets the older version took that the newer doesn't.
bets_keptintegerBets both take.
units_addednumberUnits the added bets won or lost.
units_removednumberUnits the removed bets won or lost.
units_deltanumbernullableThe newer version's units minus the older's.
roi_deltanumbernullableThe newer version's ROI minus the older's, a fraction.
same_databooleanBoth runs read the same data_version.
addedarray of objectrequiredEvery added bet, by key.
keystring{game_id}:{team_id}.
resultstringnullablewin, loss, push, or null while ungraded.
unitsnumbernullableUnits it won or lost.
removedarray of objectrequiredEvery removed bet, by key.
keystring{game_id}:{team_id}.
resultstringnullablewin, loss, push, or null while ungraded.
unitsnumbernullableUnits it won or lost.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2/diff?from=1&to=2
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_diff",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"from": {
"version": 1,
"formula": "=home",
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"created_at": "2026-10-04T17:30:00Z",
"hidden": false,
"run_id": "run_rc69h189x95wy5rhfet2",
"data_version": "snap_52",
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"level": "too_few",
"headline": "Too few bets to judge",
"params": {
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"robust": {
"off": [],
"line_prob": 0.5,
"skip_prob": 0,
"break_date": null,
"noise_runs": 200,
"input_scale": 0.1,
"trim_winners": 0,
"monte_carlo_mode": "block",
"monte_carlo_sims": null,
"random_baseline_sims": 2000
},
"sizing": {
"caps": {
"priority": "earliest",
"max_open_stake": null,
"max_bets_per_day": null,
"max_bets_per_game": null,
"daily_stop_loss_units": null,
"max_bets_per_team_week": null
},
"paths": 5000,
"percent": 0.02,
"kelly_cap": 0.05,
"ruin_share": 0.5,
"horizon_bets": 500,
"staking_plans": [
"flat",
"to_win",
"percent",
"kelly"
],
"bankroll_units": 100,
"kelly_fraction": null
},
"teaser": null,
"seasons": [
"2025-26"
],
"warm_up": "as_written",
"ci_level": 0.95,
"plan_roi": null,
"stake_ok": false,
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null
}
},
"to": {
"version": 2,
"formula": "=AND(home, rest_days >= 1)",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"created_at": "2026-10-04T17:30:00Z",
"hidden": false,
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"data_version": "snap_52",
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"level": "too_few",
"headline": "No bets",
"params": {
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"robust": {
"off": [],
"line_prob": 0.5,
"skip_prob": 0,
"break_date": null,
"noise_runs": 200,
"input_scale": 0.1,
"trim_winners": 0,
"monte_carlo_mode": "block",
"monte_carlo_sims": null,
"random_baseline_sims": 2000
},
"sizing": {
"caps": {
"priority": "earliest",
"max_open_stake": null,
"max_bets_per_day": null,
"max_bets_per_game": null,
"daily_stop_loss_units": null,
"max_bets_per_team_week": null
},
"paths": 5000,
"percent": 0.02,
"kelly_cap": 0.05,
"ruin_share": 0.5,
"horizon_bets": 500,
"staking_plans": [
"flat",
"to_win",
"percent",
"kelly"
],
"bankroll_units": 100,
"kelly_fraction": null
},
"teaser": null,
"seasons": [
"2025-26"
],
"warm_up": "as_written",
"ci_level": 0.95,
"plan_roi": null,
"stake_ok": false,
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null
}
},
"bets_added": 0,
"bets_removed": 1,
"bets_kept": 0,
"units_added": 0,
"units_removed": 0.91,
"units_delta": -0.91,
"roi_delta": null,
"same_data": true,
"added": [],
"removed": [
{
"key": "gm_1hnx5nrmk4xjpz6fzyfz:tm_b4w5y9xy5f9sgg1cmkvy",
"result": "win",
"units": 0.91
}
],
"summary": "Home teams, version 1 to 2: (0) bets added, (1) removed, -0.9 units.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_44e0850b8956a358966980cf",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2/diff?from=1&to=9
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "This rule has no version 9.",
"param": "to",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_fec3da3f6b4d35a869ac2e8a",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules:translate
public, no key needed · 0 credits, not charged during the beta · MCP tool translate_rule · status live
A betting rule in the person's words written as a rule formula (backtester T8, F43): candidate formulas from Kelly's model, each checked by the formula engine and the rule tester's refusals (look-ahead, Kelly's edge), grouped by the games they would bet; the largest group's formula with a plain-English readback written by a fixed template, and one question when the words read two ways. Words about the game's own outcome, data Kelly doesn't hold or stake sizes come back as problems. Nothing runs: test the formula with POST /v1/rules:test (translation_id). Free, and never counted toward rules tried.
Body, application/json (required):
textstringThe rule in the person's words (or a formula starting with =, read back as is), up to 1,000 characters.
min length 1 · max length 1000
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
marketstringspread, moneyline, over, under or draw; left out, the market the words name (spread when they name none).
One of: spread, moneyline, over, under, draw (open: new values may be added).
sidestringrule or fade; left out, what the words say.
One of: rule, fade (open: new values may be added).
bet_momentstringclose (default) or edition: the moment the formula is checked against.
One of: close, edition (open: new values may be added).
seasonsarray of stringSeasons to compare the candidate formulas on (default: the latest two).
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The rule translation facts.
objectstringrequiredrule_translation.
One of: rule_translation (open: new values may be added).
textstringThe words, as sent.
competitionstringThe competition.
marketstringThe market (the request's, or the one the words name; spread by default).
sidestringrule, or fade when the words bet against the teams described.
bet_momentstringThe bet moment the formula was checked at.
formulastringrequirednullableThe chosen formula; null when the words can't be a rule.
readbackstringrequirednullableThe formula in plain English, written by template.
candidatesintegerFormulas the model wrote.
validintegerThose that passed Kelly's checks.
agreementnumbernullableShare of the valid candidates that bet the same games as the chosen one.
questionstringrequirednullableOne question when the words read two ways.
choicesarray of objectThe readings the question offers.
labelstringIn words.
formulastringIts formula.
problemsarray of objectrequiredWhy there is no formula, when there isn't.
codestringmissing_data, look_ahead, stake_rule, team_named, not_a_rule, not_understood or the rule tester's own refusal code.
pathstringnullableThe field or data it names.
messagestringWhat to change.
fixesarray of objectFixes, when there are any.
translation_idstringrequiredPass it to test_rule to keep the words with the test.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
prompt_versionstringThe translator's prompt version.
seasons_checkedarray of stringSeasons the candidates were compared on.
methodstringai (words) or formula (a formula sent as is, read back without the model).
One of: ai, formula (open: new values may be added).
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large
POST https://api.askkelly.ai/v1/rules:translate
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "=AND(home, rest_days >= 2)",
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"object": "rule_translation",
"text": "=AND(home, rest_days >= 2)",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"formula": "=AND(home, rest_days >= 2)",
"readback": "Bets the team against the spread when it is at home, and its rest in days is at least 2.",
"candidates": 1,
"valid": 1,
"agreement": 1,
"question": null,
"choices": [],
"problems": [],
"translation_id": "tx_xbj639zqrcq2kg653rsf",
"prompt_version": "t8-nl-1",
"seasons_checked": [],
"method": "formula",
"summary": "Bets the team against the spread when it is at home, and its rest in days is at least 2. Formula: =AND(home, rest_days >= 2). Nothing has run yet; test it to see its record.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_65c75377c603b7dea913e749",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:translate
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "",
"competition": "nba"
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "text: '' should be non-empty",
"param": "text",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_e81de766467deb1e6e85c93c",
"retryable": false,
"display_message": "Something in that request was not understood."
}GET /v1/rules/{rule_id}/record
public, no key needed · 0 credits, not charged during the beta · MCP tool get_rule_record · status live
A rule you saved, from the moment it was saved: each bet taken at its bet moment after the save, at the line Kelly held then, graded when final, beside its backtest. Record, units, ROI, closing-line value, an anytime-valid e-value and whether live is consistent with the backtest. Never reset: an edit is a new version with its own record. No prices are returned.
rule_idstring · pathrequiredA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
versioninteger · queryThe version to read (default the newest).
min 1
200 The rule record facts.
objectconstrequiredAlways rule_record.
Always "rule_record".
rule_idstringrequiredThe saved rule.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
competitionstringIts competition.
marketstringIts market.
versionintegerrequiredThe version read.
min 1
versions_triedintegerVersions saved under the rule.
min 1
formulastringThe version's formula.
bet_momentstringedition or close.
since_savedobjectrequiredThe record since the save.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
backtestobjectThe version's latest kept backtest run.
run_idstringnullableThe run.
recordstringnullableIts record.
decisionsintegernullableWins plus losses.
unitsnumbernullableUnits.
roinumbernullableROI, a fraction.
win_ratenumbernullableWin rate.
seasonsarray of stringnullableSeasons tested.
registrationobjectnullableThe version's public registration, if any.
registration_idstringreg_….
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
statusstringIts status.
registered_atstringWhen (RFC 3339).
format date-time
betsarray of objectrequiredThe forward bets, newest first (up to 200).
keystring{game_id}:{team_id}.
labelstringThe game, from the side bet.
bet_momentstringedition or close.
moment_atstringWhen the bet was taken (RFC 3339).
format date-time
linenumbernullableThe line at the bet moment.
statusstringpending, graded, void or no_line.
One of: pending, graded, void, no_line (open: new values may be added).
resultstringnullablewin, loss, push, half_win or half_loss once graded.
unitsnumbernullableUnits won or lost.
clvnumbernullableClosing-line value, a fraction.
noticestringIt isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2/record
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_record",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"competition": "nba",
"market": "spread",
"version": 2,
"versions_tried": 2,
"formula": "=AND(home, rest_days >= 1)",
"bet_moment": "close",
"since_saved": {
"saved_at": "2026-10-04T17:30:00Z",
"record": "0-0",
"wins": 0,
"losses": 0,
"pushes": 0,
"decisions": 0,
"bets": 0,
"units": 0,
"roi": null,
"clv_mean": null,
"clv_bets": 0,
"e_value": null,
"max_e_value": null,
"e_threshold": 20,
"vs_backtest": null,
"pending": 0,
"no_line": 0,
"void": 0,
"status": "na",
"min_decisions": 30,
"last_bet_at": null
},
"backtest": {
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"record": "0-0",
"decisions": 0,
"units": 0,
"roi": null,
"win_rate": null,
"seasons": [
"2025-26"
]
},
"registration": null,
"bets": [],
"notice": "A saved rule's record from the moment it was saved, at the line Kelly held at its bet moment. It isn't a pick.",
"summary": "Home teams (version 2) since it was saved: 0-0, +0.0 units, no ROI yet (0) decisions, (0) pending.",
"assumptions": [
"A saved rule's record from the moment it was saved, at the line Kelly held at its bet moment. It isn't a pick."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_b85bb844fd56266d471b4716",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/rl_zzzzzzzzzzzzzzzzzzzz/record
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No saved rule with that rule_id on this account.",
"param": "rule_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_3853cadefa4e444ac772f889",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules/{rule_id}:register
public, no key needed · 0 credits, not charged during the beta · MCP tool register_rule · status live
Registers one of your saved versions in public with its pass bar set now: an e-value to reach (20 by default, alpha 0.05) or a fixed number of decisions. The frozen rule is hashed and the hash published; its record counts only bets after this moment, and it can never be edited, only withdrawn. A formula can stay hidden until embargo_until; the record is public from day one.
rule_idstring · pathrequiredA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
Body, application/json (required):
rule_idstringA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
versionintegerThe version to register (default the newest).
min 1
success_typestringThe pass bar: e_value (default; met when the forward e-value reaches the threshold) or fixed_n (a one-sided test at success_n decisions).
One of: e_value, fixed_n (open: new values may be added).
success_thresholdnumbere_value only: the e-value to reach, 20 (alpha 0.05) or more.
min 20 · max 10000
success_nintegerfixed_n only: forward decisions to judge it at (30 to 5000).
min 30 · max 5000
public_formulabooleanfalse hides the formula until embargo_until (the record is public anyway).
embargo_untilstringWith public_formula false: the date the formula is shown (YYYY-MM-DD, within a year).
format date
200 The rule registration facts.
objectconstrequiredAlways rule_registration.
Always "rule_registration".
registration_idstringrequiredreg_….
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
reg_hashstringrequiredsha256 of the canonical registration JSON (reg_json, keys sorted, no spaces): recompute it to check nothing changed.
pattern ^sha256:[0-9a-f]{64}$
registered_atstringrequiredWhen it was registered (RFC 3339); its record counts bets after this.
format date-time
statusstringrequiredrunning, met, failed or withdrawn.
One of: running, met, failed, withdrawn (open: new values may be added).
withdrawn_atstringnullableWhen the author withdrew it.
format date-time
embargo_untilstringnullableThe date the formula is shown.
format date
ruleobjectThe frozen rule: formula (null while hidden), competition, market, side, bet moment, price policy and stake.
successobjectrequiredThe pass bar set at registration: e_value (met when the forward e-value reaches threshold) or fixed_n (a one-sided test at n decisions).
typestringe_value or fixed_n.
One of: e_value, fixed_n (open: new values may be added).
thresholdnumbere_value: the e-value to reach (20 is alpha 0.05).
nintegerfixed_n: the forward decisions it is judged at.
alphanumberThe test's level.
progressobjectHow far it is toward the bar.
statisticnumbernullableThe running maximum e-value, or the fixed-n p-value once judged.
thresholdnumberThe bar.
progressnumber0 to 1.
backtestobjectnullableThe backtest it rests on: run, data version, tester revision, seasons, record, ROI.
author_rules_triedintegerRules the author had tried on this competition and market when registering (the rules-tried ledger).
min 1
author_registrationsobjectEvery registration the author has made, by status.
totalintegerAll of them.
runningintegerRunning.
metintegerMet.
failedintegerFailed.
withdrawnintegerWithdrawn.
recordobjectrequiredThe record since registration.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
betsarray of objectThe bets since registration, newest first (up to 200).
keystring{game_id}:{team_id}.
labelstringThe game, from the side bet.
bet_momentstringedition or close.
moment_atstringWhen the bet was taken (RFC 3339).
format date-time
linenumbernullableThe line at the bet moment.
statusstringpending, graded, void or no_line.
One of: pending, graded, void, no_line (open: new values may be added).
resultstringnullablewin, loss, push, half_win or half_loss once graded.
unitsnumbernullableUnits won or lost.
clvnumbernullableClosing-line value, a fraction.
anchorobjectnullableThe registration day's Merkle root and its RFC 3161 timestamp (null until the day is over).
daystringThe UTC day.
merkle_rootstringsha256 root.
leavesintegerRegistrations that day.
tsastringnullableThe timestamp authority.
stamped_atstringnullableWhen it was stamped.
format date-time
reg_jsonobjectnullableThe registration as hashed (null while the formula is hidden).
noticestringIt isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · not_found · conflict
POST https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2:register
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{}
{
"type": "https://docs.askkelly.ai/problems/conflict",
"title": "Conflict",
"status": 409,
"code": "conflict",
"detail": "This version is already registered; withdraw it first or save a new version.",
"param": "rule_id",
"doc_url": "https://docs.askkelly.ai/problems/conflict",
"request_id": "req_529f74a562c7fa54276958f8",
"retryable": false,
"display_message": "That's already in use."
}POST https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2:register
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"success_type": "e_value",
"success_threshold": 20
}
{
"object": "rule_registration",
"registration_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"reg_hash": "sha256:cfb004340b9bf7088544384474b3d3014851046d98f16d393fe33e9cbc94fa8d",
"registered_at": "2026-10-04T17:30:00Z",
"status": "running",
"withdrawn_at": null,
"formula_hidden": false,
"embargo_until": null,
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"progress": {
"statistic": 0,
"threshold": 20,
"progress": 0
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"author_rules_tried": 2,
"author_registrations": {
"total": 1,
"running": 1,
"met": 0,
"failed": 0,
"withdrawn": 0
},
"record": {
"saved_at": "2026-10-04T17:30:00Z",
"record": "0-0",
"wins": 0,
"losses": 0,
"pushes": 0,
"decisions": 0,
"bets": 0,
"units": 0,
"roi": null,
"clv_mean": null,
"clv_bets": 0,
"e_value": null,
"max_e_value": null,
"e_threshold": 20,
"vs_backtest": null,
"pending": 0,
"no_line": 0,
"void": 0,
"status": "na",
"min_decisions": 30,
"last_bet_at": null
},
"bets": [],
"anchor": null,
"reg_json": {
"v": 1,
"kind": "kelly.rule_registration",
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"reg_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"embargo_until": null,
"registered_at": "2026-10-04T17:30:00Z",
"public_formula": true,
"primary_outcome": "units at the registered bet moment, graded by Kelly's rule tester",
"author_rules_tried": 2
},
"notice": "This tracks a betting rule's record from the day it was registered. It isn't a pick.",
"summary": "Home teams (version 2) is registered: sha256:cfb004340b9b…, met when its e-value reaches 20. Its record counts bets after 2026-10-04T17:30:00Z.",
"assumptions": [
"This tracks a betting rule's record from the day it was registered. It isn't a pick."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_2ceb6410bd9e25bcbc2c8b66",
"live_topics": [],
"rg_line": null
}GET /v1/registrations/{registration_id}
public, no key needed · 0 credits, not charged during the beta · MCP tool get_registration · status live
A public registration: the frozen rule (its formula hidden until an embargo ends), the pass bar, the record since registration and its status (running, met, failed, withdrawn), the published hash, the day's timestamped Merkle root, and how many rules and registrations the author has made.
registration_idstring · pathrequiredA registration's ID (reg_…), from register_rule.
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
200 The rule registration facts.
objectconstrequiredAlways rule_registration.
Always "rule_registration".
registration_idstringrequiredreg_….
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
reg_hashstringrequiredsha256 of the canonical registration JSON (reg_json, keys sorted, no spaces): recompute it to check nothing changed.
pattern ^sha256:[0-9a-f]{64}$
registered_atstringrequiredWhen it was registered (RFC 3339); its record counts bets after this.
format date-time
statusstringrequiredrunning, met, failed or withdrawn.
One of: running, met, failed, withdrawn (open: new values may be added).
withdrawn_atstringnullableWhen the author withdrew it.
format date-time
embargo_untilstringnullableThe date the formula is shown.
format date
ruleobjectThe frozen rule: formula (null while hidden), competition, market, side, bet moment, price policy and stake.
successobjectrequiredThe pass bar set at registration: e_value (met when the forward e-value reaches threshold) or fixed_n (a one-sided test at n decisions).
typestringe_value or fixed_n.
One of: e_value, fixed_n (open: new values may be added).
thresholdnumbere_value: the e-value to reach (20 is alpha 0.05).
nintegerfixed_n: the forward decisions it is judged at.
alphanumberThe test's level.
progressobjectHow far it is toward the bar.
statisticnumbernullableThe running maximum e-value, or the fixed-n p-value once judged.
thresholdnumberThe bar.
progressnumber0 to 1.
backtestobjectnullableThe backtest it rests on: run, data version, tester revision, seasons, record, ROI.
author_rules_triedintegerRules the author had tried on this competition and market when registering (the rules-tried ledger).
min 1
author_registrationsobjectEvery registration the author has made, by status.
totalintegerAll of them.
runningintegerRunning.
metintegerMet.
failedintegerFailed.
withdrawnintegerWithdrawn.
recordobjectrequiredThe record since registration.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
betsarray of objectThe bets since registration, newest first (up to 200).
keystring{game_id}:{team_id}.
labelstringThe game, from the side bet.
bet_momentstringedition or close.
moment_atstringWhen the bet was taken (RFC 3339).
format date-time
linenumbernullableThe line at the bet moment.
statusstringpending, graded, void or no_line.
One of: pending, graded, void, no_line (open: new values may be added).
resultstringnullablewin, loss, push, half_win or half_loss once graded.
unitsnumbernullableUnits won or lost.
clvnumbernullableClosing-line value, a fraction.
anchorobjectnullableThe registration day's Merkle root and its RFC 3161 timestamp (null until the day is over).
daystringThe UTC day.
merkle_rootstringsha256 root.
leavesintegerRegistrations that day.
tsastringnullableThe timestamp authority.
stamped_atstringnullableWhen it was stamped.
format date-time
reg_jsonobjectnullableThe registration as hashed (null while the formula is hidden).
noticestringIt isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/registrations/reg_h7gxzwtpbbhn6rkgsrs4
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_registration",
"registration_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"reg_hash": "sha256:cfb004340b9bf7088544384474b3d3014851046d98f16d393fe33e9cbc94fa8d",
"registered_at": "2026-10-04T17:30:00Z",
"status": "running",
"withdrawn_at": null,
"formula_hidden": false,
"embargo_until": null,
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"progress": {
"statistic": 0,
"threshold": 20,
"progress": 0
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"author_rules_tried": 2,
"author_registrations": {
"total": 1,
"running": 1,
"met": 0,
"failed": 0,
"withdrawn": 0
},
"record": {
"saved_at": "2026-10-04T17:30:00Z",
"record": "0-0",
"wins": 0,
"losses": 0,
"pushes": 0,
"decisions": 0,
"bets": 0,
"units": 0,
"roi": null,
"clv_mean": null,
"clv_bets": 0,
"e_value": null,
"max_e_value": null,
"e_threshold": 20,
"vs_backtest": null,
"pending": 0,
"no_line": 0,
"void": 0,
"status": "na",
"min_decisions": 30,
"last_bet_at": null
},
"bets": [],
"anchor": null,
"reg_json": {
"v": 1,
"kind": "kelly.rule_registration",
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"reg_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"embargo_until": null,
"registered_at": "2026-10-04T17:30:00Z",
"public_formula": true,
"primary_outcome": "units at the registered bet moment, graded by Kelly's rule tester",
"author_rules_tried": 2
},
"notice": "This tracks a betting rule's record from the day it was registered. It isn't a pick.",
"summary": "Registration reg_h7gxzwtpbbhn6rkgsrs4: running, 0-0 since 2026-10-04, +0.0 units (met when its e-value reaches 20).",
"assumptions": [
"This tracks a betting rule's record from the day it was registered. It isn't a pick."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_2e1ecd758c89417812ff7acf",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/registrations/reg_zzzzzzzzzzzzzzzzzzzz
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No registration with that ID.",
"param": "registration_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_a16da7860dc9d726052ae6eb",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/registrations/{registration_id}:withdraw
public, no key needed · 0 credits, not charged during the beta · MCP tool withdraw_registration · status live
Withdraws one of your registrations. Its record to date stays public as a tombstone.
registration_idstring · pathrequiredA registration's ID (reg_…), from register_rule.
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
200 The rule registration facts.
objectconstrequiredAlways rule_registration.
Always "rule_registration".
registration_idstringrequiredreg_….
pattern ^reg_[0-9a-hjkmnp-tv-z]{20}$
reg_hashstringrequiredsha256 of the canonical registration JSON (reg_json, keys sorted, no spaces): recompute it to check nothing changed.
pattern ^sha256:[0-9a-f]{64}$
registered_atstringrequiredWhen it was registered (RFC 3339); its record counts bets after this.
format date-time
statusstringrequiredrunning, met, failed or withdrawn.
One of: running, met, failed, withdrawn (open: new values may be added).
withdrawn_atstringnullableWhen the author withdrew it.
format date-time
embargo_untilstringnullableThe date the formula is shown.
format date
ruleobjectThe frozen rule: formula (null while hidden), competition, market, side, bet moment, price policy and stake.
successobjectrequiredThe pass bar set at registration: e_value (met when the forward e-value reaches threshold) or fixed_n (a one-sided test at n decisions).
typestringe_value or fixed_n.
One of: e_value, fixed_n (open: new values may be added).
thresholdnumbere_value: the e-value to reach (20 is alpha 0.05).
nintegerfixed_n: the forward decisions it is judged at.
alphanumberThe test's level.
progressobjectHow far it is toward the bar.
statisticnumbernullableThe running maximum e-value, or the fixed-n p-value once judged.
thresholdnumberThe bar.
progressnumber0 to 1.
backtestobjectnullableThe backtest it rests on: run, data version, tester revision, seasons, record, ROI.
author_rules_triedintegerRules the author had tried on this competition and market when registering (the rules-tried ledger).
min 1
author_registrationsobjectEvery registration the author has made, by status.
totalintegerAll of them.
runningintegerRunning.
metintegerMet.
failedintegerFailed.
withdrawnintegerWithdrawn.
recordobjectrequiredThe record since registration.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
betsarray of objectThe bets since registration, newest first (up to 200).
keystring{game_id}:{team_id}.
labelstringThe game, from the side bet.
bet_momentstringedition or close.
moment_atstringWhen the bet was taken (RFC 3339).
format date-time
linenumbernullableThe line at the bet moment.
statusstringpending, graded, void or no_line.
One of: pending, graded, void, no_line (open: new values may be added).
resultstringnullablewin, loss, push, half_win or half_loss once graded.
unitsnumbernullableUnits won or lost.
clvnumbernullableClosing-line value, a fraction.
anchorobjectnullableThe registration day's Merkle root and its RFC 3161 timestamp (null until the day is over).
daystringThe UTC day.
merkle_rootstringsha256 root.
leavesintegerRegistrations that day.
tsastringnullableThe timestamp authority.
stamped_atstringnullableWhen it was stamped.
format date-time
reg_jsonobjectnullableThe registration as hashed (null while the formula is hidden).
noticestringIt isn't a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · conflict
POST https://api.askkelly.ai/v1/registrations/reg_h7gxzwtpbbhn6rkgsrs4:withdraw
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{}
{
"object": "rule_registration",
"registration_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"reg_hash": "sha256:cfb004340b9bf7088544384474b3d3014851046d98f16d393fe33e9cbc94fa8d",
"registered_at": "2026-10-04T17:30:00Z",
"status": "withdrawn",
"withdrawn_at": "2026-10-04T17:30:00Z",
"formula_hidden": false,
"embargo_until": null,
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"progress": {
"statistic": 0,
"threshold": 20,
"progress": 0
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"author_rules_tried": 2,
"author_registrations": {
"total": 1,
"running": 0,
"met": 0,
"failed": 0,
"withdrawn": 1
},
"record": {
"saved_at": "2026-10-04T17:30:00Z",
"record": "0-0",
"wins": 0,
"losses": 0,
"pushes": 0,
"decisions": 0,
"bets": 0,
"units": 0,
"roi": null,
"clv_mean": null,
"clv_bets": 0,
"e_value": null,
"max_e_value": null,
"e_threshold": 20,
"vs_backtest": null,
"pending": 0,
"no_line": 0,
"void": 0,
"status": "na",
"min_decisions": 30,
"last_bet_at": null
},
"bets": [],
"anchor": null,
"reg_json": {
"v": 1,
"kind": "kelly.rule_registration",
"rule": {
"name": "Home teams",
"side": "rule",
"price": "consensus",
"stake": "risk",
"market": "spread",
"formula": "=AND(home, rest_days >= 1)",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"version": 2,
"warm_up": "as_written",
"rule_hash": "sha256:af2069d0e4d5307d8c4ea7ac707cc0d3e7e13f5c5ea4f95806a58327400b8676",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"canonical_formula": "{\"args\":[{\"args\":[{\"k\":\"name\",\"path\":[\"rest_days\"]},{\"k\":\"num\",\"v\":1}],\"k\":\"op\",\"op\":\">=\"},{\"k\":\"name\",\"path\":[\"home\"]}],\"fn\":\"AND\",\"k\":\"call\"}"
},
"reg_id": "reg_h7gxzwtpbbhn6rkgsrs4",
"success": {
"type": "e_value",
"alpha": 0.05,
"threshold": 20
},
"backtest": {
"roi": null,
"wins": 0,
"units": 0,
"engine": 8,
"losses": 0,
"record": "0-0",
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"seasons": [
"2025-26"
],
"win_rate": null,
"decisions": 0,
"data_version": "snap_52"
},
"embargo_until": null,
"registered_at": "2026-10-04T17:30:00Z",
"public_formula": true,
"primary_outcome": "units at the registered bet moment, graded by Kelly's rule tester",
"author_rules_tried": 2
},
"notice": "This tracks a betting rule's record from the day it was registered. It isn't a pick.",
"summary": "Registration reg_h7gxzwtpbbhn6rkgsrs4 is withdrawn; its record to date (0-0) stays public.",
"assumptions": [
"This tracks a betting rule's record from the day it was registered. It isn't a pick."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_5c908e27466f1e05b5c41324",
"live_topics": [],
"rg_line": null
}GET /v1/rules/{rule_id}/matches
public, no key needed · 0 credits, not charged during the beta · MCP tool list_rule_matches · status live
The games a saved rule's alert automations hold now: matched, just unmatched, or frozen at their bet moment, with the worst price at which the backtest still breaks even. Turn alerts on with an automation whose scope names the rule (scope.rule_id).
rule_idstring · pathrequiredA saved rule's ID (rl_…), from test_rule's rule_version.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
versioninteger · queryThe version whose backtest sets the price guard (default the newest).
min 1
200 The rule matches facts.
objectconstrequiredAlways rule_matches.
Always "rule_matches".
rule_idstringrequiredThe saved rule.
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerThe version read.
min 1
break_even_priceobjectnullableThe worst price at which the backtest still breaks even (its win rate's 80% Wilson lower bound). Null without an edge to guard.
americanintegerAmerican odds.
matchedintegerrequiredGames matched now.
matchesarray of objectrequiredGames, matched first.
keystring{game_id}:{team_id}.
statestringmatched, unmatched or frozen (reached its bet moment).
One of: matched, unmatched, frozen (open: new values may be added).
versionintegerThe version alerted.
moment_atstringnullableIts bet moment.
format date-time
changed_atstringWhen its state last changed.
format date-time
alertedstringnullableThe last event sent for it.
alerts_onbooleanThe rule has an alert automation.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/rl_5bbmw66t2r1jvg6cpxt2/matches
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_matches",
"rule_id": "rl_5bbmw66t2r1jvg6cpxt2",
"name": "Home teams",
"version": 2,
"break_even_price": null,
"matched": 0,
"matches": [],
"alerts_on": false,
"summary": "Home teams: (0) games match now.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_0a4c579dc55200489daee103",
"live_topics": [],
"rg_line": null
}POST /v1/rules:sweep
public, no key needed · 1 credit, not charged during the beta · MCP tool sweep_rule · status live
Up to three of a rule's numbers swept as slots ({1..6}, {-7..-3 step 0.5}, {1|2|4}; 256 settings at most), every setting bet on the same games and the search charged (backtester T8: F41, F16, F29, F30). The headline is the median ROI across every setting; the plateau pick sits beside the single best; the newest 30% of the bets are sealed and opened once for the picked setting; PBO, SPA and Reality Check p, the deflated ROI, walk-forward over seasons and purged cross-validation; then the picked setting's full rule test, its verdict reading the sweep. Every setting counts toward the account's rules tried. Never a pick or a ranking.
Body, application/json (required):
formulastringThe rule with up to three slots where numbers go: {1..6}, {-7..-3 step 0.5} or {1|2|4}, e.g. =AND(home, rest_days - opponent.rest_days >= {1..6}). At most 256 settings.
min length 2 · max length 2000
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...; racing: horse_racing, greyhounds or harness (seasons are calendar years).
pattern ^[a-z0-9_]{2,24}$
seasonsarray of stringSeasons to test, oldest first (1 to 25), e.g. ["2021", "2022"]; or use from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
sidestringrule (default) bets the team the rule matched; fade bets the other side.
One of: rule, fade (open: new values may be added).
bet_momentstringclose (default), edition (football: the weekly sheet's Monday 10 AM ET line) or day_before (other sports: 24 hours before the start).
One of: close, edition, day_before (open: new values may be added).
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
stakestringrisk (default): 1 unit risked a bet; to_win: each bet sized to win 1 unit.
One of: risk, to_win (open: new values may be added).
pricestringconsensus (default): Kelly's price at the bet moment; flat_110: every bet at -110.
One of: consensus, flat_110 (open: new values may be added).
testsintegerRules tried on this data outside Kelly, this one included (default 1). Kelly counts the account's own; this may only raise that count.
min 1 · max 100000
base_formulastringThe base the rule filters, e.g. =AND(NOT(home), spread_close > 0); it must hold on every game the rule bets. Default: Kelly takes the rule's who-or-when conditions.
min length 2 · max length 2000
plan_roinumberAlso report the bets needed to prove this ROI (a fraction, 0.05 is 5%).
ci_levelnumberThe ROI interval's level, 0.8 to 0.99 (default 0.95).
min 0.8 · max 0.99
max_stale_minutesintegerLeave out a price whose snapshot is older than this at the bet moment (default 60 at the close, 360 earlier; never under 10 for games before October 2022).
min 5 · max 10080
bankroll_unitsnumberBankroll in units for the ruin chance and the staking replays (default 100).
max 1000000
horizon_betsintegerBets ahead the ruin chance looks (default 500).
min 1 · max 1000
pathsintegerSimulated futures for the ruin chance (default 5000).
min 1000 · max 10000
staking_plansarray of stringStaking plans to replay: flat, to_win, percent, kelly (default all four).
min items 1 · max items 4
percentnumberShare of the bankroll the percent plan bets (default 0.02).
max 0.5
kelly_fractionnumberMultiple of the Kelly criterion the kelly plan bets; omitted, the honest multiple from the record's sample size.
max 1
kelly_capnumberMost of the bankroll the kelly plan puts on one bet (default 0.05).
max 1
max_bets_per_dayintegerCap: bets a day (ET).
min 1 · max 1000
max_bets_per_gameintegerCap: bets a game.
min 1 · max 10
max_bets_per_team_weekintegerCap: bets on a team a week.
min 1 · max 50
max_open_stakenumberCap: share of the bankroll at risk at once.
max 1
daily_stop_loss_unitsnumberCap: no more bets that day once this many units are lost.
max 100000
cap_prioritystringWhich bets a cap keeps: earliest (default) or random.
One of: earliest, random (open: new values may be added).
warm_upstringas_written (default) grades every bet; require_full grades only bets whose windows had all the earlier games they read (ROLLING_AVG(margin, 5) needs 5).
One of: as_written, require_full (open: new values may be added).
settlestringRacing only: the price bets settle at: sp (industry SP; win default, each_way), bsp (Betfair SP; place, lay and lay_place default) or tote.
One of: sp, bsp, tote (open: new values may be added).
commissionnumberRacing only: Betfair commission on a winning BSP bet's winnings, as a share (default 0.05).
min 0 · max 0.2
ew_placesintegerRacing each_way only: places paid, instead of the standard terms by field size.
min 1 · max 8
ew_fractionnumberRacing each_way only: the place fraction of the win odds (0.2 is 1/5, 0.25 is 1/4), instead of the standard terms.
max 1
robustness_offarray of stringRobustness blocks to leave out (all of them with "all"); every block runs by default.
max items 10
ablationstringshapley (default): each condition's share of the extra units (Shapley values); leave_one_out: only the rule without each condition.
One of: shapley, leave_one_out (open: new values may be added).
break_datestringA date to test for a change in the edge (a Chow test), e.g. 2023-09-01.
format date
random_baseline_simsintegerRandom rules drawn for the baseline (default 2,000).
min 100 · max 10000
monte_carlo_simsintegerReorderings of the bets (default 10,000; fewer, never under 2,000, for long records).
min 100 · max 20000
monte_carlo_modestringblock (default: whole bet-days resampled in runs of about a week), iid or permutation.
One of: block, iid, permutation (open: new values may be added).
skip_probnumberChance each bet is missed in the Monte Carlo and noise runs (default 0).
min 0 · max 0.5
trim_winnersnumberShare of the biggest winners removed before the Monte Carlo (default 0).
min 0 · max 0.2
noise_runsintegerRuns a noise family (default 200).
min 10 · max 1000
input_scalenumberNoise on the numbers the rule compares, as a share of their spread (default 0.10).
min 0 · max 0.5
line_probnumberChance each bet's line moves half a point either way in the line noise (default 0.5).
min 0 · max 1
savebooleantrue keeps this rule as a version of a saved rule named name (or of rule_id): the same rule again keeps its version, a different one is the next.
namestringThe saved rule's name, for a new one (1 to 120 characters).
min length 1 · max length 120
rule_idstringAdd this test as a version of a rule already saved (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
transferbooleanAlso run the rule on its other market (spread and moneyline; over and under) and on other leagues, with its numbers moved to the same percentile there, pooled across leagues (transfer). Evidence beside the verdict; these runs don't count toward rules tried.
transfer_leaguesarray of stringLeagues to carry it to (default: sibling leagues first, then the rest).
min items 1 · max items 13
transfer_marketsstringauto (default) adds the other market of the rule's own league; none leaves it out.
One of: auto, none (open: new values may be added).
transfer_thresholdsstringboth (default), literal (the rule's numbers as written) or quantile (each number moved to the same percentile of its field in the other league).
One of: both, literal, quantile (open: new values may be added).
translation_idstringThe translation_id a translate_rule result gave, when the formula came from the person's words; it is kept with the test.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
chartsbooleantrue adds the equity, drawdown and rolling ROI charts as Kelly Chart Specs (detail full).
marketstringspread (default), moneyline, over, under or draw (soccer).
One of: spread, moneyline, over, under, draw (open: new values may be added).
detailstringfull (default) or summary (no grid, windows, paths, bets list, splits or condition impact).
One of: full, summary (open: new values may be added).
pickstringplateau (default): the setting whose neighbours did best; peak: the single best setting.
One of: plateau, peak (open: new values may be added).
walk_forward_windowstringanchored (default): train on every earlier season; rolling: the last walk_forward_train_seasons only.
One of: anchored, rolling (open: new values may be added).
walk_forward_train_seasonsintegerTraining seasons (anchored: the fewest, default 2; rolling: the window, default 3).
min 1 · max 10
walk_forward_stepstringseason (default) or half (NBA, WNBA, NHL, MLB: half seasons).
One of: season, half (open: new values may be added).
walk_forward_objectivestringWhat each window chooses by: z (default; with 30 decisions), units or roi.
One of: z, units, roi (open: new values may be added).
cv_groupsintegerPurged cross-validation's time groups (default 6).
min 3 · max 10
cv_test_groupsintegerGroups held out in each split (default 2).
min 1 · max 4
200 The rule sweep facts.
objectconstrequiredAlways rule_sweep.
Always "rule_sweep".
ruleobjectrequiredThe rule as swept.
formulastringThe formula with its slots.
competitionstringLeague.
marketstringMarket.
sidestringrule or fade.
bet_momentstringclose or edition.
seasonsarray of stringSeasons.
season_typestringSeason type.
stakestringrisk or to_win.
pricestringconsensus or flat_110.
headline_roinumberrequirednullableThe headline ROI: the median across every setting (= sweep.spp_median_roi).
sweepobjectrequiredThe sweep (F41): every setting on the same games, the search charged.
slotsarray of objectThe swept numbers (up to three).
namestringa, b or c, in the order written.
writtenstringThe slot as written, e.g. {1..6}.
valuesarray of numberIts values, in order.
variantsintegerSettings tried (up to 256); each one counts as a rule tried.
tests_addedintegerSettings new to the account's rules tried (0 when the same sweep runs again).
gridarray of objectnullableEvery setting before the sealed window, in grid order (summary detail: null).
settingobjectA setting: each slot's name (a, b, c) and its value.
formulastringThe rule at this setting.
decisionsintegerWins plus losses.
recordstringWins-losses(-pushes).
unitsnumberUnits won or lost.
roinumbernullableUnits over units risked.
p_valuenumbernullableOne-sided p against no skill at these prices (z = Σ units / √Σ s²(d − 1)).
spp_median_roinumbernullableThe headline: the median ROI across every setting (System Parameter Permutation), what the idea made, not its luckiest setting.
roi_p05numbernullable5th percentile of the settings' ROIs.
roi_p95numbernullable95th percentile of the settings' ROIs.
pick_rulestringplateau (default): the best neighbourhood; peak: the best single setting.
One of: plateau, peak (open: new values may be added).
min_decisions_to_pickintegerDecisions a setting needs before it can be picked.
fast_pathbooleanSettings were moved on the rule's circuit (true) or each evaluated.
bestobjectnullableThe best single setting before the sealed window (shown beside the pick, not instead).
indexintegerThe setting's place in the grid (row-major, the last slot moving fastest).
settingobjectA setting: each slot's name (a, b, c) and its value.
formulastringThe rule at this setting.
roinumbernullableROI before the sealed window.
decisionsintegerWins plus losses before the sealed window.
pickedobjectnullableThe picked setting.
indexintegerIts place in the grid.
settingobjectA setting: each slot's name (a, b, c) and its value.
formulastringThe rule at this setting.
roinumbernullableROI before the sealed window.
decisionsintegerWins plus losses.
reasonstringplateau or peak.
neighbourhood_roinumbernullableThe mean ROI of its neighbourhood (plateau pick).
edge_of_rangearray of stringSlots where the pick is the end of the range: the results may keep rising past it; widening the range counts as more rules tried.
pbonumbernullableProbability of backtest overfitting (F16, CSCV on pooled ROI): the chance the in-sample best setting does no better than the median setting out of sample. Over 0.5 is worse than picking at random.
blocksintegernullableTime blocks PBO split the bets into (16; 10 under 300 bets; 8 under 160).
splitsintegernullableHalves PBO tried: C(blocks, blocks/2).
degradation_slopenumbernullableSlope of out-of-sample ROI on in-sample ROI for the in-sample best (negative: what looks best does worse).
prob_oos_lossnumbernullableShare of halves where the in-sample best lost money out of sample.
logit_mediannumbernullableMedian logit of the in-sample best's out-of-sample rank.
spa_p_valuenumbernullableHansen's SPA p for the best setting (stationary bootstrap over days).
reality_check_pnumbernullableWhite's Reality Check p for the best setting.
p_adjusted_pickednumbernullableRomano-Wolf adjusted p for the picked setting.
k_effectivenumbernullableEffective settings (Li-Ji): near-duplicate settings count as fewer.
deflated_roinumbernullableThe best ROI less the best a search of k_effective settings finds by chance.
expected_max_roinumbernullableThat expected best by chance.
selectionobjectThe window the settings were chosen on.
fromstringnullableFirst day (YYYY-MM-DD).
tostringnullableLast day.
betsintegerRows any setting bet in it.
forwardobjectnullableThe sealed newest 30% (by day), opened once for the picked setting only.
fromstringFirst sealed day.
tostringLast day.
betsintegerBets.
decisionsintegerWins plus losses.
recordstringWins-losses(-pushes).
unitsnumberUnits won or lost.
roinumbernullableUnits over units risked.
p_valuenumbernullableOne-sided p against no skill.
nastringnullableWhy there is no pick, when there isn't.
walk_forwardobjectrequirednullableWalk-forward (F29): the setting chosen on earlier seasons, bet on the next.
windowstringanchored or rolling.
One of: anchored, rolling (open: new values may be added).
train_seasonsintegerTraining seasons (anchored: at least this many).
stepstringseason or half.
One of: season, half (open: new values may be added).
objectivestringz, units or roi.
One of: z, units, roi (open: new values may be added).
min_decisionsintegerDecisions a setting needs to be chosen.
windowsarray of objectEach window in time order.
trainarray of stringFirst and last training season (or half).
teststringThe test season (or half).
chosenobjectnullableThe setting chosen on the training seasons.
chosen_formulastringThat setting's rule.
carriedbooleanNo setting had enough decisions; the previous choice was kept.
is_roinumbernullableIts ROI on the training seasons.
is_decisionsintegerIts decisions on the training seasons.
oosobjectnullableIts record on the test season.
recordstringWins-losses(-pushes).
decisionsintegerWins plus losses.
unitsnumberUnits won or lost.
roinumbernullableUnits over units risked.
notestringWhy the window has no choice.
oosobjectnullableThe test seasons stitched together.
recordstringWins-losses(-pushes).
decisionsintegerWins plus losses.
unitsnumberUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% interval (normal, at break-even variance).
p_valuenumbernullableOne-sided p against no skill.
wfenumbernullableWalk-forward efficiency: out-of-sample ROI over the chosen settings' in-sample ROI.
windows_profitablenumbernullableShare of windows that made money.
parameter_stabilitynumbernullableShare of windows that chose the most chosen setting.
matrixarray of objectnullableRolling windows of 2, 3 and 4 seasons by season (and half season where the league allows).
train_seasonsintegerRolling training seasons.
stepstringseason or half.
windowsintegerWindows.
oos_unitsnumbernullableOut-of-sample units.
wfenumbernullableWalk-forward efficiency.
passbooleanPardo's four criteria passed.
matrix_passedintegernullableCells passed.
matrix_cellsintegernullableCells run.
matrix_robustbooleannullableAt least two thirds of the cells passed.
statusstringpass, fail or na.
One of: pass, fail, na (open: new values may be added).
nastringnullableWhy it can't run.
cross_validationobjectrequirednullablePurged cross-validation (F30): settings chosen on the rest, bet on the held-out groups.
groupsintegerContiguous groups of about equal bets.
test_groupsintegerGroups held out a split.
splitsintegernullableSplits: C(groups, test_groups).
pathsintegernullableFull out-of-sample paths.
weakbooleanFewer than three seasons: 4 groups, 1 held out.
lookback_daysnumberDays the rule's longest window reads back.
season_to_datebooleanThe rule reads a season-to-date window (purged to the season's start).
embargo_daysnumberDays after each held-out group left out of training.
purged_rowsintegerBets left out of some training set by the purge or embargo.
path_resultsarray of objectnullableEach path (summary detail: null).
pathintegerPath number.
decisionsintegerWins plus losses.
recordstringWins-losses(-pushes).
unitsnumberUnits won or lost.
roinumbernullableUnits over units risked.
incompletebooleanA split on the path had no setting with enough decisions.
roi_paths_minnumbernullableLowest path ROI.
roi_paths_mediannumbernullableMedian path ROI.
roi_paths_maxnumbernullableHighest path ROI.
paths_profitablenumbernullableShare of paths that made money.
pbonumbernullablePBO over the splits.
degradationnumbernullableMedian path ROI over the chosen settings' median in-sample ROI.
statusstringpass (PBO < 0.5 and 60% of paths profitable), fail or na.
One of: pass, fail, na (open: new values may be added).
nastringnullableWhy it can't run.
picked_testobjectrequirednullableThe picked setting's full rule test (as test_rule returns it), its verdict reading the sweep: not just luck needs PBO ≤ 0.25, SPA p ≤ 0.05 and cross-validation; still working is the sealed window and walk-forward.
verdictobjectVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
overallobjectEvery graded bet.
recordstringWins-losses(-pushes).
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
betsintegerGraded bets.
decisionsintegerWins plus losses.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_intervalarray of numbernullableROI's 95% interval at 1 unit risked a bet.
win_ratenumbernullableWins over decisions.
break_evennumbernullableThe win rate that breaks even at these prices.
significanceobjectnullableThe record test (kmath.records.record_test) and verdict v2's statistics (kmath.significance): the p-value, adjusted for the rules tried, and the anytime-valid test.
p_valuenumbernullableThe larger of the exact and the clustered one-sided p against no skill at these prices (F12, F23).
methodstringnullableexact_binomial, exact_simulation, saddlepoint or z_score.
znumbernullableThe score z: units over the square root of Σ(d − 1).
t_statnumbernullableThe t statistic with the sample variance.
p_value_exactnumbernullableThe exact (or z) tail alone.
p_value_clusterednumbernullableThe tail with bets clustered by slate and team-season (F23).
p_value_vs_vignumbernullableAgainst a random bet at the no-vig close (needs closes).
p_adjustednumbernullableThe p-value adjusted for the rules tried (Romano-Wolf on the ledger's rules, Bonferroni on the count otherwise; F14).
adjust_methodstringnullableromano_wolf, holm or none.
testsintegernullableRules tried in this family, this one included.
tests_effectivenumbernullableEffective tests (Li-Ji): near-duplicate rules count as fewer.
reality_check_pnumbernullableWhite's Reality Check p (when the ledger holds the rules' units).
spa_pnumbernullableHansen's SPA p (when the ledger holds the rules' units).
clears_t3booleannullablez at or above 3 (Harvey-Liu-Zhu's bar).
haircutnumbernullableHarvey-Liu haircut: the share of the ROI the rules tried take away.
haircut_roinumbernullableROI after the haircut, per decision.
anytime_validobjectnullableThe record as a bet against no skill (F17): e-value, anytime p, ROI range.
e_valuenumbernullableThe e-value now.
p_valuenumbernullable1 over the largest e-value so far.
max_enumbernullableThe largest e-value along the path.
roi_cs95array of numbernullableAn ROI range valid at every bet.
roi_ci95array of numbernullableROI's 95% score interval, clustered (F13, F23).
units_ci95array of numbernullableUnits' 95% interval: decisions times the ROI interval.
roi_ciobjectnullableROI intervals at ci_level: score (the one shown), BCa, block bootstrap over slates and anytime-valid (F13).
deflatedobjectnullablePer-bet Sharpe, PSR and DSR at the effective tests, and the bets needed: deflated (sample_needed), single test and 80% power, each times the design effect (F15).
sample_neededintegernullableDecisions needed, deflated for the rules tried; null with no edge to prove.
sample_needed_singleintegernullableDecisions needed for one rule tried.
sample_needed_power80integernullableDecisions for 80% power at this ROI.
sample_needed_planintegernullableDecisions for 80% power at plan_roi.
dsrnumbernullableDeflated Sharpe ratio.
psrnumbernullableProbabilistic Sharpe ratio.
sample_neededobjectnullableBets needed to tell this win rate from break-even (one-sided, with 80% power, ...).
small_samplebooleanUnder 30 decisions.
few_decisionsbooleanUnder 5 decisions.
honestobjectnullableThe honest forward ROI (F22, the headline ROI): empirical Bayes toward a prior mean (the rule's own closing-line value when at least half its bets have a close, else the usual two-way margin), less the winner's curse for the rules tried and the out-of-sample decay.
roinumbernullableHonest forward ROI.
ci80array of numbernullable80% range.
raw_roinumbernullableThe past ROI per decision, beside it.
prior_meannumbernullableThe prior mean the ROI is shrunk toward (prior.source says which).
prior_sdnumbernullableThe spread of true ROIs (0.02 until fitted on the ledger).
shrinknumbernullableWeight on the prior, se²/(se² + τ²).
selection_adjnumbernullableWinner's-curse correction for the rules tried.
decay_factornumbernullableOut-of-sample decay on the excess (0.74).
stabilizes_at_betsintegernullableDecisions at which the record and the prior weigh the same.
uses_clvbooleanThe prior mean is the rule's own closing-line value.
prior_basisstringfitted (prior_sd from the ledger's fit) or seed.
priorobjectWhere the prior mean comes from.
sourcestringclv (the rule's mean closing-line value), margin (the two-way margin, -4.55%) or fitted.
meannumbernullableThe prior mean used.
closesintegerDecided bets with a no-vig close.
cappedbooleanA closing-line value above 0 was held at 0: under 50 bets have a close.
wordsstringThe prior, said plainly.
dependenceobjectnullableBets clustered by slate (the date; the week in the NFL and college football) and team-season (F23).
clustersintegerSlates.
design_effectnumberClustered over iid variance.
effective_betsintegerDecisions over the design effect (at least 1).
methodstringcr1_twoway, cr1_slate, cr1_larger_oneway, wild_bootstrap or wild_bootstrap_webb.
baseobjectnullableThe rule against the rest of its own base (F19): the who-or-when conditions alone, or base_formula.
formulastringnullableThe base, or null for the whole market.
sourcestringauto, user or market.
One of: auto, user, market (open: new values may be added).
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
roinumbernullableThe base's ROI per decision.
liftnumbernullableRule ROI less base ROI.
p_value_vs_basenumbernullableSeason-stratified permutation p of the rule within its base.
units_from_basenumbernullableUnits the base alone would have made on the rule's stakes.
units_from_filternumbernullableThe rule's units less units_from_base.
clvobjectClosing-line value as evidence (F21): each bet's no-vig CLV, q_close x price - 1, on bets Kelly holds a close for. Empty when betting at the close.
bets_with_closeintegerBets with a close.
bets_without_closeintegerBets without one.
referencestringThe close used: consensus (the no-vig average across books).
devigstringHow the vig was removed: multiplicative.
lines_convertedintegerBets whose close was carried from the closing line to the bet's line (NFL spreads and totals).
coveragenumbernullableShare of bets with a close probability (null with no bets).
closes_staleintegerCloses left out as older than 60 minutes at the start, or seen after it (audit F9): those bets keep their result but have no CLV.
points_coveragenumbernullableShare of bets with points beaten (spreads and totals).
points_testobjectnullableThe clustered test on points beaten (audit F3).
betsintegerBets with points beaten.
meannumbernullableMean points the bet's line beat the close by.
std_error_clusterednumbernullableStandard error clustered by slate and team-season.
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean points beaten > 0.
cluster_methodstringHow the standard error was clustered.
basisstringWhat Beats the close is judged on: probability, or points when under 80% of bets have a close probability and more have points beaten (audit F3).
One of: probability, points (open: new values may be added).
meannumbernullableMean CLV (fraction): the bets' expected ROI if the close is right.
ci95array of numbernullableClustered 95% interval on the mean.
min items 2 · max items 2
p_valuenumbernullableOne-sided clustered p of mean CLV > 0.
beat_ratenumbernullableShare of bets that beat the close, ties left out.
beat_rate_ci95array of numbernullableWilson 95% interval on beat_rate.
min items 2 · max items 2
sign_test_pnumbernullableExact one-sided sign test against half.
mean_probnumbernullableMean of the close's no-vig probability less the bet moment's.
mean_pointsnumbernullableMean points the bet's line beat the closing line by (spreads and totals).
points_betsintegerBets mean_points is over.
profit_slopenumbernullablePer-bet units regressed on CLV: near 1 when CLV turned into profit.
profit_slope_senumbernullableIts slate-clustered standard error.
sample_neededintegernullableBets CLV needs for a 95% interval to clear 0 at this mean and spread (null unless the mean is above 0).
sample_needed_roiintegernullableBets profit would need at the same edge.
significanceobjectnullableThe CLV mean's tests (null under two bets).
tnumbernullableMean over its iid standard error.
ci95array of numbernullableStudent 95% interval (iid).
min items 2 · max items 2
std_errornumbernullableiid standard error.
bets_needednumbernullableBets a t of 2 needs at this mean and spread.
std_error_clusterednumbernullableStandard error clustered by slate and team-season (F23).
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean CLV > 0 (wild cluster bootstrap under 42 slates, else Student t); beats_close reads it.
cluster_methodstringHow the standard error was clustered.
design_effectnumbernullableClustered variance over iid variance.
notestringWhy the block is empty, when it is.
luckobjectThe luck-adjusted result (F20): the units the market's fair probabilities expected the decided bets to make against what they made. Pushes count zero both ways.
readingstringluck (bets before the close: the gap is outcome noise) or versus_the_market (bets at the close: the gap is the rule's edge on the closing market).
One of: luck, versus_the_market (open: new values may be added).
close_sourcestringconsensus.
devigstringmultiplicative.
betsintegerDecided bets with a fair probability.
bets_without_probintegerDecided bets without one.
left_outobjectBets left out.
pushintegerPushes.
halfintegerHalf results (quarter lines).
lines_convertedintegerBets whose close was carried to the bet's line.
expected_unitsnumbernullableSum of stake x (q x (price - 1) - (1 - q)).
actual_unitsnumbernullableUnits the bets made.
luck_unitsnumbernullableactual_units less expected_units.
luck_se_unitsnumbernullableluck's standard error given the probabilities.
expected_winsnumbernullableSum of q.
actual_winsintegernullableWins.
luck_percentilenumbernullableWhere the wins sit among the market's (Poisson-binomial, mid-p): 0.95 is luckier than (95) of (100) identical paths.
expected_roinumbernullableexpected_units over units risked.
actual_roinumbernullableactual_units over units risked.
thetanumbernullablePer-bet units regressed on expected units (30 bets or more): near 1 when the close is calibrated on these bets.
by_seasonarray of objectEach season, in calendar order; luck_units add up to the total.
seasonstringSeason.
betsintegerDecided bets with a probability.
expected_unitsnumberUnits the market expected.
actual_unitsnumberUnits made.
luck_unitsnumberactual_units less expected_units.
capacityobjectnullableF8: the limits Kelly measured at the bet moment (ADR-K783); evidence beside the verdict, never in it. No book publishes its limits, so only measured ones are shown.
availablebooleanWhether measured limits are shown for this market and moment.
marketstringnullableThe market as line history names it (spreads, totals, h2h).
before_startstringnullableThe time before the start the bet moment falls in.
One of: 0_1h, 1_12h, 12_36h, 36_72h, 72h_plus (open: new values may be added).
booksarray of objectPer book, highest median first.
bookstringnullableThe book as the vendor names it.
vendorstringnullableWho measured it.
currencystringnullableThe stake currency, null when the vendor doesn't say.
mediannumbernullableThe median limit.
p25numbernullableIts 25th percentile.
p75numbernullableIts 75th percentile.
minnumbernullableThe smallest.
maxnumbernullableThe largest.
snapshotsintegerLimit readings behind it.
gamesintegerGames behind it.
first_seenstringnullableThe first reading's date.
last_seenstringnullableThe last reading's date.
max_stake_per_betobjectnullableThe highest book median: the most one book typically took on one bet.
bookstringnullableThe book.
amountnumbernullableThe median limit.
currencystringnullableIts currency.
per_season_at_limitnumbernullablemax_stake_per_bet times the rule's bets a season.
bets_per_seasonnumbernullableThe rule's bets a season.
limits_heldintegerLimit readings Kelly holds for this market and moment.
reasonstringnullableWhy none are shown, when available is false.
stressobjectThe same bets graded harder.
half_point_worseobjectnullableEvery line half a point worse (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
flat_110objectnullableEvery bet at -110 (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
other_sideobjectThe other side of every bet.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
break_evenobjectnullableF7: how much worse the price or line can get before the profit is gone.
price_multipliernumbernullableThe share of each payout (d - 1) that breaks even (under 1: the room to spare).
american_avgnumbernullableThe flat American price that breaks even.
decimal_avgnumbernullableThe same, decimal.
line_pointsnumbernullableThe most points worse (in halves, up to 3) the line could be with a profit left; 0 when none; null for moneylines.
gridarray of objectnullableF7: the bets at -105 to -120 with the line 0 to 1 point worse (moneylines: payouts cut 0-6%); left out at detail summary.
vigintegernullableEvery bet at this American price.
line_shiftnumbernullablePoints the line moved (negative: worse).
price_multipliernumbernullableMoneylines: the share of each payout kept.
unitsnumbernullableUnits.
roinumbernullableROI.
slippage_mcobjectnullableF7: each bet's line worsened by a draw from the market's own Monday-to-close moves, 1,000 seeded runs; null without such moves.
runsintegerRuns.
roi_p05numbernullable5th percentile ROI.
roi_p50numbernullableMedian ROI.
moves_fromintegerLine moves the draws come from.
move_mediannumbernullableTheir median size, points.
price_worseobjectnullableTeasers: every ticket ten cents worse (price_american). Racing: every bet's winnings 5% smaller.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
legsobjectnullableTeasers: every teased leg, judged against the per-leg break-even (1/D)^(1/n).
recordstringLegs won-lost(-pushed).
decisionsintegerLegs won plus lost.
cover_ratenumbernullableLegs won over legs decided.
break_even_per_legnumberThe cover rate a leg needs at the ticket's price.
significanceobjectnullableThe legs' record test against the per-leg break-even.
sample_neededobjectnullableLegs needed to tell this cover rate from the break-even.
random_baselineobjectnullableF18: the rule against random bets of the same count on the same games, matched by season and price bucket.
simsintegerRandom rules drawn.
betsintegerThe rule's bets.
poolintegerBets the market offered.
percentilenumbernullableShare of random rules with a lower ROI.
beats_per_100integerRandom rules out of 100 the rule beats.
p_valuenumbernullable(1 + random rules at or above) / (sims + 1).
analytic_pnumbernullableThe finite-population normal cross-check.
random_roi_mediannumbernullableMedian random ROI (about minus the vig).
random_roi_p05numbernullable5th percentile random ROI.
random_roi_p95numbernullable95th percentile.
best_of_testsobjectWould the best of M random rules (M = rules tried) do this well?
testsintegerM.
p_valuenumbernullable1 - F(ROI)^M.
histogramobjectnullableRandom ROIs in 30 bins (edges, counts); full detail only.
decayobjectnullableF24: is the rule still making money? Rolling ROI and CLV, one unknown break (sup-F, block permutation p), a CUSUM alarm, a known-date test, the trend and the decay ratio.
betsintegerGraded bets.
windowintegernullableRolling window, bets.
rollingarray of objectnullableUp to 200 points; full detail only.
breakobjectnullableThe most likely date the edge changed.
atstringDate.
bet_indexintegerBets before it.
sup_fnumbernullablesup-F.
p_valuenumbernullableBlock permutation p-value.
at_edgebooleanWithin the first or last 15%: not a confirmed break.
roi_beforenumbernullableROI before.
roi_afternumbernullableROI after.
bets_afterintegerBets after.
cusumobjectnullablePage's lower CUSUM alarm.
known_breakobjectnullableA Chow test at break_date.
trend_per_100_betsnumbernullableProfit per bet's change per 100 bets (1 unit risked).
decay_rationumbernullableROI after the break (or the newest 30%) over ROI before.
clv_breakobjectnullableThe break test on CLV.
roi_newest_30numbernullableROI of the newest 30% of bets.
ablationobjectnullableF27: each top-level AND condition's Shapley share of the units beyond the base's ROI.
baseobjectEvery game with a line: decisions and ROI.
totalnumbernullableThe rule's units beyond the base's ROI (the Shapley values add up to it).
clausesarray of objectOne a condition, in the order written.
conditionstringThe condition.
shapley_unitsnumbernullableIts share of the extra units.
labelstringdead, harmful, load_bearing or helps.
One of: dead, harmful, load_bearing, helps (open: new values may be added).
fixobjectnullableRemove it: the rule without it.
pairsarray of objectPairs (up to 6 conditions).
methodstringexact or sampled.
One of: exact, sampled (open: new values may be added).
nearbyobjectnullableF28: every number in the rule moved two steps each way (steps fixed by the field).
numbersarray of objectThe rule's numbers.
variantsintegerVariants in the grid.
distinct_bet_setsintegerVariants with their own bets.
roi_mediannumbernullableMedian ROI of the variants.
roi_p05numbernullable5th percentile.
roi_p95numbernullable95th percentile.
spike_rationumbernullable(ROI - median) / IQR of the variants.
readingstringnullableplateau, slope or spike.
nastringnullableWhy there are no numbers to move.
monte_carloobjectnullableF31: drawdowns and losing runs the same bets give in other orders.
simsintegerSimulations.
modestringblock, iid or permutation.
max_drawdown_unitsobjectMax drawdown, units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
longest_losing_runobjectLongest losing run, bets.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
final_unitsobjectFinal units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
prob_below_zeronumbernullableShare of runs finishing below zero.
worst_100_bets_units_p05numbernullable5th percentile of the worst 100-bet stretch.
noiseobjectnullableF32: jittered inputs, lines moved half a point either way, and an earlier bet moment.
inputsobjectnullableThe rule's compared numbers jittered.
runsintegerRuns.
profitable_runsintegerRuns with a profit.
roi_mediannumbernullableMedian ROI.
roi_p05numbernullable5th percentile ROI.
overlap_mediannumbernullableMedian Jaccard overlap of the bets with the rule's.
linesobjectnullableLines (moneylines: payouts) moved at random.
delayobjectnullableThe rule bet at the earlier moment.
concentrationobjectnullableF34: the profit without the best season, team, opponent, month and weekday.
byarray of objectGroupings in a fixed order.
groupstringseason, team, opponent, month or weekday.
bestobjectnullableThe best group (3+ bets).
without_bestobjectnullableThe rule without it.
effective_groupsnumbernullable1 / Herfindahl of the profitable groups' shares.
p_more_concentratednumbernullablePermutation p-value (teams and opponents).
trim_top_0_05objectnullableWithout the top 5% of bets (moneylines) or bet-days.
split_dimensionsarray of objectnullableF35: the rule by fixed dimensions with shrunk ROI and intervals.
dimensionstringThe dimension.
heterogeneity_pnumbernullableChi-square p-value of win rates across its levels.
heterogeneity_qnumbernullableBenjamini-Hochberg q across the dimensions.
real_differencebooleanq < 0.10.
where_profitstringnullableThe level with the most shrunk extra units, when real.
cellsarray of objectnullableLevels in a fixed order; full detail only.
robustnessobjectThe suite's run: the verdict v2 checks it feeds and its seed.
checksarray of objecthalf_point_worse, nearby_holds, not_concentrated and still_working.
keystringrequiredThe check.
One of: half_point_worse, nearby_holds, not_concentrated, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na.
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumbernullableThe figure the status compares.
thresholdnumbernullableWhat it is compared with.
fast_pathbooleanThe rule's numbers were re-bet from its compiled columns.
seedintegerThe seed (a hash of the rule and its options).
evaluationsintegerFormula evaluations the suite ran.
offarray of stringBlocks switched off.
pathobjectThe units path in date order: drawdown and longest losing run; at detail full its series (F47).
max_drawdown_unitsnumberThe deepest fall from a peak, units.
longest_losing_runintegerMost losses in a row (pushes don't break it).
seriesobjectnullableBet-indexed lines of the graded bets (detail full): index the bet numbers kept (past 600 bets thinned by LTTB, the deepest drawdown always kept), dates, units, drawdown, base_units (the market base's ROI at the same stakes, null unless measured), band_p025 and band_p975 (the no-skill range: 1,000 seeded runs of the same bets at their fair chances; null under 20 bets or for racing), rolling_roi_100 with its no-skill range rolling_p025 and rolling_p975, and markers (season starts, the holdout).
xstringAlways bet_index.
One of: bet_index (open: new values may be added).
indexarray of integerBet numbers (1 is the first graded bet).
datesarray of stringEach kept bet's date (YYYY-MM-DD, UTC).
unitsarray of numberUnits to date.
drawdownarray of numberUnits below the running peak (0 or below).
base_unitsarray of numbernullableThe market base at the same stakes.
band_p025array of numbernullableNo-skill 2.5th percentile.
band_p975array of numbernullableNo-skill 97.5th percentile.
rolling_roi_100array of numberROI over the last 100 bets (null before).
rolling_p025array of numberIts no-skill 2.5th percentile.
rolling_p975array of numberIts no-skill 97.5th percentile.
markersarray of objectWhere each season and the holdout start.
typestringseason or holdout.
One of: season, holdout (open: new values may be added).
atintegerThe bet number it starts at.
labelstringnullableThe season, or newest 30%.
downsampledbooleanThe series were thinned to 600 points.
expected_max_drawdown_zero_edgenumbernullableThe median deepest drawdown of the no-skill runs, units.
holdoutobjectnullableThe latest 30% of bets against the earlier 70%.
seasonsarray of objectSeasons in calendar order.
splitsarray of objectnullableHome, away, favorite, underdog and line bands, in a fixed order; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
withoutarray of objectnullableOne row a top-level AND condition: the rule's record without it; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
p_value_vs_rulenumbernullableThe rule against the bets the condition removes (F19).
p_value_vs_rule_adjustednumbernullablep_value_vs_rule, Holm-adjusted over the conditions shown.
betsarray of objectEvery bet in date order (left out at detail summary).
keystring{game_id}:{team_id} (racing: {race_id}:{runner_id}).
starts_atstringKickoff (RFC 3339).
format date-time
seasonstringThe season.
labelstringDate, team and opponent.
priceobjectnullableThe price at the bet moment, decimal.
linenumbernullableThe bet's line as graded (the team's spread, the total).
resultstringnullablewin, loss or push (soccer quarter lines: half_win or half_loss); null until played.
close_probnumbernullableThe fair closing probability of the same side, for CLV.
units_to_datenumberUnits after this bet, in date order.
how_gradedstringnullableHow the bet was graded, in words.
unitsnumbernullableUnits the bet won or lost at the run's staking (racing: its exact profit or loss with dead heats, each way, Rule 4 and commission); null until played.
drawdown_to_datenumbernullableUnits to date less their running peak (0 or below); null until played.
sidestringnullablehome or away (spread, moneyline), over, under or draw.
marketstringThe market graded (the other total when fading a total).
bet_momentstringWhen the bet was taken: edition, day_before or close.
clv_probnumbernullableThe fair closing chance less the fair chance at the bet moment, when Kelly holds both.
samplestringnullableholdout for the latest 30% of graded bets by time, else in_sample; null with no holdout or until played.
One of: in_sample, holdout (open: new values may be added).
whyarray of objectWhy it matched: up to six conditions (F46).
clausestringThe condition as written.
valuenumber | boolean | stringnullableThe compared expression's number for a comparison (rest_days - opponent.rest_days for rest_days - opponent.rest_days >= 3), else true or false.
settlementarray of stringThe settlement rules that decided the bet (IDs in settlement.rules).
leak_testobjectThe leak test (F3): the rule and each part of it evaluated again on a sample of rows with the data cut at the bet moment. A leak refuses the test (422 rule_not_testable, problem leak_found), so a report always says clean or not_run.
statusstringrequiredclean, leak_found or not_run.
One of: clean, leak_found, not_run (open: new values may be added).
rows_checkedintegerrequiredRows rebuilt at the bet moment.
matched_checkedintegerOf them, rows the rule matched.
matches_flippedintegerRows whose match changed with the data cut.
partsarray of objectrequiredParts that changed, in the formula's order (empty when clean).
namestringThe part, as written (rule is the whole rule).
rows_changedintegerRows where it changed.
max_changenumbernullableThe largest change in a number.
examplestringOne changed row, {game_id}:{team_id}.
warm_upobjectThe warm-up check (F4): bets taken before the rule's windows had the earlier games they read (one season at a time; windows never reach into last season).
policystringThe policy graded in the report.
One of: as_written, require_full (open: new values may be added).
games_neededintegerEarlier games every window needs (0: the rule has no window).
windowsarray of objectEach window in the formula's order.
exprstringThe window, as written.
games_neededintegernullableEarlier games it reads (null: season to date).
short_rowsintegerRows in scope with fewer earlier games this season.
short_betsintegerBets the rule took with fewer.
short_betsintegerBets taken with a window short.
by_policyarray of objectnullableas_written, then require_full; null when the rule has no window.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
warningbooleanShort bets are over 10% of the bets, or leaving them out moves ROI by more than half its 95% interval's half-width.
settlementobjectHow bets were settled (F9): Kelly's own grading rules as data, and each rule's bets.
profilestringThe rules profile (kelly_standard).
rules_versionstringThe rules table's date.
appliedarray of objectRules that decided at least one bet, in table order.
rulestringRule ID.
betsintegerBets.
rulesarray of objectThe rules for this league and market.
idstringRule ID.
textstringThe rule in words.
effective_fromstringThe date Kelly's grader took it up.
coverageobjectCoverage (F10): from every game in scope to the bets graded, each season's priced share, blank inputs and whether the rows left out for no line differ from the bets kept.
flowarray of objectThe steps in order; rule_true rows = decisions + every row out after it.
stepstringin_scope, rule_true, one_bet_a_game, full_windows, priced, graded or decisions.
gamesintegerGames (in_scope only).
rowsintegerRows.
outobjectRows that left at this step, by reason.
by_seasonarray of objectEvery season asked for, in order.
seasonstringSeason.
gamesintegerGames in scope.
rowsintegerRows.
rule_trueintegerRows the rule matched.
betsintegerBets graded.
inputs_missingarray of objectInputs read bare on the row, with blanks.
attributestringThe input.
rows_blankintegerRows with no value.
left_out_differsobjectRows left out for no line against the bets kept: standardized mean difference over home, neutral site and rest days; flagged over 0.1 with 10 or more left out.
flagbooleanThey differ.
max_smdnumbernullableThe largest difference.
covariatestringnullableWhere it is.
left_outintegerRows.
keptintegerRows.
by_covariatearray of objectEach covariate's difference.
momentobjectThe bet moment's clock (F5): its target time, the prices' snapshot age at it and where the lines came from.
bet_momentstringedition, day_before or close.
targetstringnullableThe moment's time for a game, in words.
max_stale_minutesintegerOldest price kept, in minutes before the target.
floor_minutes_before_oct_2022integerThe limit's floor for games before October 2022 (the Odds API's 10-minute snapshots).
rows_pricedintegerRows the rule took that had a price.
rows_agedintegerRows whose price carries its snapshot time and was kept.
median_snapshot_age_minutesnumbernullableMedian minutes from snapshot to target, kept rows.
max_snapshot_age_minutesnumbernullableOldest kept snapshot, in minutes before the target.
stale_priceintegerRows left out: price older than max_stale_minutes.
after_momentintegerRows left out: price seen after the target.
age_unknownintegerRows kept whose price has no snapshot time (Kelly's own frozen lines).
line_sourcesobjectKept rows by line source (sheet_line, odds_api_early, odds_api_close).
left_outobjectGames the rule matched but didn't bet.
no_lineintegerNo Kelly line at the bet moment.
both_sidesintegerThe rule matched both teams.
stale_priceintegerThe price was older than max_stale_minutes at the bet moment, or seen after it (F5).
unpairedintegerTeasers: legs left off a ticket (a week with too few, or the leftover in chronological pairing).
no_priceintegerRacing: runners with no price to settle at.
non_runnersintegerRacing: non-runners (void).
void_racesintegerRacing: abandoned or void races.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
capsobjectnullableExposure caps applied in time order (F39); the record and checks are on the capped ledger.
skippedintegerBets the caps skipped.
by_reasonobjectSkipped bets by cap.
uncappedobjectThe record without the caps.
random_priority_mean_roinumbernullableMean ROI over random tie-breaks.
ruleobjectThe rule and the options it was tested with.
rule_hashstringThe rule's identity: SHA-256 of its canonical formula (spacing, case and the order of AND and OR conditions don't change it) with the competition, market, side, bet moment, seasons, season type, stake and price. Detail and tests are not part of it.
pattern ^sha256:[0-9a-f]{64}$
translation_idstringThe rules:translate result this formula came from (F43), when the request named one.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
readsarray of objectEvery name the rule reads, once.
namestringThe attribute, with its relation (opponent.rest_days).
knownstringIts moment: static, pregame, edition, close or final (a window's input is read before the game).
windowedbooleanA backward window reads it (earlier games only).
formula_versionstringThe formula language version the rule was read with (F45).
catalogue_versionstringnullableThe stat catalogue version its names were bound against.
reportobjectA shared report's own facts (get_rule_report only, F54).
report_idstringThe report.
report_urlstringIts page.
format uri
visibilitystringunlisted or public.
One of: unlisted, public (open: new values may be added).
author_rules_triedintegerThe author's rules tried on this market when shared.
frozenbooleanAlways true.
rerunobjectThe run's request, to test it again on today's data.
report_urlstringnullableA frozen report of this run anyone with the link can open (F54): only when the request asked (share) and the call has an account; else null. share_rule_run makes or changes one later.
format uri
chartsarray of objectOnly when the request asked (charts) at detail full: equity against the no-skill band, the drawdown and the rolling ROI as Kelly Chart Specs (F47).
transferobjectnullableOther markets and sports (F33): the same rule on its other market and on other leagues, numbers mapped to the same meaning, pooled with DerSimonian-Laird random effects. Evidence beside the verdict, never part of it; these runs don't count toward rules tried. Present only when the request asks for transfer.
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
whystringnullableWhy nothing ran, when nothing did.
optionsobjectThe leagues, markets and thresholds asked for.
cellsarray of objectEach market and league in a fixed order (the other market first, then sibling leagues), never sorted by a result; up to 12.
competitionstringThe league.
marketstringThe market.
thresholdsstringliteral (the rule's own numbers) or quantile (each number moved to the same percentile of its field in this league).
One of: literal, quantile (open: new values may be added).
formulastringThe formula this cell ran (quantile cells carry the moved numbers).
seasonsarray of stringSeasons read, as this league labels them.
mappedarray of objectnullableQuantile cells only.
numberstringThe number as written.
fieldstringWhat it is compared with.
percentilenumbernullableIts percentile in the rule's own league (a fraction).
valuenumbernullableThe number at the same percentile here.
decisionsintegerWins plus losses.
betsintegerGraded bets.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% score interval.
senumbernullableROI's standard error at 1 unit risked a bet.
p_valuenumbernullableThe cell's own one-sided p against no skill at its prices.
p_holmnumbernullableHolm's adjusted p across the cells.
pooledbooleanThe cell is in the pool (another league, one a league, 10+ decisions).
pooledobjectnullableThe pool of the other leagues' cells.
roinumbernullableRandom-effects ROI.
roi_ci95array of numbernullableIts 95% interval.
senumbernullableIts standard error.
p_valuenumbernullableTwo-sided p of the pooled ROI.
i2numbernullableI²: the share of the spread across leagues beyond chance (a fraction).
tau2numbernullableτ²: the between-league variance of true ROI.
qnumbernullableCochran's Q.
cellsintegerCells pooled.
same_directionobjectnullablePooled cells whose ROI has the rule's own sign.
agreeintegerCells the same way.
cellsintegerCells.
p_valuenumbernullableTwo-sided sign-test p.
textstring"4 of 5".
not_testablearray of objectLeagues and markets left out, with why.
competitionstringThe league.
marketstringnullableThe market.
missingarray of stringFields that league lacks.
whystringWhy.
notestringWhat the panel is and isn't.
run_idstringnullableThis run, kept on the account (run_…): the same request on the same data is the same run. GET /v1/rules/runs/{run_id} reads it back. Null without an account.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
data_versionstringThe data the run read: Kelly's lake snapshot (or the data's as-of time). A new snapshot is a new run of the same rule.
manifestobjectnullableWhat this run read and ran, so the same request can be run again and checked against it (ADR-K885). No prices, lines or rows of source data.
versionintegerThe manifest's own layout version.
codeobjectThe code that graded the run.
engineintegerThe tester's revision; it rises with any fix that can move a result on the same data.
releasestringnullableThe deployed commit.
formula_versionstringnullableThe formula language's version.
catalogue_versionstringnullableThe stat catalogue's version.
attributes_sha256stringnullableA hash of the attribute registry: what each input means and when it is known.
settlement_rulesstringnullableThe settlement rules' version.
dataobjectThe data the run read.
versionstringThe same value as data_version.
snapshotsarray of stringThe lake snapshots read, in order.
one_snapshotbooleanWhether every season was read on one snapshot.
as_ofstringnullableWhen the read was made (RFC 3339).
format date-time
data_throughobjectHow far Kelly's history went, by league.
latest_resultstringnullableThe newest game read that had a result.
rows_built_atstringnullableWhen the prepared rows were built, where the run read them.
format date-time
queriesobjectEach reviewed lake query the tester reads, with a hash of its text.
requestobjectThe request.
request_hashstringThe request's hash.
rule_hashstringThe rule's hash.
seedintegerThe seed every simulation in the run used.
sourcesobjectWhere the run's inputs came from.
pricesobjectBets priced, by the source of the line.
inputsarray of objectEach input the rule read, with the moment it is known.
namestringThe input.
knownstringnullableThe moment it is known.
seasonsarray of objectEach season read.
seasonstringThe season.
rowsintegerTeam-games in scope.
left_outobjectMatched games that were not bet, by reason.
sha256stringThe hash of this block. Two runs with the same hash read the same data with the same code.
since_runobjectnullableOnly on a run read back: what has changed in Kelly's tester since the run was kept.
statusstringcurrent: nothing that can move its numbers changed. superseded: something did. no_manifest: the run was kept before manifests.
One of: current, superseded, no_manifest (open: new values may be added).
engine_changedbooleannullableWhether the tester's revision has risen since.
changedarray of objectThe parts that differ, each with its value then (from) and now (to).
partstringengine, release, formula_language, catalogue, attributes, settlement_rules or queries.
tested_atstringWhen the run was kept (RFC 3339); only on a run read back.
format date-time
rule_versionobjectnullableThe saved rule this test was kept under (save, or rule_id): its rule_id, name, version and versions_tried. Null when the test wasn't saved.
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerrequiredThis rule's version under it: the same rule again keeps its version.
min 1
new_versionbooleanThis test made a new version.
versions_triedintegerrequiredVersions saved under it, this one included.
min 1
since_savedobjectnullableThe saved version's record since it was saved (backtester T6, F50), when the test names its rule_id and repeats that version: verdict v2's since_saved check reads it. Null for a new version (its clock starts now) or an unsaved test.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
tests_on_accountobjectnullableRules this account has tried on this competition and market, this one included (the rules-tried ledger: every distinct rule tested on the site, the API or Kelly MCP, kept for good). The not-luck check adjusts for the larger of this count and the request's own tests. Null when the call has no account.
competitionstringrequiredThe competition.
marketstringrequiredThe market.
countintegerrequiredDistinct rules tried.
min 1
shown_countedintegerVariants an earlier report showed (its condition-removed forms, nearby numbers and home or away splits) that joined the count because this rule is one of them (audit F6).
stakingstringHow each bet was staked.
matchedintegerBets taken.
ungradedintegerBets not yet played.
noticestringThis tests a rule on past games; it isn't a pick.
calibrationobjectnullableoutput: probability only. The formula's chances against the market's no-vig close of the same side (moneyline: the team wins; spread: it covers the bet's line), on the newest 30% of the scored team-games (whole days), and the bets it makes at the edge threshold.
formulastringThe probability formula as sent.
bet_formulastringThe rule it bets: the chance at the bet's own price worth more than the edge threshold (the rule tested, counted and kept).
edge_thresholdnumberExpected return a bet needs, p × decimal price − 1 (0.02 by default).
basisstringnewest_30: scored on the holdout only.
One of: newest_30 (open: new values may be added).
gamesintegerGames scored.
rowsintegerTeam-games scored (both teams of a game).
in_sample_rowsintegerTeam-games before the holdout, not scored.
holdout_fromstringnullableThe holdout's first day.
left_outobjectTeam-games not scored.
pushesintegerPushes and half results.
outside_0_1integerValues below 0 or above 1.
no_valueintegerRows the formula gave no number for.
clippedintegerValues of 0 or 1 clipped to 1e-6 from the edge.
briernumbernullableBrier score of the formula.
brier_marketnumbernullableBrier score of the close.
brier_skillnumbernullable1 − formula / close (above 0: the formula scores better).
log_lossnumbernullableLog loss of the formula.
log_loss_marketnumbernullableLog loss of the close.
log_loss_skillnumbernullable1 − formula / close.
dm_p_valuenumbernullableDiebold–Mariano p on the per-game log-loss difference (two-sided).
dm_statisticnumbernullableThe HLN-corrected statistic (positive: the formula's loss is higher).
mcbnumbernullableCORP miscalibration (Brier).
dscnumbernullableCORP discrimination.
uncnumbernullableUncertainty; brier = mcb − dsc + unc.
mcb_p_valuenumbernullableThe MCB test: the chance a calibrated formula looks this miscalibrated (999 resamples).
reliabilityarray of objectnullableThe CORP reliability curve at up to 20 forecast values.
pnumbernullableThe formula's probability.
observednumbernullableHow often it happened (PAV fit).
lonumbernullable90% consistency band, low.
hinumbernullable90% consistency band, high.
ecenumbernullableExpected calibration error over ten equal-width bins (bin-dependent; for familiarity).
distinct_forecastsintegerDistinct values the formula gave.
slopenumbernullableCalibration slope b in y ~ a + b·logit p (1 is ideal; under 1 overconfident).
slope_senumbernullableIts game-clustered standard error.
interceptnumbernullableCalibration intercept a (0 is ideal).
encompassingobjectnullabley ~ a + b_q·logit q + b_p·logit p, game-clustered: does the formula know something the close doesn't?
b_modelnumbernullableb_p.
b_marketnumbernullableb_q.
se_modelnumbernullableb_p's standard error.
p_valuenumbernullableTwo-sided p of b_p.
edge_binsarray of objectnullableRealized results against the formula's edge.
edgenumbernullableMean of p − q in the bin.
realized_minus_marketnumbernullableMean of y − q in the bin.
rowsintegerTeam-games in the bin.
statusstringThe calibrated check: pass, fail or na (under 200 games, or one value).
One of: pass, fail, na (open: new values may be added).
valuestringThe check's line.
statisticnumbernullableThe encompassing p.
thresholdnumbernullable0.05.
testsintegerrequiredRules tried behind the picked setting's adjusted p (every setting included).
tests_on_accountobjectnullableThe account's rules tried on this league and market.
competitionstringLeague.
marketstringMarket.
countintegerRules tried, every setting of this sweep included.
data_versionstringnullableThe lake snapshot the picked setting's test read.
run_idstringnullableThe picked setting's kept run (get_rule_run), with an account.
noticestringWhat this is and isn't.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/rules:sweep
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=home",
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"type": "https://docs.askkelly.ai/problems/rule_not_testable",
"title": "Rule can't be tested",
"status": 422,
"code": "rule_not_testable",
"detail": "The formula has no slot to sweep.",
"param": "formula",
"doc_url": "https://docs.askkelly.ai/problems/rule_not_testable",
"request_id": "req_c691351597e5cf2c7f567da2",
"retryable": false,
"display_message": "That rule can't be tested on past games as written.",
"problems": [
{
"code": "no_sweep",
"path": null,
"message": "The formula has no slot to sweep.",
"fixes": [
"Put a range where a number goes, e.g. rest_days >= {1..6}."
]
}
]
}POST https://api.askkelly.ai/v1/rules:sweep
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"formula": "=AND(home, spread_close <= {-6..6 step 3})",
"competition": "nba",
"seasons": [
"2025-26"
]
}
{
"object": "rule_sweep",
"rule": {
"formula": "=AND(home, spread_close <= {-6..6 step 3})",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus"
},
"headline_roi": 0.9099999999999999,
"sweep": {
"slots": [
{
"name": "a",
"written": "{-6..6 step 3}",
"values": [
-6,
-3,
0,
3,
6
]
}
],
"variants": 5,
"grid": [
{
"setting": {
"a": -6
},
"formula": "=AND(home, spread_close <= -6)",
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
{
"setting": {
"a": -3
},
"formula": "=AND(home, spread_close <= -3)",
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.17005720935900537
},
{
"setting": {
"a": 0
},
"formula": "=AND(home, spread_close <= 0)",
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.17005720935900537
},
{
"setting": {
"a": 3
},
"formula": "=AND(home, spread_close <= 3)",
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.17005720935900537
},
{
"setting": {
"a": 6
},
"formula": "=AND(home, spread_close <= 6)",
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.17005720935900537
}
],
"spp_median_roi": 0.9099999999999999,
"roi_p05": 0.9099999999999999,
"roi_p95": 0.9099999999999999,
"share_profitable": 1,
"pick_rule": "plateau",
"min_decisions_to_pick": 30,
"fast_path": true,
"best": null,
"picked": null,
"pbo": null,
"blocks": null,
"splits": null,
"degradation_slope": null,
"prob_oos_loss": null,
"logit_median": null,
"spa_p_value": null,
"reality_check_p": null,
"p_adjusted_picked": null,
"k_effective": null,
"deflated_roi": null,
"expected_max_roi": null,
"selection": {
"from": "2025-10-21",
"to": "2025-10-21",
"bets": 1
},
"forward": null,
"na": "no setting has (30) decisions before the sealed window",
"tests_added": 5
},
"walk_forward": null,
"cross_validation": null,
"picked_test": null,
"tests_on_account": {
"competition": "nba",
"market": "spread",
"count": 10
},
"tests": 10,
"data_version": null,
"run_id": null,
"as_of": "2026-10-04T17:30:00Z",
"assumptions": [
"Every setting is bet at the same prices on the same games; the newest 30% of the bets (by day) are sealed and shown once, for the picked setting only.",
"The headline is the median ROI across all (5) settings; settings with under (30) decisions before the sealed window count but can't be picked.",
"Every setting counts as a rule tried (5); a setting tried before isn't counted again.",
"Walk-forward and cross-validation choose the setting again inside each training window, on every season.",
"Settings are graded as written (warm-up and exposure caps apply to the picked setting's full test)."
],
"notice": "This tests settings of your rule on past games. It isn't a pick and doesn't rank selections; the headline is the typical setting's result, not the best one's.",
"summary": "(5) settings against the spread, 2025-26: no setting has (30) decisions before the sealed window; the typical setting made +91.0% ROI.",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_0a9085dd03633c71481499f8",
"live_topics": [],
"rg_line": null
}GET /v1/rules/runs/{run_id}/export
public, no key needed · 0 credits, not charged during the beta · MCP tool export_rule_run · status live
One of your kept rule tests' bets as CSV or JSON Lines text: the game, side, start, season, result, units and units to date. Lines and prices are never in a file (the data licences). The same run gives the same bytes.
run_idstring · pathrequiredThe run_id a test_rule result gave (run_…).
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
formatstring · querycsv (default) or jsonl.
One of: csv, jsonl (open: new values may be added).
200 The rule export facts.
objectstringrequiredrule_export.
One of: rule_export (open: new values may be added).
run_idstringrequiredThe run.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
formatstringrequiredcsv (RFC 4180, a header row, CRLF) or jsonl (one bet a line).
One of: csv, jsonl (open: new values may be added).
rowsintegerrequiredBets in the file.
columnsarray of stringrequiredThe file's columns, in order.
omittedarray of stringrequiredColumns never written to a file (licensed: lines and prices).
rule_hashstringnullableThe rule's identity.
data_versionstringThe data the run read.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
tested_atstringWhen the run was kept (RFC 3339).
format date-time
filenamestringA file name for it.
sha256stringThe content's SHA-256.
pattern ^sha256:[0-9a-f]{64}$
contentstringrequiredThe file's text (UTF-8). A text cell a spreadsheet would read as a formula starts with an apostrophe.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/runs/run_mkytv75qe5a297z7y3af/export?format=csv
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_export",
"run_id": "run_mkytv75qe5a297z7y3af",
"format": "csv",
"rows": 1,
"columns": [
"game_key",
"game_id",
"team_id",
"starts_at",
"season",
"market",
"bet_moment",
"result",
"units",
"units_to_date"
],
"omitted": [
"line",
"price",
"close_line",
"close_price",
"close_prob",
"book"
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"data_version": "snap_52",
"as_of": "2026-11-06T00:00:00Z",
"tested_at": "2026-10-04T17:30:00Z",
"filename": "kelly-run_mkytv75qe5a297z7y3af.csv",
"sha256": "sha256:f11c63caf507c6bd6f20f128aabc88b5c7f09d39c31d2978a5219e3481007cf3",
"content": "game_key,game_id,team_id,starts_at,season,market,bet_moment,result,units,units_to_date\r\ngm_1hnx5nrmk4xjpz6fzyfz:tm_b4w5y9xy5f9sgg1cmkvy,gm_1hnx5nrmk4xjpz6fzyfz,tm_b4w5y9xy5f9sgg1cmkvy,2025-10-21T23:30:00Z,2025-26,spread,close,win,0.91,0.91\r\n",
"summary": "(1) bets from run run_mkytv75qe5a297z7y3af as CSV: the game, side, result and units; lines and prices are left out (the data licences).",
"assumptions": [
"Lines and prices are never in an export file."
],
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_aefedea15a9544fda2c60fc7",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/runs/run_zzzzzzzzzzzzzzzzzzzz/export
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No run run_zzzzzzzzzzzzzzzzzzzz on this account.",
"param": "run_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_8af022b41b7d69af3cec4ec5",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules:compare
public, no key needed · 0 credits, not charged during the beta · MCP tool compare_rules · status live
Two to six of your kept rule tests on common ground: the bets they share, opposite sides on the same game, the paired ROI difference with its 95% range (a bootstrap over game days) and Holm-adjusted p against the reference, how their days move together, and what betting them all would have done. Runs on different settings say so and aren't tested against each other.
Body, application/json (required):
runsarray of stringTwo to six run_ids (run_…), in the order to show them.
min items 2 · max items 6
referenceintegerThe index in runs the others are tested against (default 0).
min 0 · max 5
200 The rule compare facts.
objectstringrequiredrule_compare.
One of: rule_compare (open: new values may be added).
alignedbooleanrequiredSame competition, market, bet moment, seasons, season type, price and stake.
differs_onarray of stringThe settings that differ (empty when aligned).
settingsobjectnullableThe shared settings.
referenceintegerThe run the others are tested against.
runsarray of objectrequiredThe runs, as asked.
run_idstringThe run.
formulastringnullableIts rule.
sidestringnullablerule or fade.
betsintegerBets it took.
recordstringnullableW-L-P.
unitsnumbernullableUnits.
roinumbernullableROI, a fraction.
honest_roinumbernullableIts honest forward ROI (F22).
verdict_levelstringnullableIts verdict's level.
data_versionstringThe data it read.
tested_atstringWhen it was kept.
format date-time
pairsarray of objectrequiredEach other run against the reference.
aintegerThe reference's index in runs.
bintegerThe other run's index.
jaccardnumbernullableShared bets over the bets either took.
conflictsintegerGames where they bet opposite sides.
roi_diffnumbernullableThe reference's ROI less the other's (units per graded bet); null when not aligned.
roi_diff_ci95array of numbernullableIts 95% range: a paired bootstrap over game days (2,000 draws).
p_valuenumbernullableTwo-sided bootstrap p for no difference.
p_holmnumbernullableHolm-adjusted across the pairs.
betterstringnullablea or b when its range excludes 0 and p_holm < 0.05; else null.
One of: a, b (open: new values may be added).
daily_pnl_corrnumbernullablePearson correlation of their units per game day (null under 3 days).
combinedobjectrequiredBetting every run's bets together: a shared bet once, opposite sides netted to none.
betsintegerBets.
decisionsintegerDecided bets.
recordstringW-L-P.
unitsnumberUnits.
roinumbernullableROI.
max_drawdown_unitsnumbernullableDeepest drawdown (null for runs kept before T5).
conflicts_nettedintegerGames netted to no bet.
same_databooleanEvery run read the same data.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · not_found
POST https://api.askkelly.ai/v1/rules:compare
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"runs": [
"run_mkytv75qe5a297z7y3af",
"run_ewabjhkycqv5zk6mvbkn"
]
}
{
"object": "rule_compare",
"aligned": true,
"differs_on": [],
"settings": {
"competition": "nba",
"market": "spread",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"price": "consensus",
"stake": "risk"
},
"reference": 0,
"runs": [
{
"run_id": "run_mkytv75qe5a297z7y3af",
"formula": "=home",
"side": "rule",
"bets": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"honest_roi": -0.04514385341429974,
"verdict_level": "too_few",
"data_version": "snap_52",
"tested_at": "2026-10-04T17:30:00Z"
},
{
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"formula": "=AND(home, rest_days >= 1)",
"side": "rule",
"bets": 0,
"record": "0-0",
"units": 0,
"roi": null,
"honest_roi": null,
"verdict_level": "too_few",
"data_version": "snap_52",
"tested_at": "2026-10-04T17:30:00Z"
}
],
"pairs": [
{
"a": 0,
"b": 1,
"shared_bets": 0,
"jaccard": 0,
"conflicts": 0,
"roi_diff": null,
"roi_diff_ci95": null,
"p_value": null,
"daily_pnl_corr": null,
"p_holm": null,
"better": null
}
],
"combined": {
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.91,
"roi": 0.91,
"max_drawdown_units": 0,
"conflicts_netted": 0
},
"same_data": true,
"summary": "(2) runs compared: run 2 shares (0) bets with run 1. No run's ROI is clearly different from the reference's.",
"assumptions": [
"The range on each ROI difference resamples whole game days (2,000 draws); ROI here is units per graded bet at one unit risked."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_00b063f1e4edf67cd34d2576",
"live_topics": [],
"rg_line": null
}POST https://api.askkelly.ai/v1/rules:compare
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"runs": [
"run_mkytv75qe5a297z7y3af",
"run_zzzzzzzzzzzzzzzzzzzz"
]
}
{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No run run_zzzzzzzzzzzzzzzzzzzz on this account.",
"param": "run_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_023c15582310f2b78f834165",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules:share
public, no key needed · 0 credits, not charged during the beta · MCP tool share_rule_run · status live
A report link for one of your kept rule tests: the run frozen as it was tested, with your rules tried on its market, for anyone with the link (unlisted), anyone (public), or no one (private: the link reads as not found). Sharing again changes only the visibility.
Body, application/json (required):
200 The rule report link facts.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · not_found
POST https://api.askkelly.ai/v1/rules:share
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"run_id": "run_mkytv75qe5a297z7y3af",
"visibility": "unlisted"
}
{
"object": "rule_report_link",
"report_id": "rpt_8kqqvwd602ere4fy0nv6",
"report_url": "https://askkelly.ai/rule-tester/r/rpt_8kqqvwd602ere4fy0nv6",
"run_id": "run_mkytv75qe5a297z7y3af",
"visibility": "unlisted",
"author_rules_tried": 2,
"shared_at": "2026-10-04T17:30:00Z",
"tested_at": "2026-10-04T17:30:00Z",
"data_version": "snap_52",
"frozen": true,
"summary": "Report link for run run_mkytv75qe5a297z7y3af: anyone with the link can open it. It shows the run as it was tested, with your (2) rules tried on this market.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_dbba199df7186c2aa793e858",
"live_topics": [],
"rg_line": null
}GET /v1/rules/reports/{report_id}
public, no key needed · 0 credits, not charged during the beta · MCP tool get_rule_report · status live
A rule test someone shared: the run exactly as it was tested (its data and as-of time), the author's rules tried on that market, the bets by game and side (never a price) and its charts, with the request to test it again on today's data. A private or unknown report is not_found.
report_idstring · pathrequiredThe report's ID (rpt_…).
pattern ^rpt_[0-9a-hjkmnp-tv-z]{20}$
200 The rule test facts.
objectconstrequiredAlways rule_test.
Always "rule_test".
verdictobjectrequiredVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
overallobjectrequiredEvery graded bet.
recordstringWins-losses(-pushes).
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
betsintegerGraded bets.
decisionsintegerWins plus losses.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_intervalarray of numbernullableROI's 95% interval at 1 unit risked a bet.
win_ratenumbernullableWins over decisions.
break_evennumbernullableThe win rate that breaks even at these prices.
significanceobjectnullableThe record test (kmath.records.record_test) and verdict v2's statistics (kmath.significance): the p-value, adjusted for the rules tried, and the anytime-valid test.
p_valuenumbernullableThe larger of the exact and the clustered one-sided p against no skill at these prices (F12, F23).
methodstringnullableexact_binomial, exact_simulation, saddlepoint or z_score.
znumbernullableThe score z: units over the square root of Σ(d − 1).
t_statnumbernullableThe t statistic with the sample variance.
p_value_exactnumbernullableThe exact (or z) tail alone.
p_value_clusterednumbernullableThe tail with bets clustered by slate and team-season (F23).
p_value_vs_vignumbernullableAgainst a random bet at the no-vig close (needs closes).
p_adjustednumbernullableThe p-value adjusted for the rules tried (Romano-Wolf on the ledger's rules, Bonferroni on the count otherwise; F14).
adjust_methodstringnullableromano_wolf, holm or none.
testsintegernullableRules tried in this family, this one included.
tests_effectivenumbernullableEffective tests (Li-Ji): near-duplicate rules count as fewer.
reality_check_pnumbernullableWhite's Reality Check p (when the ledger holds the rules' units).
spa_pnumbernullableHansen's SPA p (when the ledger holds the rules' units).
clears_t3booleannullablez at or above 3 (Harvey-Liu-Zhu's bar).
haircutnumbernullableHarvey-Liu haircut: the share of the ROI the rules tried take away.
haircut_roinumbernullableROI after the haircut, per decision.
anytime_validobjectnullableThe record as a bet against no skill (F17): e-value, anytime p, ROI range.
e_valuenumbernullableThe e-value now.
p_valuenumbernullable1 over the largest e-value so far.
max_enumbernullableThe largest e-value along the path.
roi_cs95array of numbernullableAn ROI range valid at every bet.
roi_ci95array of numbernullableROI's 95% score interval, clustered (F13, F23).
units_ci95array of numbernullableUnits' 95% interval: decisions times the ROI interval.
roi_ciobjectnullableROI intervals at ci_level: score (the one shown), BCa, block bootstrap over slates and anytime-valid (F13).
deflatedobjectnullablePer-bet Sharpe, PSR and DSR at the effective tests, and the bets needed: deflated (sample_needed), single test and 80% power, each times the design effect (F15).
sample_neededintegernullableDecisions needed, deflated for the rules tried; null with no edge to prove.
sample_needed_singleintegernullableDecisions needed for one rule tried.
sample_needed_power80integernullableDecisions for 80% power at this ROI.
sample_needed_planintegernullableDecisions for 80% power at plan_roi.
dsrnumbernullableDeflated Sharpe ratio.
psrnumbernullableProbabilistic Sharpe ratio.
sample_neededobjectnullableBets needed to tell this win rate from break-even (one-sided, with 80% power, ...).
small_samplebooleanUnder 30 decisions.
few_decisionsbooleanUnder 5 decisions.
honestobjectnullableThe honest forward ROI (F22, the headline ROI): empirical Bayes toward a prior mean (the rule's own closing-line value when at least half its bets have a close, else the usual two-way margin), less the winner's curse for the rules tried and the out-of-sample decay.
roinumbernullableHonest forward ROI.
ci80array of numbernullable80% range.
raw_roinumbernullableThe past ROI per decision, beside it.
prior_meannumbernullableThe prior mean the ROI is shrunk toward (prior.source says which).
prior_sdnumbernullableThe spread of true ROIs (0.02 until fitted on the ledger).
shrinknumbernullableWeight on the prior, se²/(se² + τ²).
selection_adjnumbernullableWinner's-curse correction for the rules tried.
decay_factornumbernullableOut-of-sample decay on the excess (0.74).
stabilizes_at_betsintegernullableDecisions at which the record and the prior weigh the same.
uses_clvbooleanThe prior mean is the rule's own closing-line value.
prior_basisstringfitted (prior_sd from the ledger's fit) or seed.
priorobjectWhere the prior mean comes from.
sourcestringclv (the rule's mean closing-line value), margin (the two-way margin, -4.55%) or fitted.
meannumbernullableThe prior mean used.
closesintegerDecided bets with a no-vig close.
cappedbooleanA closing-line value above 0 was held at 0: under 50 bets have a close.
wordsstringThe prior, said plainly.
dependenceobjectnullableBets clustered by slate (the date; the week in the NFL and college football) and team-season (F23).
clustersintegerSlates.
design_effectnumberClustered over iid variance.
effective_betsintegerDecisions over the design effect (at least 1).
methodstringcr1_twoway, cr1_slate, cr1_larger_oneway, wild_bootstrap or wild_bootstrap_webb.
baseobjectnullableThe rule against the rest of its own base (F19): the who-or-when conditions alone, or base_formula.
formulastringnullableThe base, or null for the whole market.
sourcestringauto, user or market.
One of: auto, user, market (open: new values may be added).
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
roinumbernullableThe base's ROI per decision.
liftnumbernullableRule ROI less base ROI.
p_value_vs_basenumbernullableSeason-stratified permutation p of the rule within its base.
units_from_basenumbernullableUnits the base alone would have made on the rule's stakes.
units_from_filternumbernullableThe rule's units less units_from_base.
clvobjectrequiredClosing-line value as evidence (F21): each bet's no-vig CLV, q_close x price - 1, on bets Kelly holds a close for. Empty when betting at the close.
bets_with_closeintegerBets with a close.
bets_without_closeintegerBets without one.
referencestringThe close used: consensus (the no-vig average across books).
devigstringHow the vig was removed: multiplicative.
lines_convertedintegerBets whose close was carried from the closing line to the bet's line (NFL spreads and totals).
coveragenumbernullableShare of bets with a close probability (null with no bets).
closes_staleintegerCloses left out as older than 60 minutes at the start, or seen after it (audit F9): those bets keep their result but have no CLV.
points_coveragenumbernullableShare of bets with points beaten (spreads and totals).
points_testobjectnullableThe clustered test on points beaten (audit F3).
betsintegerBets with points beaten.
meannumbernullableMean points the bet's line beat the close by.
std_error_clusterednumbernullableStandard error clustered by slate and team-season.
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean points beaten > 0.
cluster_methodstringHow the standard error was clustered.
basisstringWhat Beats the close is judged on: probability, or points when under 80% of bets have a close probability and more have points beaten (audit F3).
One of: probability, points (open: new values may be added).
meannumbernullableMean CLV (fraction): the bets' expected ROI if the close is right.
ci95array of numbernullableClustered 95% interval on the mean.
min items 2 · max items 2
p_valuenumbernullableOne-sided clustered p of mean CLV > 0.
beat_ratenumbernullableShare of bets that beat the close, ties left out.
beat_rate_ci95array of numbernullableWilson 95% interval on beat_rate.
min items 2 · max items 2
sign_test_pnumbernullableExact one-sided sign test against half.
mean_probnumbernullableMean of the close's no-vig probability less the bet moment's.
mean_pointsnumbernullableMean points the bet's line beat the closing line by (spreads and totals).
points_betsintegerBets mean_points is over.
profit_slopenumbernullablePer-bet units regressed on CLV: near 1 when CLV turned into profit.
profit_slope_senumbernullableIts slate-clustered standard error.
sample_neededintegernullableBets CLV needs for a 95% interval to clear 0 at this mean and spread (null unless the mean is above 0).
sample_needed_roiintegernullableBets profit would need at the same edge.
significanceobjectnullableThe CLV mean's tests (null under two bets).
tnumbernullableMean over its iid standard error.
ci95array of numbernullableStudent 95% interval (iid).
min items 2 · max items 2
std_errornumbernullableiid standard error.
bets_needednumbernullableBets a t of 2 needs at this mean and spread.
std_error_clusterednumbernullableStandard error clustered by slate and team-season (F23).
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean CLV > 0 (wild cluster bootstrap under 42 slates, else Student t); beats_close reads it.
cluster_methodstringHow the standard error was clustered.
design_effectnumbernullableClustered variance over iid variance.
notestringWhy the block is empty, when it is.
luckobjectThe luck-adjusted result (F20): the units the market's fair probabilities expected the decided bets to make against what they made. Pushes count zero both ways.
readingstringluck (bets before the close: the gap is outcome noise) or versus_the_market (bets at the close: the gap is the rule's edge on the closing market).
One of: luck, versus_the_market (open: new values may be added).
close_sourcestringconsensus.
devigstringmultiplicative.
betsintegerDecided bets with a fair probability.
bets_without_probintegerDecided bets without one.
left_outobjectBets left out.
pushintegerPushes.
halfintegerHalf results (quarter lines).
lines_convertedintegerBets whose close was carried to the bet's line.
expected_unitsnumbernullableSum of stake x (q x (price - 1) - (1 - q)).
actual_unitsnumbernullableUnits the bets made.
luck_unitsnumbernullableactual_units less expected_units.
luck_se_unitsnumbernullableluck's standard error given the probabilities.
expected_winsnumbernullableSum of q.
actual_winsintegernullableWins.
luck_percentilenumbernullableWhere the wins sit among the market's (Poisson-binomial, mid-p): 0.95 is luckier than (95) of (100) identical paths.
expected_roinumbernullableexpected_units over units risked.
actual_roinumbernullableactual_units over units risked.
thetanumbernullablePer-bet units regressed on expected units (30 bets or more): near 1 when the close is calibrated on these bets.
by_seasonarray of objectEach season, in calendar order; luck_units add up to the total.
seasonstringSeason.
betsintegerDecided bets with a probability.
expected_unitsnumberUnits the market expected.
actual_unitsnumberUnits made.
luck_unitsnumberactual_units less expected_units.
capacityobjectnullableF8: the limits Kelly measured at the bet moment (ADR-K783); evidence beside the verdict, never in it. No book publishes its limits, so only measured ones are shown.
availablebooleanWhether measured limits are shown for this market and moment.
marketstringnullableThe market as line history names it (spreads, totals, h2h).
before_startstringnullableThe time before the start the bet moment falls in.
One of: 0_1h, 1_12h, 12_36h, 36_72h, 72h_plus (open: new values may be added).
booksarray of objectPer book, highest median first.
bookstringnullableThe book as the vendor names it.
vendorstringnullableWho measured it.
currencystringnullableThe stake currency, null when the vendor doesn't say.
mediannumbernullableThe median limit.
p25numbernullableIts 25th percentile.
p75numbernullableIts 75th percentile.
minnumbernullableThe smallest.
maxnumbernullableThe largest.
snapshotsintegerLimit readings behind it.
gamesintegerGames behind it.
first_seenstringnullableThe first reading's date.
last_seenstringnullableThe last reading's date.
max_stake_per_betobjectnullableThe highest book median: the most one book typically took on one bet.
bookstringnullableThe book.
amountnumbernullableThe median limit.
currencystringnullableIts currency.
per_season_at_limitnumbernullablemax_stake_per_bet times the rule's bets a season.
bets_per_seasonnumbernullableThe rule's bets a season.
limits_heldintegerLimit readings Kelly holds for this market and moment.
reasonstringnullableWhy none are shown, when available is false.
stressobjectrequiredThe same bets graded harder.
half_point_worseobjectnullableEvery line half a point worse (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
flat_110objectnullableEvery bet at -110 (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
other_sideobjectThe other side of every bet.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
break_evenobjectnullableF7: how much worse the price or line can get before the profit is gone.
price_multipliernumbernullableThe share of each payout (d - 1) that breaks even (under 1: the room to spare).
american_avgnumbernullableThe flat American price that breaks even.
decimal_avgnumbernullableThe same, decimal.
line_pointsnumbernullableThe most points worse (in halves, up to 3) the line could be with a profit left; 0 when none; null for moneylines.
gridarray of objectnullableF7: the bets at -105 to -120 with the line 0 to 1 point worse (moneylines: payouts cut 0-6%); left out at detail summary.
vigintegernullableEvery bet at this American price.
line_shiftnumbernullablePoints the line moved (negative: worse).
price_multipliernumbernullableMoneylines: the share of each payout kept.
unitsnumbernullableUnits.
roinumbernullableROI.
slippage_mcobjectnullableF7: each bet's line worsened by a draw from the market's own Monday-to-close moves, 1,000 seeded runs; null without such moves.
runsintegerRuns.
roi_p05numbernullable5th percentile ROI.
roi_p50numbernullableMedian ROI.
moves_fromintegerLine moves the draws come from.
move_mediannumbernullableTheir median size, points.
price_worseobjectnullableTeasers: every ticket ten cents worse (price_american). Racing: every bet's winnings 5% smaller.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
legsobjectnullableTeasers: every teased leg, judged against the per-leg break-even (1/D)^(1/n).
recordstringLegs won-lost(-pushed).
decisionsintegerLegs won plus lost.
cover_ratenumbernullableLegs won over legs decided.
break_even_per_legnumberThe cover rate a leg needs at the ticket's price.
significanceobjectnullableThe legs' record test against the per-leg break-even.
sample_neededobjectnullableLegs needed to tell this cover rate from the break-even.
random_baselineobjectnullableF18: the rule against random bets of the same count on the same games, matched by season and price bucket.
simsintegerRandom rules drawn.
betsintegerThe rule's bets.
poolintegerBets the market offered.
percentilenumbernullableShare of random rules with a lower ROI.
beats_per_100integerRandom rules out of 100 the rule beats.
p_valuenumbernullable(1 + random rules at or above) / (sims + 1).
analytic_pnumbernullableThe finite-population normal cross-check.
random_roi_mediannumbernullableMedian random ROI (about minus the vig).
random_roi_p05numbernullable5th percentile random ROI.
random_roi_p95numbernullable95th percentile.
best_of_testsobjectWould the best of M random rules (M = rules tried) do this well?
testsintegerM.
p_valuenumbernullable1 - F(ROI)^M.
histogramobjectnullableRandom ROIs in 30 bins (edges, counts); full detail only.
decayobjectnullableF24: is the rule still making money? Rolling ROI and CLV, one unknown break (sup-F, block permutation p), a CUSUM alarm, a known-date test, the trend and the decay ratio.
betsintegerGraded bets.
windowintegernullableRolling window, bets.
rollingarray of objectnullableUp to 200 points; full detail only.
breakobjectnullableThe most likely date the edge changed.
atstringDate.
bet_indexintegerBets before it.
sup_fnumbernullablesup-F.
p_valuenumbernullableBlock permutation p-value.
at_edgebooleanWithin the first or last 15%: not a confirmed break.
roi_beforenumbernullableROI before.
roi_afternumbernullableROI after.
bets_afterintegerBets after.
cusumobjectnullablePage's lower CUSUM alarm.
known_breakobjectnullableA Chow test at break_date.
trend_per_100_betsnumbernullableProfit per bet's change per 100 bets (1 unit risked).
decay_rationumbernullableROI after the break (or the newest 30%) over ROI before.
clv_breakobjectnullableThe break test on CLV.
roi_newest_30numbernullableROI of the newest 30% of bets.
ablationobjectnullableF27: each top-level AND condition's Shapley share of the units beyond the base's ROI.
baseobjectEvery game with a line: decisions and ROI.
totalnumbernullableThe rule's units beyond the base's ROI (the Shapley values add up to it).
clausesarray of objectOne a condition, in the order written.
conditionstringThe condition.
shapley_unitsnumbernullableIts share of the extra units.
labelstringdead, harmful, load_bearing or helps.
One of: dead, harmful, load_bearing, helps (open: new values may be added).
fixobjectnullableRemove it: the rule without it.
pairsarray of objectPairs (up to 6 conditions).
methodstringexact or sampled.
One of: exact, sampled (open: new values may be added).
nearbyobjectnullableF28: every number in the rule moved two steps each way (steps fixed by the field).
numbersarray of objectThe rule's numbers.
variantsintegerVariants in the grid.
distinct_bet_setsintegerVariants with their own bets.
roi_mediannumbernullableMedian ROI of the variants.
roi_p05numbernullable5th percentile.
roi_p95numbernullable95th percentile.
spike_rationumbernullable(ROI - median) / IQR of the variants.
readingstringnullableplateau, slope or spike.
nastringnullableWhy there are no numbers to move.
monte_carloobjectnullableF31: drawdowns and losing runs the same bets give in other orders.
simsintegerSimulations.
modestringblock, iid or permutation.
max_drawdown_unitsobjectMax drawdown, units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
longest_losing_runobjectLongest losing run, bets.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
final_unitsobjectFinal units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
prob_below_zeronumbernullableShare of runs finishing below zero.
worst_100_bets_units_p05numbernullable5th percentile of the worst 100-bet stretch.
noiseobjectnullableF32: jittered inputs, lines moved half a point either way, and an earlier bet moment.
inputsobjectnullableThe rule's compared numbers jittered.
runsintegerRuns.
profitable_runsintegerRuns with a profit.
roi_mediannumbernullableMedian ROI.
roi_p05numbernullable5th percentile ROI.
overlap_mediannumbernullableMedian Jaccard overlap of the bets with the rule's.
linesobjectnullableLines (moneylines: payouts) moved at random.
delayobjectnullableThe rule bet at the earlier moment.
concentrationobjectnullableF34: the profit without the best season, team, opponent, month and weekday.
byarray of objectGroupings in a fixed order.
groupstringseason, team, opponent, month or weekday.
bestobjectnullableThe best group (3+ bets).
without_bestobjectnullableThe rule without it.
effective_groupsnumbernullable1 / Herfindahl of the profitable groups' shares.
p_more_concentratednumbernullablePermutation p-value (teams and opponents).
trim_top_0_05objectnullableWithout the top 5% of bets (moneylines) or bet-days.
split_dimensionsarray of objectnullableF35: the rule by fixed dimensions with shrunk ROI and intervals.
dimensionstringThe dimension.
heterogeneity_pnumbernullableChi-square p-value of win rates across its levels.
heterogeneity_qnumbernullableBenjamini-Hochberg q across the dimensions.
real_differencebooleanq < 0.10.
where_profitstringnullableThe level with the most shrunk extra units, when real.
cellsarray of objectnullableLevels in a fixed order; full detail only.
robustnessobjectThe suite's run: the verdict v2 checks it feeds and its seed.
checksarray of objecthalf_point_worse, nearby_holds, not_concentrated and still_working.
keystringrequiredThe check.
One of: half_point_worse, nearby_holds, not_concentrated, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na.
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumbernullableThe figure the status compares.
thresholdnumbernullableWhat it is compared with.
fast_pathbooleanThe rule's numbers were re-bet from its compiled columns.
seedintegerThe seed (a hash of the rule and its options).
evaluationsintegerFormula evaluations the suite ran.
offarray of stringBlocks switched off.
pathobjectThe units path in date order: drawdown and longest losing run; at detail full its series (F47).
max_drawdown_unitsnumberThe deepest fall from a peak, units.
longest_losing_runintegerMost losses in a row (pushes don't break it).
seriesobjectnullableBet-indexed lines of the graded bets (detail full): index the bet numbers kept (past 600 bets thinned by LTTB, the deepest drawdown always kept), dates, units, drawdown, base_units (the market base's ROI at the same stakes, null unless measured), band_p025 and band_p975 (the no-skill range: 1,000 seeded runs of the same bets at their fair chances; null under 20 bets or for racing), rolling_roi_100 with its no-skill range rolling_p025 and rolling_p975, and markers (season starts, the holdout).
xstringAlways bet_index.
One of: bet_index (open: new values may be added).
indexarray of integerBet numbers (1 is the first graded bet).
datesarray of stringEach kept bet's date (YYYY-MM-DD, UTC).
unitsarray of numberUnits to date.
drawdownarray of numberUnits below the running peak (0 or below).
base_unitsarray of numbernullableThe market base at the same stakes.
band_p025array of numbernullableNo-skill 2.5th percentile.
band_p975array of numbernullableNo-skill 97.5th percentile.
rolling_roi_100array of numberROI over the last 100 bets (null before).
rolling_p025array of numberIts no-skill 2.5th percentile.
rolling_p975array of numberIts no-skill 97.5th percentile.
markersarray of objectWhere each season and the holdout start.
typestringseason or holdout.
One of: season, holdout (open: new values may be added).
atintegerThe bet number it starts at.
labelstringnullableThe season, or newest 30%.
downsampledbooleanThe series were thinned to 600 points.
expected_max_drawdown_zero_edgenumbernullableThe median deepest drawdown of the no-skill runs, units.
holdoutobjectnullableThe latest 30% of bets against the earlier 70%.
seasonsarray of objectSeasons in calendar order.
splitsarray of objectnullableHome, away, favorite, underdog and line bands, in a fixed order; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
withoutarray of objectnullableOne row a top-level AND condition: the rule's record without it; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
p_value_vs_rulenumbernullableThe rule against the bets the condition removes (F19).
p_value_vs_rule_adjustednumbernullablep_value_vs_rule, Holm-adjusted over the conditions shown.
betsarray of objectEvery bet in date order (left out at detail summary).
keystring{game_id}:{team_id} (racing: {race_id}:{runner_id}).
starts_atstringKickoff (RFC 3339).
format date-time
seasonstringThe season.
labelstringDate, team and opponent.
priceobjectnullableThe price at the bet moment, decimal.
linenumbernullableThe bet's line as graded (the team's spread, the total).
resultstringnullablewin, loss or push (soccer quarter lines: half_win or half_loss); null until played.
close_probnumbernullableThe fair closing probability of the same side, for CLV.
units_to_datenumberUnits after this bet, in date order.
how_gradedstringnullableHow the bet was graded, in words.
unitsnumbernullableUnits the bet won or lost at the run's staking (racing: its exact profit or loss with dead heats, each way, Rule 4 and commission); null until played.
drawdown_to_datenumbernullableUnits to date less their running peak (0 or below); null until played.
sidestringnullablehome or away (spread, moneyline), over, under or draw.
marketstringThe market graded (the other total when fading a total).
bet_momentstringWhen the bet was taken: edition, day_before or close.
clv_probnumbernullableThe fair closing chance less the fair chance at the bet moment, when Kelly holds both.
samplestringnullableholdout for the latest 30% of graded bets by time, else in_sample; null with no holdout or until played.
One of: in_sample, holdout (open: new values may be added).
whyarray of objectWhy it matched: up to six conditions (F46).
clausestringThe condition as written.
valuenumber | boolean | stringnullableThe compared expression's number for a comparison (rest_days - opponent.rest_days for rest_days - opponent.rest_days >= 3), else true or false.
settlementarray of stringThe settlement rules that decided the bet (IDs in settlement.rules).
leak_testobjectThe leak test (F3): the rule and each part of it evaluated again on a sample of rows with the data cut at the bet moment. A leak refuses the test (422 rule_not_testable, problem leak_found), so a report always says clean or not_run.
statusstringrequiredclean, leak_found or not_run.
One of: clean, leak_found, not_run (open: new values may be added).
rows_checkedintegerrequiredRows rebuilt at the bet moment.
matched_checkedintegerOf them, rows the rule matched.
matches_flippedintegerRows whose match changed with the data cut.
partsarray of objectrequiredParts that changed, in the formula's order (empty when clean).
namestringThe part, as written (rule is the whole rule).
rows_changedintegerRows where it changed.
max_changenumbernullableThe largest change in a number.
examplestringOne changed row, {game_id}:{team_id}.
warm_upobjectThe warm-up check (F4): bets taken before the rule's windows had the earlier games they read (one season at a time; windows never reach into last season).
policystringThe policy graded in the report.
One of: as_written, require_full (open: new values may be added).
games_neededintegerEarlier games every window needs (0: the rule has no window).
windowsarray of objectEach window in the formula's order.
exprstringThe window, as written.
games_neededintegernullableEarlier games it reads (null: season to date).
short_rowsintegerRows in scope with fewer earlier games this season.
short_betsintegerBets the rule took with fewer.
short_betsintegerBets taken with a window short.
by_policyarray of objectnullableas_written, then require_full; null when the rule has no window.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
warningbooleanShort bets are over 10% of the bets, or leaving them out moves ROI by more than half its 95% interval's half-width.
settlementobjectHow bets were settled (F9): Kelly's own grading rules as data, and each rule's bets.
profilestringThe rules profile (kelly_standard).
rules_versionstringThe rules table's date.
appliedarray of objectRules that decided at least one bet, in table order.
rulestringRule ID.
betsintegerBets.
rulesarray of objectThe rules for this league and market.
idstringRule ID.
textstringThe rule in words.
effective_fromstringThe date Kelly's grader took it up.
coverageobjectCoverage (F10): from every game in scope to the bets graded, each season's priced share, blank inputs and whether the rows left out for no line differ from the bets kept.
flowarray of objectThe steps in order; rule_true rows = decisions + every row out after it.
stepstringin_scope, rule_true, one_bet_a_game, full_windows, priced, graded or decisions.
gamesintegerGames (in_scope only).
rowsintegerRows.
outobjectRows that left at this step, by reason.
by_seasonarray of objectEvery season asked for, in order.
seasonstringSeason.
gamesintegerGames in scope.
rowsintegerRows.
rule_trueintegerRows the rule matched.
betsintegerBets graded.
inputs_missingarray of objectInputs read bare on the row, with blanks.
attributestringThe input.
rows_blankintegerRows with no value.
left_out_differsobjectRows left out for no line against the bets kept: standardized mean difference over home, neutral site and rest days; flagged over 0.1 with 10 or more left out.
flagbooleanThey differ.
max_smdnumbernullableThe largest difference.
covariatestringnullableWhere it is.
left_outintegerRows.
keptintegerRows.
by_covariatearray of objectEach covariate's difference.
momentobjectThe bet moment's clock (F5): its target time, the prices' snapshot age at it and where the lines came from.
bet_momentstringedition, day_before or close.
targetstringnullableThe moment's time for a game, in words.
max_stale_minutesintegerOldest price kept, in minutes before the target.
floor_minutes_before_oct_2022integerThe limit's floor for games before October 2022 (the Odds API's 10-minute snapshots).
rows_pricedintegerRows the rule took that had a price.
rows_agedintegerRows whose price carries its snapshot time and was kept.
median_snapshot_age_minutesnumbernullableMedian minutes from snapshot to target, kept rows.
max_snapshot_age_minutesnumbernullableOldest kept snapshot, in minutes before the target.
stale_priceintegerRows left out: price older than max_stale_minutes.
after_momentintegerRows left out: price seen after the target.
age_unknownintegerRows kept whose price has no snapshot time (Kelly's own frozen lines).
line_sourcesobjectKept rows by line source (sheet_line, odds_api_early, odds_api_close).
left_outobjectrequiredGames the rule matched but didn't bet.
no_lineintegerNo Kelly line at the bet moment.
both_sidesintegerThe rule matched both teams.
stale_priceintegerThe price was older than max_stale_minutes at the bet moment, or seen after it (F5).
unpairedintegerTeasers: legs left off a ticket (a week with too few, or the leftover in chronological pairing).
no_priceintegerRacing: runners with no price to settle at.
non_runnersintegerRacing: non-runners (void).
void_racesintegerRacing: abandoned or void races.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
capsobjectnullableExposure caps applied in time order (F39); the record and checks are on the capped ledger.
skippedintegerBets the caps skipped.
by_reasonobjectSkipped bets by cap.
uncappedobjectThe record without the caps.
random_priority_mean_roinumbernullableMean ROI over random tie-breaks.
ruleobjectrequiredThe rule and the options it was tested with.
rule_hashstringThe rule's identity: SHA-256 of its canonical formula (spacing, case and the order of AND and OR conditions don't change it) with the competition, market, side, bet moment, seasons, season type, stake and price. Detail and tests are not part of it.
pattern ^sha256:[0-9a-f]{64}$
translation_idstringThe rules:translate result this formula came from (F43), when the request named one.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
readsarray of objectEvery name the rule reads, once.
namestringThe attribute, with its relation (opponent.rest_days).
knownstringIts moment: static, pregame, edition, close or final (a window's input is read before the game).
windowedbooleanA backward window reads it (earlier games only).
formula_versionstringThe formula language version the rule was read with (F45).
catalogue_versionstringnullableThe stat catalogue version its names were bound against.
reportobjectA shared report's own facts (get_rule_report only, F54).
report_idstringThe report.
report_urlstringIts page.
format uri
visibilitystringunlisted or public.
One of: unlisted, public (open: new values may be added).
author_rules_triedintegerThe author's rules tried on this market when shared.
frozenbooleanAlways true.
rerunobjectThe run's request, to test it again on today's data.
report_urlstringnullableA frozen report of this run anyone with the link can open (F54): only when the request asked (share) and the call has an account; else null. share_rule_run makes or changes one later.
format uri
chartsarray of objectOnly when the request asked (charts) at detail full: equity against the no-skill band, the drawdown and the rolling ROI as Kelly Chart Specs (F47).
transferobjectnullableOther markets and sports (F33): the same rule on its other market and on other leagues, numbers mapped to the same meaning, pooled with DerSimonian-Laird random effects. Evidence beside the verdict, never part of it; these runs don't count toward rules tried. Present only when the request asks for transfer.
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
whystringnullableWhy nothing ran, when nothing did.
optionsobjectThe leagues, markets and thresholds asked for.
cellsarray of objectEach market and league in a fixed order (the other market first, then sibling leagues), never sorted by a result; up to 12.
competitionstringThe league.
marketstringThe market.
thresholdsstringliteral (the rule's own numbers) or quantile (each number moved to the same percentile of its field in this league).
One of: literal, quantile (open: new values may be added).
formulastringThe formula this cell ran (quantile cells carry the moved numbers).
seasonsarray of stringSeasons read, as this league labels them.
mappedarray of objectnullableQuantile cells only.
numberstringThe number as written.
fieldstringWhat it is compared with.
percentilenumbernullableIts percentile in the rule's own league (a fraction).
valuenumbernullableThe number at the same percentile here.
decisionsintegerWins plus losses.
betsintegerGraded bets.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% score interval.
senumbernullableROI's standard error at 1 unit risked a bet.
p_valuenumbernullableThe cell's own one-sided p against no skill at its prices.
p_holmnumbernullableHolm's adjusted p across the cells.
pooledbooleanThe cell is in the pool (another league, one a league, 10+ decisions).
pooledobjectnullableThe pool of the other leagues' cells.
roinumbernullableRandom-effects ROI.
roi_ci95array of numbernullableIts 95% interval.
senumbernullableIts standard error.
p_valuenumbernullableTwo-sided p of the pooled ROI.
i2numbernullableI²: the share of the spread across leagues beyond chance (a fraction).
tau2numbernullableτ²: the between-league variance of true ROI.
qnumbernullableCochran's Q.
cellsintegerCells pooled.
same_directionobjectnullablePooled cells whose ROI has the rule's own sign.
agreeintegerCells the same way.
cellsintegerCells.
p_valuenumbernullableTwo-sided sign-test p.
textstring"4 of 5".
not_testablearray of objectLeagues and markets left out, with why.
competitionstringThe league.
marketstringnullableThe market.
missingarray of stringFields that league lacks.
whystringWhy.
notestringWhat the panel is and isn't.
run_idstringnullableThis run, kept on the account (run_…): the same request on the same data is the same run. GET /v1/rules/runs/{run_id} reads it back. Null without an account.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
data_versionstringThe data the run read: Kelly's lake snapshot (or the data's as-of time). A new snapshot is a new run of the same rule.
manifestobjectnullableWhat this run read and ran, so the same request can be run again and checked against it (ADR-K885). No prices, lines or rows of source data.
versionintegerThe manifest's own layout version.
codeobjectThe code that graded the run.
engineintegerThe tester's revision; it rises with any fix that can move a result on the same data.
releasestringnullableThe deployed commit.
formula_versionstringnullableThe formula language's version.
catalogue_versionstringnullableThe stat catalogue's version.
attributes_sha256stringnullableA hash of the attribute registry: what each input means and when it is known.
settlement_rulesstringnullableThe settlement rules' version.
dataobjectThe data the run read.
versionstringThe same value as data_version.
snapshotsarray of stringThe lake snapshots read, in order.
one_snapshotbooleanWhether every season was read on one snapshot.
as_ofstringnullableWhen the read was made (RFC 3339).
format date-time
data_throughobjectHow far Kelly's history went, by league.
latest_resultstringnullableThe newest game read that had a result.
rows_built_atstringnullableWhen the prepared rows were built, where the run read them.
format date-time
queriesobjectEach reviewed lake query the tester reads, with a hash of its text.
requestobjectThe request.
request_hashstringThe request's hash.
rule_hashstringThe rule's hash.
seedintegerThe seed every simulation in the run used.
sourcesobjectWhere the run's inputs came from.
pricesobjectBets priced, by the source of the line.
inputsarray of objectEach input the rule read, with the moment it is known.
namestringThe input.
knownstringnullableThe moment it is known.
seasonsarray of objectEach season read.
seasonstringThe season.
rowsintegerTeam-games in scope.
left_outobjectMatched games that were not bet, by reason.
sha256stringThe hash of this block. Two runs with the same hash read the same data with the same code.
since_runobjectnullableOnly on a run read back: what has changed in Kelly's tester since the run was kept.
statusstringcurrent: nothing that can move its numbers changed. superseded: something did. no_manifest: the run was kept before manifests.
One of: current, superseded, no_manifest (open: new values may be added).
engine_changedbooleannullableWhether the tester's revision has risen since.
changedarray of objectThe parts that differ, each with its value then (from) and now (to).
partstringengine, release, formula_language, catalogue, attributes, settlement_rules or queries.
tested_atstringWhen the run was kept (RFC 3339); only on a run read back.
format date-time
rule_versionobjectnullableThe saved rule this test was kept under (save, or rule_id): its rule_id, name, version and versions_tried. Null when the test wasn't saved.
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerrequiredThis rule's version under it: the same rule again keeps its version.
min 1
new_versionbooleanThis test made a new version.
versions_triedintegerrequiredVersions saved under it, this one included.
min 1
since_savedobjectnullableThe saved version's record since it was saved (backtester T6, F50), when the test names its rule_id and repeats that version: verdict v2's since_saved check reads it. Null for a new version (its clock starts now) or an unsaved test.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
tests_on_accountobjectnullableRules this account has tried on this competition and market, this one included (the rules-tried ledger: every distinct rule tested on the site, the API or Kelly MCP, kept for good). The not-luck check adjusts for the larger of this count and the request's own tests. Null when the call has no account.
competitionstringrequiredThe competition.
marketstringrequiredThe market.
countintegerrequiredDistinct rules tried.
min 1
shown_countedintegerVariants an earlier report showed (its condition-removed forms, nearby numbers and home or away splits) that joined the count because this rule is one of them (audit F6).
stakingstringHow each bet was staked.
matchedintegerBets taken.
ungradedintegerBets not yet played.
noticestringrequiredThis tests a rule on past games; it isn't a pick.
calibrationobjectnullableoutput: probability only. The formula's chances against the market's no-vig close of the same side (moneyline: the team wins; spread: it covers the bet's line), on the newest 30% of the scored team-games (whole days), and the bets it makes at the edge threshold.
formulastringThe probability formula as sent.
bet_formulastringThe rule it bets: the chance at the bet's own price worth more than the edge threshold (the rule tested, counted and kept).
edge_thresholdnumberExpected return a bet needs, p × decimal price − 1 (0.02 by default).
basisstringnewest_30: scored on the holdout only.
One of: newest_30 (open: new values may be added).
gamesintegerGames scored.
rowsintegerTeam-games scored (both teams of a game).
in_sample_rowsintegerTeam-games before the holdout, not scored.
holdout_fromstringnullableThe holdout's first day.
left_outobjectTeam-games not scored.
pushesintegerPushes and half results.
outside_0_1integerValues below 0 or above 1.
no_valueintegerRows the formula gave no number for.
clippedintegerValues of 0 or 1 clipped to 1e-6 from the edge.
briernumbernullableBrier score of the formula.
brier_marketnumbernullableBrier score of the close.
brier_skillnumbernullable1 − formula / close (above 0: the formula scores better).
log_lossnumbernullableLog loss of the formula.
log_loss_marketnumbernullableLog loss of the close.
log_loss_skillnumbernullable1 − formula / close.
dm_p_valuenumbernullableDiebold–Mariano p on the per-game log-loss difference (two-sided).
dm_statisticnumbernullableThe HLN-corrected statistic (positive: the formula's loss is higher).
mcbnumbernullableCORP miscalibration (Brier).
dscnumbernullableCORP discrimination.
uncnumbernullableUncertainty; brier = mcb − dsc + unc.
mcb_p_valuenumbernullableThe MCB test: the chance a calibrated formula looks this miscalibrated (999 resamples).
reliabilityarray of objectnullableThe CORP reliability curve at up to 20 forecast values.
pnumbernullableThe formula's probability.
observednumbernullableHow often it happened (PAV fit).
lonumbernullable90% consistency band, low.
hinumbernullable90% consistency band, high.
ecenumbernullableExpected calibration error over ten equal-width bins (bin-dependent; for familiarity).
distinct_forecastsintegerDistinct values the formula gave.
slopenumbernullableCalibration slope b in y ~ a + b·logit p (1 is ideal; under 1 overconfident).
slope_senumbernullableIts game-clustered standard error.
interceptnumbernullableCalibration intercept a (0 is ideal).
encompassingobjectnullabley ~ a + b_q·logit q + b_p·logit p, game-clustered: does the formula know something the close doesn't?
b_modelnumbernullableb_p.
b_marketnumbernullableb_q.
se_modelnumbernullableb_p's standard error.
p_valuenumbernullableTwo-sided p of b_p.
edge_binsarray of objectnullableRealized results against the formula's edge.
edgenumbernullableMean of p − q in the bin.
realized_minus_marketnumbernullableMean of y − q in the bin.
rowsintegerTeam-games in the bin.
statusstringThe calibrated check: pass, fail or na (under 200 games, or one value).
One of: pass, fail, na (open: new values may be added).
valuestringThe check's line.
statisticnumbernullableThe encompassing p.
thresholdnumbernullable0.05.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · not_found
GET https://api.askkelly.ai/v1/rules/reports/rpt_8kqqvwd602ere4fy0nv6
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_test",
"clv": {
"ci95": null,
"mean": null,
"note": "Bets at the close have no closing-line value by definition.",
"basis": "probability",
"devig": "multiplicative",
"p_value": null,
"coverage": 0,
"beat_rate": null,
"mean_prob": null,
"reference": "consensus",
"mean_points": null,
"points_bets": 0,
"points_test": null,
"sign_test_p": null,
"closes_stale": 0,
"profit_slope": null,
"significance": null,
"sample_needed": null,
"beat_rate_ci95": null,
"bets_with_close": 0,
"lines_converted": 0,
"points_coverage": 0,
"profit_slope_se": null,
"beat_close_share": null,
"sample_needed_roi": null,
"bets_without_close": 1
},
"base": {
"why": "the rule has no who-or-when condition to filter, so its base is the whole market",
"source": "market",
"status": "na",
"formula": null
},
"caps": null,
"luck": {
"bets": 0,
"devig": "multiplicative",
"theta": null,
"reading": "versus_the_market",
"left_out": {
"half": 0,
"push": 0
},
"by_season": [],
"actual_roi": null,
"luck_units": null,
"actual_wins": null,
"actual_units": null,
"close_source": "consensus",
"expected_roi": null,
"expected_wins": null,
"luck_se_units": null,
"expected_units": null,
"lines_converted": 0,
"luck_percentile": null,
"bets_without_prob": 1
},
"path": {
"series": {
"x": "bet_index",
"dates": [
"2025-10-21"
],
"index": [
1
],
"units": [
0.91
],
"markers": [
{
"at": 1,
"type": "season",
"label": "2025-26"
}
],
"drawdown": [
0
],
"band_p025": null,
"band_p975": null,
"base_units": null,
"rolling_p025": [
null
],
"rolling_p975": [
null
],
"rolling_roi_100": [
null
]
},
"downsampled": false,
"longest_losing_run": 0,
"max_drawdown_units": 0,
"expected_max_drawdown_zero_edge": null
},
"risk": {
"seed": 6899137762779026,
"curve": [],
"paths": 5000,
"method": "lundberg+posterior_monte_carlo",
"status": "too_few",
"percent": {
"status": "too_few",
"drawdown_p95": null,
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null
},
"staking": "flat",
"warnings": [
"too_few_decisions"
],
"ruin_share": 0.5,
"drawdown_p95": null,
"horizon_bets": 500,
"lose_all_ever": 0,
"bankroll_units": 100,
"stake_fraction": null,
"down_after_horizon": null,
"drawdown_p95_units": null,
"estimation_premium": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"ruin_ever_if_rate_exact": 0,
"bankroll_for_ruin_005_units": null,
"lose_all_ever_allowing_error": 0.277131008418
},
"rule": {
"side": "rule",
"price": "consensus",
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
}
],
"stake": "risk",
"market": "spread",
"formula": "=home",
"seasons": [
"2025-26"
],
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0"
},
"as_of": "2026-11-06T00:00:00Z",
"decay": {
"bets": 1,
"break": null,
"cusum": null,
"window": 50,
"rolling": [],
"clv_break": null,
"decay_ratio": null,
"known_break": null,
"roi_newest_30": null,
"status_reason": "fewer than (100) graded bets",
"trend_per_100_bets": null
},
"noise": {
"seed": 3397954809,
"delay": {
"na": "day-before lines for this league are still loading",
"bet_moment": "day_before"
},
"lines": {
"runs": 200,
"roi_p05": 0.9099999999999999,
"line_prob": 0.5,
"roi_median": 0.9099999999999999,
"profitable_runs": 200,
"profitable_share": 1
},
"inputs": null,
"inputs_na": "the rule compares no numbers",
"low_power": true
},
"stale": false,
"honest": {
"roi": -0.04514385341429974,
"ci80": [
-0.07077022186419332,
-0.019517484964406154
],
"prior": {
"mean": -0.04545450206611379,
"words": "shrunk toward the market's usual margin of -4.5%",
"capped": false,
"closes": 0,
"source": "margin"
},
"tries": 1,
"shrink": 0.999560632688928,
"raw_roi": 0.9099999999999999,
"prior_sd": 0.02,
"uses_clv": false,
"prior_mean": -0.04545450206611379,
"basis_rules": 0,
"prior_basis": "seed",
"decay_factor": 0.74,
"posterior_sd": 0.019995605844174143,
"selection_adj": 0,
"stabilizes_at_bets": 2275
},
"moment": {
"target": "the start",
"rows_aged": 0,
"bet_moment": "close",
"age_unknown": 1,
"rows_priced": 1,
"stale_price": 0,
"after_moment": 0,
"line_sources": {
"sheet_line": 1
},
"max_stale_minutes": 60,
"max_snapshot_age_minutes": null,
"median_snapshot_age_minutes": null,
"floor_minutes_before_oct_2022": 10
},
"nearby": {
"na": "the rule compares no numbers",
"grid": null,
"numbers": [],
"reading": null,
"roi_p05": null,
"roi_p95": null,
"variants": 0,
"roi_median": null,
"spike_ratio": null,
"plateau_share": null,
"distinct_bet_sets": 0
},
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"run_id": "run_mkytv75qe5a297z7y3af",
"splits": [
{
"roi": 0.9099999999999999,
"bets": 1,
"label": "Home",
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1,
"p_adjusted": 1,
"shown_rows": 2
},
{
"roi": 0.9099999999999999,
"bets": 1,
"label": "Favorite",
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1,
"p_adjusted": 1,
"shown_rows": 2
}
],
"stress": {
"grid": [
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": 0
},
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": -0.5
},
{
"roi": 0.9523809523809523,
"vig": -105,
"units": 0.9523809523809523,
"line_shift": -1
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": 0
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": -0.5
},
{
"roi": 0.9090909090909092,
"vig": -110,
"units": 0.9090909090909092,
"line_shift": -1
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": 0
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": -0.5
},
{
"roi": 0.8695652173913042,
"vig": -115,
"units": 0.8695652173913042,
"line_shift": -1
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": 0
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": -0.5
},
{
"roi": 0.8333333333333335,
"vig": -120,
"units": 0.8333333333333335,
"line_shift": -1
}
],
"flat_110": {
"roi": 0.9090909090909092,
"bets": 1,
"units": 0.9090909090909092,
"record": "1-0",
"p_value": 0.5238095238095238,
"decisions": 1
},
"break_even": {
"decimal_avg": null,
"line_points": 3,
"american_avg": null,
"price_multiplier": 0
},
"other_side": {
"roi": -1,
"bets": 1,
"units": -1,
"record": "0-1",
"p_value": 1,
"decisions": 1
},
"slippage_mc": null,
"half_point_worse": {
"roi": 0.9099999999999999,
"bets": 1,
"units": 0.9099999999999999,
"record": "1-0",
"p_value": 0.5235602094240838,
"decisions": 1
}
},
"holdout": null,
"matched": 1,
"overall": {
"roi": 0.9099999999999999,
"bets": 1,
"wins": 1,
"units": 0.9099999999999999,
"losses": 0,
"pushes": 0,
"record": "1-0",
"roi_ci": {
"bca": null,
"level": 0.95,
"score": [
-0.6054908095394765,
0.9099999999999999
],
"method": "score",
"iid_score": [
-0.6054908095394765,
0.9099999999999999
],
"anytime_valid": [
-0.999045,
0.9090450000000001
],
"block_bootstrap": null
},
"deflated": {
"dsr": null,
"psr": null,
"skew": null,
"roi_bar": null,
"kurtosis": null,
"sample_needed": null,
"sharpe_per_bet": null,
"trials_effective": 1,
"sample_needed_plan": null,
"sharpe_expected_max": null,
"sample_needed_single": null,
"sample_needed_power80": null
},
"formulas": [
"F3.2@1",
"F12.1@1",
"F12.3@1",
"F12.4@1",
"F36.2@1"
],
"roi_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"warnings": [
"fixed_n_p_value",
"small_sample"
],
"win_rate": 1,
"decisions": 1,
"break_even": 0.5235602094240838,
"units_ci95": [
-0.6054908095394765,
0.9099999999999999
],
"roi_interval": [
-0.6054908095394765,
0.9099999999999999
],
"significance": {
"z": 0.9539392014169455,
"null": "break_even",
"seed": null,
"sims": null,
"alpha": 0.05,
"spa_p": null,
"tests": 1,
"method": "exact_binomial",
"t_stat": null,
"haircut": null,
"p_value": 0.5235602094240838,
"clears_t3": false,
"p_adjusted": 0.5235602094240838,
"haircut_roi": null,
"significant": false,
"adjust_method": "none",
"anytime_valid": {
"max_e": 1.0580125,
"method": "kelly_betting_mixture",
"crossed": false,
"e_value": 1.0580125,
"p_value": 0.9451684171973394,
"roi_cs95": [
-0.999045,
0.9090450000000001
],
"kelly_fraction": 0.5
},
"p_value_exact": 0.5235602094240838,
"p_value_vs_vig": null,
"reality_check_p": null,
"tests_effective": 1,
"p_value_clustered": 0.17005720935900537,
"coin_match_per_100": 50,
"multiple_comparisons": null,
"no_skill_match_per_100": 52.35602094240838
},
"small_sample": true,
"few_decisions": true,
"sample_needed": null,
"over_break_even": 0.47643979057591623
},
"seasons": [
{
"roi": 0.9099999999999999,
"bets": 1,
"wins": 1,
"units": 0.9099999999999999,
"losses": 0,
"pushes": 0,
"record": "1-0",
"season": "2025-26",
"clv_mean": null,
"win_rate": 1,
"decisions": 1,
"break_even": 0.5235602094240838,
"roi_interval": [
0.9099999999999999,
0.9099999999999999
],
"significance": {
"z": 0.9539392014169455,
"alpha": 0.05,
"p_value": 0.5235602094240838,
"significant": false,
"anytime_valid": {
"crossed": false,
"e_value": 1.0500500000000001,
"p_value": 0.9523356030665205
},
"coin_match_per_100": 50,
"multiple_comparisons": null,
"no_skill_match_per_100": 52.35602094240838
},
"small_sample": true,
"few_decisions": true,
"sample_needed": null,
"over_break_even": 0.47643979057591623
}
],
"staking": "flat, 1 unit risked a bet",
"verdict": {
"gate": {
"key": "no_look_ahead",
"value": "every value the rule reads is known at the bet moment",
"status": "pass",
"leak_test": "clean"
},
"level": "too_few",
"checks": [
{
"key": "enough_bets",
"zone": "red",
"group": "real",
"label": "Enough bets",
"value": "(1) decisions are too few to size the sample",
"status": "fail",
"statistic": 1,
"threshold": null
},
{
"key": "not_luck",
"zone": "red",
"group": "real",
"label": "Not just luck",
"value": "p 0.524",
"status": "fail",
"statistic": 0.5235602094240838,
"threshold": 0.05
},
{
"key": "beats_base",
"zone": "na",
"group": "real",
"label": "Beats its own base",
"value": "the rule has no who-or-when condition to filter, so its base is the whole market",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "half_point_worse",
"zone": "na",
"group": "pressure",
"label": "Holds ½ point worse",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"zone": "na",
"group": "pressure",
"label": "Nearby thresholds hold",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"zone": "na",
"group": "pressure",
"label": "Not one lucky stretch",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "beats_close",
"zone": "na",
"group": "still_working",
"label": "Beats the close",
"value": "bets at the close can't beat it",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"zone": "na",
"group": "still_working",
"label": "Still working",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "since_saved",
"zone": "na",
"group": "still_working",
"label": "Since saved",
"value": "(0) of (30) bets since it was saved",
"status": "na",
"statistic": null,
"threshold": null
}
],
"passed": 0,
"version": 2,
"headline": "Too few bets to judge",
"applicable": 2,
"held_back_by": []
},
"warm_up": {
"policy": "as_written",
"warning": false,
"windows": [],
"by_policy": null,
"short_bets": 0,
"games_needed": 0
},
"without": [],
"ablation": null,
"capacity": {
"books": [],
"market": "spreads",
"reason": "Kelly holds no measured spreads limits for nba at this moment yet.",
"available": false,
"limits_held": 0,
"before_start": "0_1h",
"bets_per_season": null,
"max_stake_per_bet": null,
"per_season_at_limit": null
},
"coverage": {
"flow": [
{
"rows": 6,
"step": "in_scope",
"games": 3
},
{
"rows": 3,
"step": "rule_true"
},
{
"out": {
"both_sides": 0,
"same_game_total": 0
},
"rows": 3,
"step": "one_bet_a_game"
},
{
"out": {
"no_line": 2
},
"rows": 1,
"step": "priced"
},
{
"out": {
"void": 0,
"not_played_yet": 0
},
"rows": 1,
"step": "graded"
},
{
"out": {
"push": 0
},
"rows": 1,
"step": "decisions"
}
],
"by_season": [
{
"bets": 1,
"rows": 6,
"games": 3,
"season": "2025-26",
"rule_true": 3,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"rows": 6,
"share": 0,
"attribute": "home",
"rows_blank": 0,
"seasons_blank": []
}
],
"left_out_differs": {
"flag": false,
"kept": 1,
"max_smd": null,
"left_out": 2,
"covariate": null,
"by_covariate": []
}
},
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
},
"manifest": {
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"settlement_rules": "2026-11-08",
"attributes_sha256": "0000000000000000",
"catalogue_version": "1:0"
},
"data": {
"as_of": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
},
"version": "snap_52",
"snapshots": [
"snap_52"
],
"data_through": {},
"one_snapshot": true,
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z"
},
"sha256": "sha256:d78d99c7954d4b2367b7baa69328756b5aa4464bb52ddcb24cdf05a0484557ad",
"request": {
"seed": 18147943352127158000,
"rule_hash": "sha256:c9a3bcfd2d08b46b86935391523bd2671a885a30b3b1f15bf69e12c5e399441b",
"request_hash": "sha256:13e69ec9c84eb683918ee67e91ae21ac69cc813cfbc4758f625524596be5d3da"
},
"seasons": [
{
"rows": 6,
"season": "2025-26",
"priced_share": 0.3333
}
],
"sources": {
"inputs": [
{
"name": "home",
"known": "pregame"
}
],
"prices": {
"sheet_line": 1
}
},
"version": 1,
"left_out": {
"no_line": 2,
"both_sides": 0,
"stale_price": 0
}
},
"ungraded": 0,
"leak_test": {
"parts": [],
"status": "clean",
"rows_checked": 6,
"matched_checked": 3,
"matches_flipped": 0
},
"dependence": {
"se": null,
"slate": "date",
"method": "none",
"se_iid": null,
"clusters": 1,
"cluster_by": [
"slate",
"team_season"
],
"design_effect": 1,
"effective_bets": 1
},
"robustness": {
"off": [],
"seed": 769905824,
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"value": "fewer than (30) decided bets",
"status": "na",
"statistic": null,
"threshold": null
}
],
"fast_path": true,
"evaluations": 1
},
"settlement": {
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
],
"applied": [
{
"bets": 1,
"rule": "final_score"
}
],
"profile": "kelly_standard",
"rules_version": "2026-11-08"
},
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (2).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"monte_carlo": null,
"since_saved": null,
"data_through": {},
"data_version": "snap_52",
"concentration": {
"by": [
{
"best": null,
"group": "season",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "team",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "opponent",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "month",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
},
{
"best": null,
"group": "weekday",
"groups": 1,
"without_best": null,
"effective_groups": 1,
"jackknife_roi_se": null,
"without_best_two": null,
"p_more_concentrated": null
}
],
"bets": 1,
"units": 0.9099999999999999,
"trim_top_0_05": null
},
"latest_result": "2025-10-25T23:00:00Z",
"risk_adjusted": {
"sqn": null,
"bets": 1,
"warnings": [
"no_losing_bets",
"no_spread_in_returns",
"small_sample"
],
"ulcer_index": 0,
"martin_ratio": null,
"sharpe_daily": null,
"avg_win_units": 0.9099999999999999,
"profit_factor": null,
"avg_loss_units": null,
"sharpe_per_bet": null,
"recovery_factor": null,
"sortino_per_bet": null,
"sharpe_per_season": null,
"max_drawdown_units": 0,
"profit_factor_ci95": null,
"return_on_drawdown": null,
"sharpe_per_bet_ci95": null,
"probabilistic_sharpe": null,
"sortino_per_bet_ci95": null,
"current_drawdown_units": 0,
"expected_units_per_bet": 0.9099999999999999,
"longest_underwater_bets": 0,
"longest_underwater_days": 0
},
"rows_built_at": "2026-11-06T00:00:00Z",
"staking_plans": {
"kelly": {
"basis": "record",
"capped": null,
"shrink": 1,
"win_rate": 1,
"withheld": "stake_first_party_only",
"push_rate": 0,
"multiplier": null,
"simultaneous": false,
"zero_because": "too_few",
"full_fraction": null,
"honest_fraction": null,
"multiplier_rule": "honest"
},
"plans": [
{
"roi": 0.9099999999999966,
"bets": 1,
"plan": "flat",
"units": 0.9099999999999966,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.009058844488345926,
"stopped_at_bet": null,
"end_bankroll_units": 100.91
},
{
"roi": 0.9099999999999999,
"bets": 1,
"plan": "to_win",
"units": 1,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.009950330853168092,
"stopped_at_bet": null,
"end_bankroll_units": 101
},
{
"roi": 0.9099999999999966,
"bets": 1,
"plan": "percent",
"units": 1.8199999999999932,
"status": "ok",
"max_drawdown": 0,
"growth_per_bet": 0.018036362486098575,
"stake_fraction": 0.02,
"stopped_at_bet": null,
"end_bankroll_units": 101.82
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"percent": 0.02,
"warnings": [
"small_sample"
],
"kelly_cap": 0.05,
"bankroll_units": 100
},
"random_baseline": null,
"split_dimensions": [],
"matched_both_sides": 0,
"matched_without_line": 2,
"tested_at": "2026-10-04T17:30:00Z",
"report": {
"report_id": "rpt_8kqqvwd602ere4fy0nv6",
"report_url": "https://askkelly.ai/rule-tester/r/rpt_8kqqvwd602ere4fy0nv6",
"visibility": "unlisted",
"author_rules_tried": 2,
"shared_at": "2026-10-04T17:30:00Z",
"frozen": true,
"rerun": {
"side": "rule",
"alpha": 0.05,
"price": "consensus",
"stake": "risk",
"detail": "full",
"market": "spread",
"robust": {
"off": [],
"line_prob": 0.5,
"skip_prob": 0,
"break_date": null,
"noise_runs": 200,
"input_scale": 0.1,
"trim_winners": 0,
"monte_carlo_mode": "block",
"monte_carlo_sims": null,
"random_baseline_sims": 2000
},
"sizing": {
"caps": {
"priority": "earliest",
"max_open_stake": null,
"max_bets_per_day": null,
"max_bets_per_game": null,
"daily_stop_loss_units": null,
"max_bets_per_team_week": null
},
"paths": 5000,
"percent": 0.02,
"kelly_cap": 0.05,
"ruin_share": 0.5,
"horizon_bets": 500,
"staking_plans": [
"flat",
"to_win",
"percent",
"kelly"
],
"bankroll_units": 100,
"kelly_fraction": null
},
"teaser": null,
"formula": "=home",
"seasons": [
"2025-26"
],
"warm_up": "as_written",
"ci_level": 0.95,
"plan_roi": null,
"stake_ok": false,
"bet_moment": "close",
"competition": "nba",
"season_type": "regular",
"base_formula": null
}
},
"bets": [
{
"game_key": "gm_1hnx5nrmk4xjpz6fzyfz:tm_b4w5y9xy5f9sgg1cmkvy",
"game_id": "gm_1hnx5nrmk4xjpz6fzyfz",
"team_id": "tm_b4w5y9xy5f9sgg1cmkvy",
"starts_at": "2025-10-21T23:30:00Z",
"season": "2025-26",
"market": "spread",
"bet_moment": "close",
"result": "win",
"units": 0.91,
"units_to_date": 0.91
}
],
"charts": [
{
"kcs": 1,
"kind": "equity",
"title": "Units, bet by bet",
"data": {
"key": "bet",
"columns": [
{
"id": "bet",
"type": "int",
"label": "Bet"
},
{
"id": "date",
"type": "date",
"label": "Date"
},
{
"id": "units",
"type": "float",
"label": "Units",
"unit": "units",
"format": "+.2f"
}
],
"rows": [
{
"bet": 1,
"date": "2025-10-21",
"units": 0.91
}
]
},
"x": {
"field": "bet",
"type": "quantitative",
"title": "Bet number",
"units": []
},
"y": {
"field": "units",
"type": "quantitative",
"title": "Units",
"units": []
},
"series": [
{
"field": "units",
"role": "market",
"label": "Your rule",
"dash": false
}
],
"bands": [],
"intervals": [],
"rules": [
{
"y": 0,
"role": "zero"
}
],
"tooltip": [
"bet",
"date",
"units"
],
"summary": "Units over (1) bets end at +0.9.",
"provenance": {
"tool": "test_rule",
"engine": "ruletest 8",
"as_of": "2026-11-06T00:00:00Z",
"sources": [
"kelly"
]
},
"rights": {
"export_rows": true,
"export_columns_blocked": []
}
},
{
"kcs": 1,
"kind": "underwater",
"title": "Drawdown from the peak",
"data": {
"key": "bet",
"columns": [
{
"id": "bet",
"type": "int",
"label": "Bet"
},
{
"id": "date",
"type": "date",
"label": "Date"
},
{
"id": "drawdown",
"type": "float",
"label": "Below the peak",
"unit": "units",
"format": "+.2f"
}
],
"rows": [
{
"bet": 1,
"date": "2025-10-21",
"drawdown": 0
}
]
},
"x": {
"field": "bet",
"type": "quantitative",
"title": "Bet number",
"units": []
},
"y": {
"field": "drawdown",
"type": "quantitative",
"title": "Units below the peak",
"units": []
},
"series": [
{
"field": "drawdown",
"role": "market",
"label": "Below the peak",
"dash": false
}
],
"bands": [],
"intervals": [],
"rules": [
{
"y": 0,
"role": "zero"
}
],
"tooltip": [
"bet",
"date",
"drawdown"
],
"summary": "The deepest drawdown was 0.0 units; the longest losing run (0) bets.",
"provenance": {
"tool": "test_rule",
"engine": "ruletest 8",
"as_of": "2026-11-06T00:00:00Z",
"sources": [
"kelly"
]
},
"rights": {
"export_rows": true,
"export_columns_blocked": []
}
}
],
"summary": "Too few bets to judge: honest forward ROI -4.5%, 1-0 against the spread, 2025-26, +0.9 units, +91.0% ROI past (0 of 2 checks passed).",
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_45e792ee6c71434c2737d1dc",
"live_topics": [],
"rg_line": null
}GET https://api.askkelly.ai/v1/rules/reports/rpt_zzzzzzzzzzzzzzzzzzzz
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No shared report with that ID.",
"param": "report_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_e7bac8c330c92e060e233cba",
"retryable": false,
"display_message": "That could not be found."
}GET /v1/rules/runs
public, no key needed · 0 credits, not charged during the beta · MCP tool list_rule_runs · status live
Your kept rule tests, newest first, a page at a time: each rule, its options, record, units, ROI, p-value, honest forward ROI and verdict. Every run is listed, so the rules you tried are honest; nothing is sorted by a result.
competitionstring · queryOnly this competition's runs.
min length 2 · max length 32
limitinteger · queryRuns a page (default 50).
min 1 · max 200
cursorstring · querynext_cursor from the previous page.
max length 120
200 The rule runs facts.
objectstringrequiredrule_runs.
One of: rule_runs (open: new values may be added).
runsarray of objectrequiredRuns, newest first.
run_idstringThe run.
tested_atstringWhen.
format date-time
data_versionstringThe data it read.
rule_idstringnullableThe saved rule it was kept under.
versionintegernullableIts version there.
formulastringnullableThe rule.
competitionstringnullableCompetition.
marketstringnullableMarket.
sidestringnullablerule or fade.
bet_momentstringnullableBet moment.
seasonsarray of stringnullableSeasons.
betsintegernullableBets.
recordstringnullableW-L-P.
unitsnumbernullableUnits.
roinumbernullableROI.
p_valuenumbernullableThe record's p-value.
honest_roinumbernullableHonest forward ROI.
verdict_levelstringnullableVerdict level.
headlinestringnullableVerdict headline.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/runs?limit=5
Authorization: Bearer $KELLY_API_KEY{
"object": "rule_runs",
"runs": [
{
"run_id": "run_rc69h189x95wy5rhfet2",
"tested_at": "2026-10-04T17:30:00Z",
"data_version": "snap_52",
"rule_id": null,
"version": null,
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"bets": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"honest_roi": -0.04514385341429974,
"verdict_level": "too_few",
"headline": "Too few bets to judge"
},
{
"run_id": "run_mkytv75qe5a297z7y3af",
"tested_at": "2026-10-04T17:30:00Z",
"data_version": "snap_52",
"rule_id": null,
"version": null,
"formula": "=home",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"bets": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"honest_roi": -0.04514385341429974,
"verdict_level": "too_few",
"headline": "Too few bets to judge"
},
{
"run_id": "run_ewabjhkycqv5zk6mvbkn",
"tested_at": "2026-10-04T17:30:00Z",
"data_version": "snap_52",
"rule_id": null,
"version": null,
"formula": "=AND(home, rest_days >= 1)",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"bets": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"honest_roi": null,
"verdict_level": "too_few",
"headline": "No bets"
}
],
"summary": "(3) of your kept rule tests, newest first.",
"assumptions": [],
"has_more": false,
"next_cursor": null,
"as_of": "2026-10-04T17:30:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_8f944202e84f29465c532c17",
"live_topics": [],
"rg_line": null
}GET /v1/trends
public, no key needed · 0 credits, not charged during the beta · MCP tool list_trends · status live
Kelly's nightly trend screen for one league and market (backtester T8, F42): every condition and pair of conditions from a fixed list, bet at the consensus close on the seasons but the newest two, charged with Benjamini-Yekutieli at q 0.10 over every rule ever screened there, then replicated on the newest two seasons. Each trend's record, raw and shrunk ROI, p and q, its replication, its status and its record since it was first found; the counts of rules screened, discovered and replicated beside it. Listed by formula, never ranked; never a pick.
competitionstring · queryrequiredThe league: nfl, ncaaf, nba, ncaab, wnba, nhl, mlb.
One of: nfl, ncaaf, nba, ncaab, wnba, nhl, mlb (open: new values may be added).
marketstring · queryspread (default), moneyline, over or under.
One of: spread, moneyline, over, under (open: new values may be added).
statusstring · queryOnly trends with this status: replicated, waiting_for_bets or not_replicated.
One of: replicated, waiting_for_bets, not_replicated (open: new values may be added).
200 The trends facts.
objectconstrequiredAlways trends.
Always "trends".
competitionstringThe league.
marketstringThe market.
bet_momentstringclose: every bet at Kelly's consensus close.
One of: close (open: new values may be added).
run_idstringnullableThe screen's run (tr_…); null before the first run.
pattern ^tr_[0-9a-z_]{1,80}$
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
candidates_screenedintegerRules ever screened in this league and market: the m the false discovery rate is charged on. It only grows.
screened_this_runintegerRules this run screened.
fdr_qnumbernullableThe Benjamini-Yekutieli level (0.10).
discoveredintegerRules that passed the false discovery bar.
replicatedintegerOf those listed, how many held up on the newest two seasons.
expected_falsenumberAbout how many discoveries are false: q × discovered.
train_seasonsarray of stringSeasons screened.
replicate_seasonsarray of stringSeasons held back for replication.
market_roinumbernullableThe market's own ROI at these prices on the training seasons (every side bet).
trendsarray of objectThe trends, by formula (near-duplicates collapsed to the more general).
formulastringThe rule, a formula test_rule runs as written.
decisionsintegerDecisions on the training seasons.
recordstringIts record on the training seasons.
unitsnumberUnits on the training seasons.
roi_observednumbernullableIts raw ROI on the training seasons.
roi_shrunknumbernullableThe ROI shown: shrunk toward the market's own ROI at these prices (empirical Bayes); always between the two.
p_valuenumbernullableOne-sided p on the training seasons, before the search is charged.
q_valuenumbernullableBenjamini-Yekutieli q over every rule ever screened in this league and market.
replicationobjectIts record on the newest two seasons, which screening never saw.
recordstringWins-losses(-pushes).
decisionsintegerWins plus losses.
unitsnumberUnits won or lost, one unit risked a bet.
roinumbernullableUnits over units risked.
p_valuenumbernullableOne-sided p against no skill at these prices.
statusstringreplicated (p 0.05 or under, ROI at least 0, same sign), waiting_for_bets (under 40 decisions so far) or not_replicated.
One of: replicated, waiting_for_bets, not_replicated (open: new values may be added).
found_onstringThe day the screen first found it.
format date
since_foundobjectnullableIts record on games after it was first found.
recordstringWins-losses(-pushes).
decisionsintegerWins plus losses.
unitsnumberUnits won or lost, one unit risked a bet.
roinumbernullableUnits over units risked.
p_valuenumbernullableOne-sided p against no skill at these prices.
noticestringA trend is a test of the past, never a pick.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled
GET https://api.askkelly.ai/v1/trends?competition=nfl&market=parlay
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "market: 'parlay' is not one of ['spread', 'moneyline', 'over', 'under']",
"param": "market",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_b59411a65d0e307e81a90272",
"retryable": false,
"display_message": "Something in that request was not understood."
}GET https://api.askkelly.ai/v1/trends?competition=nfl&market=spread
Authorization: Bearer $KELLY_API_KEY{
"object": "trends",
"competition": "nfl",
"market": "spread",
"bet_moment": "close",
"run_id": null,
"as_of": "2026-10-04T17:30:00Z",
"candidates_screened": 0,
"screened_this_run": 0,
"fdr_q": null,
"discovered": 0,
"replicated": 0,
"expected_false": 0,
"train_seasons": [],
"replicate_seasons": [],
"market_roi": null,
"trends": [],
"notice": "Rules found by screening past games, charged for every rule screened. A trend is a test of the past, never a pick.",
"summary": "No trend run yet for NFL spread: the nightly screen hasn't covered it.",
"assumptions": [
"Each trend bets every match at Kelly's consensus close, one unit a bet.",
"The ROI shown is shrunk toward the market's own ROI at these prices; the raw ROI is beside it.",
"Listed by formula, never ranked."
],
"has_more": false,
"next_cursor": null,
"source": "kelly",
"attribution": null,
"request_id": "req_3e850b991a33fb89b458b782",
"live_topics": [],
"rg_line": null
}POST /v1/trends:audit
public, no key needed · 1 credit, not charged during the beta · MCP tool audit_trend · status live
A trend someone published, re-tested (backtester T8, F42 audit): its words written as a rule formula (as POST /v1/rules:translate writes them; one question back when they read two ways, and nothing runs), the formula tested as POST /v1/rules:test tests it, the claimed record and ROI (or the record at -110) beside Kelly's with the difference, and the record on games after the publication date. A claim with a publication date is tested as one pre-registered rule and isn't counted toward your rules tried; without one it counts like any test. Kelly holds closes from June 2020. Never a pick.
Body, application/json (required):
textstringThe trend as it was published, in its own words (or a formula starting with =), up to 1,000 characters.
min length 1 · max length 1000
competitionstringThe competition's slug: nfl, nba, mlb, nhl, ncaaf, ncaab, wnba, epl...
pattern ^[a-z0-9_]{2,24}$
claimed_recordstringThe record the trend claims, such as 60-40 or 60-40-2 (wins, losses, pushes).
pattern ^\s*\d{1,6}\s*-\s*\d{1,6}(\s*-\s*\d{1,6})?\s*$
claimed_roinumberThe ROI the trend claims, as a fraction (0.08 is 8%); left out, the claimed record at -110.
min -1 · max 100
published_onstringWhen the trend was published, YYYY-MM-DD: the record since counts games after it.
format date
marketstringspread, moneyline, over, under or draw; left out, the market the words name (spread when they name none).
One of: spread, moneyline, over, under, draw (open: new values may be added).
sidestringrule or fade; left out, what the words say.
One of: rule, fade (open: new values may be added).
seasonsarray of stringSeasons to test (default: every season Kelly holds).
min items 1 · max items 25
from_seasonstringFirst season, e.g. 1999 or 1999-00.
pattern ^[0-9]{4}(-[0-9]{2})?$
to_seasonstringLast season.
pattern ^[0-9]{4}(-[0-9]{2})?$
season_typestringregular (default), postseason or both.
One of: regular, postseason, both (open: new values may be added).
200 The trend audit facts.
objectconstrequiredAlways trend_audit.
Always "trend_audit".
statusstringtested; needs_answer (the words read two ways: answer the translation's question); not_testable (the words can't be a rule: see the translation's problems).
One of: tested, needs_answer, not_testable (open: new values may be added).
claimobjectThe claim as sent.
textstringThe trend's words.
claimed_recordstringnullableThe claimed record, normalised (60-40 or 60-40-2).
claimed_roinumbernullableThe claimed ROI as sent.
published_onstringnullableThe publication date.
format date
translationobjectThe words as a formula (translate_rule's facts; the formula, readback and translation_id once tested).
formulastringnullableThe formula.
readbackstringnullableThe formula in plain English, written by template.
translation_idstringnullableThe translation's id, kept with the test.
questionstringnullableOne question when the words read two ways (needs_answer).
problemsarray of objectWhy the words can't be a rule (not_testable).
formulastringnullableThe formula tested.
claim_checkobjectnullableThe claim beside Kelly's record; null without a claimed record or ROI.
claimed_recordstringnullableThe claimed record.
claimed_roinumbernullableThe claim's ROI: as claimed, or the claimed record at -110.
claimed_roi_basisstringas claimed, or the claimed record at -110.
One of: as claimed, the claimed record at -110 (open: new values may be added).
kelly_recordstringnullableKelly's record on the formula.
kelly_roinumbernullableKelly's ROI at its consensus prices.
claim_inflation_roinumbernullableThe claim's ROI minus Kelly's.
since_publishedobjectnullableThe record on games after the publication date: the part nobody could have mined; null without published_on.
fromstringThe first day counted.
format date
first_gamestringnullableThe first game counted.
recordstringWins-losses(-pushes).
decisionsintegerWins plus losses.
unitsnumberUnits won or lost, one unit risked a bet.
roinumbernullableUnits over units risked.
coverageobjectWhat Kelly could check.
seasons_testedarray of stringSeasons tested.
held_fromstringThe first close Kelly holds.
format date
notestringSeasons before then can't be checked.
testobjectnullableThe formula's full rule test, as test_rule returns it, without the bets list; null when nothing ran.
verdictobjectVerdict v2: the leak gate, nine checks in three groups and the headline.
versioninteger2.
headlinestringrequiredThe verdict in words.
gateobjectThe leak gate: a rule reading anything not known at the bet moment is refused first; leak_test is the cut-data rerun's status (clean or not_run).
held_back_byarray of stringWhy a rule with no failed check isn't holds_up (honest forward ROI at or under 0, PBO over 0.5 across 10 or more variants).
levelstringrequiredtoo_few, no_profit, no_evidence, luck, promising or holds_up.
One of: too_few, no_profit, no_evidence, luck, promising, holds_up (open: new values may be added).
passedintegerrequiredChecks passed.
applicableintegerrequiredChecks that apply.
checksarray of objectrequiredThe checks, in a fixed order.
keystringrequiredThe check; a teaser keeps v1's six (newest_30, seasons_won, price_worse among them).
One of: enough_bets, not_luck, beats_base, half_point_worse, nearby_holds, not_concentrated, beats_close, still_working, since_saved, calibrated, newest_30, seasons_won, price_worse (open: new values may be added).
groupstringreal (is it real?), pressure or still_working; not on a teaser's checks.
One of: real, pressure, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na (it can't apply).
One of: pass, fail, na (open: new values may be added).
zonestringgreen, yellow (within one standard error of the threshold), red, or na.
One of: green, yellow, red, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumber | integernullableThe number the check compares.
thresholdnumber | integernullableWhat it is compared with.
basisstringseasons_won or newest_30 while the full check is still being built; points when Beats the close is judged on points beaten (audit F3).
overallobjectEvery graded bet.
recordstringWins-losses(-pushes).
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
betsintegerGraded bets.
decisionsintegerWins plus losses.
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_intervalarray of numbernullableROI's 95% interval at 1 unit risked a bet.
win_ratenumbernullableWins over decisions.
break_evennumbernullableThe win rate that breaks even at these prices.
significanceobjectnullableThe record test (kmath.records.record_test) and verdict v2's statistics (kmath.significance): the p-value, adjusted for the rules tried, and the anytime-valid test.
p_valuenumbernullableThe larger of the exact and the clustered one-sided p against no skill at these prices (F12, F23).
methodstringnullableexact_binomial, exact_simulation, saddlepoint or z_score.
znumbernullableThe score z: units over the square root of Σ(d − 1).
t_statnumbernullableThe t statistic with the sample variance.
p_value_exactnumbernullableThe exact (or z) tail alone.
p_value_clusterednumbernullableThe tail with bets clustered by slate and team-season (F23).
p_value_vs_vignumbernullableAgainst a random bet at the no-vig close (needs closes).
p_adjustednumbernullableThe p-value adjusted for the rules tried (Romano-Wolf on the ledger's rules, Bonferroni on the count otherwise; F14).
adjust_methodstringnullableromano_wolf, holm or none.
testsintegernullableRules tried in this family, this one included.
tests_effectivenumbernullableEffective tests (Li-Ji): near-duplicate rules count as fewer.
reality_check_pnumbernullableWhite's Reality Check p (when the ledger holds the rules' units).
spa_pnumbernullableHansen's SPA p (when the ledger holds the rules' units).
clears_t3booleannullablez at or above 3 (Harvey-Liu-Zhu's bar).
haircutnumbernullableHarvey-Liu haircut: the share of the ROI the rules tried take away.
haircut_roinumbernullableROI after the haircut, per decision.
anytime_validobjectnullableThe record as a bet against no skill (F17): e-value, anytime p, ROI range.
e_valuenumbernullableThe e-value now.
p_valuenumbernullable1 over the largest e-value so far.
max_enumbernullableThe largest e-value along the path.
roi_cs95array of numbernullableAn ROI range valid at every bet.
roi_ci95array of numbernullableROI's 95% score interval, clustered (F13, F23).
units_ci95array of numbernullableUnits' 95% interval: decisions times the ROI interval.
roi_ciobjectnullableROI intervals at ci_level: score (the one shown), BCa, block bootstrap over slates and anytime-valid (F13).
deflatedobjectnullablePer-bet Sharpe, PSR and DSR at the effective tests, and the bets needed: deflated (sample_needed), single test and 80% power, each times the design effect (F15).
sample_neededintegernullableDecisions needed, deflated for the rules tried; null with no edge to prove.
sample_needed_singleintegernullableDecisions needed for one rule tried.
sample_needed_power80integernullableDecisions for 80% power at this ROI.
sample_needed_planintegernullableDecisions for 80% power at plan_roi.
dsrnumbernullableDeflated Sharpe ratio.
psrnumbernullableProbabilistic Sharpe ratio.
sample_neededobjectnullableBets needed to tell this win rate from break-even (one-sided, with 80% power, ...).
small_samplebooleanUnder 30 decisions.
few_decisionsbooleanUnder 5 decisions.
honestobjectnullableThe honest forward ROI (F22, the headline ROI): empirical Bayes toward a prior mean (the rule's own closing-line value when at least half its bets have a close, else the usual two-way margin), less the winner's curse for the rules tried and the out-of-sample decay.
roinumbernullableHonest forward ROI.
ci80array of numbernullable80% range.
raw_roinumbernullableThe past ROI per decision, beside it.
prior_meannumbernullableThe prior mean the ROI is shrunk toward (prior.source says which).
prior_sdnumbernullableThe spread of true ROIs (0.02 until fitted on the ledger).
shrinknumbernullableWeight on the prior, se²/(se² + τ²).
selection_adjnumbernullableWinner's-curse correction for the rules tried.
decay_factornumbernullableOut-of-sample decay on the excess (0.74).
stabilizes_at_betsintegernullableDecisions at which the record and the prior weigh the same.
uses_clvbooleanThe prior mean is the rule's own closing-line value.
prior_basisstringfitted (prior_sd from the ledger's fit) or seed.
priorobjectWhere the prior mean comes from.
sourcestringclv (the rule's mean closing-line value), margin (the two-way margin, -4.55%) or fitted.
meannumbernullableThe prior mean used.
closesintegerDecided bets with a no-vig close.
cappedbooleanA closing-line value above 0 was held at 0: under 50 bets have a close.
wordsstringThe prior, said plainly.
dependenceobjectnullableBets clustered by slate (the date; the week in the NFL and college football) and team-season (F23).
clustersintegerSlates.
design_effectnumberClustered over iid variance.
effective_betsintegerDecisions over the design effect (at least 1).
methodstringcr1_twoway, cr1_slate, cr1_larger_oneway, wild_bootstrap or wild_bootstrap_webb.
baseobjectnullableThe rule against the rest of its own base (F19): the who-or-when conditions alone, or base_formula.
formulastringnullableThe base, or null for the whole market.
sourcestringauto, user or market.
One of: auto, user, market (open: new values may be added).
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
roinumbernullableThe base's ROI per decision.
liftnumbernullableRule ROI less base ROI.
p_value_vs_basenumbernullableSeason-stratified permutation p of the rule within its base.
units_from_basenumbernullableUnits the base alone would have made on the rule's stakes.
units_from_filternumbernullableThe rule's units less units_from_base.
clvobjectClosing-line value as evidence (F21): each bet's no-vig CLV, q_close x price - 1, on bets Kelly holds a close for. Empty when betting at the close.
bets_with_closeintegerBets with a close.
bets_without_closeintegerBets without one.
referencestringThe close used: consensus (the no-vig average across books).
devigstringHow the vig was removed: multiplicative.
lines_convertedintegerBets whose close was carried from the closing line to the bet's line (NFL spreads and totals).
coveragenumbernullableShare of bets with a close probability (null with no bets).
closes_staleintegerCloses left out as older than 60 minutes at the start, or seen after it (audit F9): those bets keep their result but have no CLV.
points_coveragenumbernullableShare of bets with points beaten (spreads and totals).
points_testobjectnullableThe clustered test on points beaten (audit F3).
betsintegerBets with points beaten.
meannumbernullableMean points the bet's line beat the close by.
std_error_clusterednumbernullableStandard error clustered by slate and team-season.
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean points beaten > 0.
cluster_methodstringHow the standard error was clustered.
basisstringWhat Beats the close is judged on: probability, or points when under 80% of bets have a close probability and more have points beaten (audit F3).
One of: probability, points (open: new values may be added).
meannumbernullableMean CLV (fraction): the bets' expected ROI if the close is right.
ci95array of numbernullableClustered 95% interval on the mean.
min items 2 · max items 2
p_valuenumbernullableOne-sided clustered p of mean CLV > 0.
beat_ratenumbernullableShare of bets that beat the close, ties left out.
beat_rate_ci95array of numbernullableWilson 95% interval on beat_rate.
min items 2 · max items 2
sign_test_pnumbernullableExact one-sided sign test against half.
mean_probnumbernullableMean of the close's no-vig probability less the bet moment's.
mean_pointsnumbernullableMean points the bet's line beat the closing line by (spreads and totals).
points_betsintegerBets mean_points is over.
profit_slopenumbernullablePer-bet units regressed on CLV: near 1 when CLV turned into profit.
profit_slope_senumbernullableIts slate-clustered standard error.
sample_neededintegernullableBets CLV needs for a 95% interval to clear 0 at this mean and spread (null unless the mean is above 0).
sample_needed_roiintegernullableBets profit would need at the same edge.
significanceobjectnullableThe CLV mean's tests (null under two bets).
tnumbernullableMean over its iid standard error.
ci95array of numbernullableStudent 95% interval (iid).
min items 2 · max items 2
std_errornumbernullableiid standard error.
bets_needednumbernullableBets a t of 2 needs at this mean and spread.
std_error_clusterednumbernullableStandard error clustered by slate and team-season (F23).
df_clusteredintegerDegrees of freedom: the fewer clusters less one.
ci95_clusteredarray of numbernullableClustered 95% interval.
min items 2 · max items 2
p_value_clusterednumbernullableOne-sided p of mean CLV > 0 (wild cluster bootstrap under 42 slates, else Student t); beats_close reads it.
cluster_methodstringHow the standard error was clustered.
design_effectnumbernullableClustered variance over iid variance.
notestringWhy the block is empty, when it is.
luckobjectThe luck-adjusted result (F20): the units the market's fair probabilities expected the decided bets to make against what they made. Pushes count zero both ways.
readingstringluck (bets before the close: the gap is outcome noise) or versus_the_market (bets at the close: the gap is the rule's edge on the closing market).
One of: luck, versus_the_market (open: new values may be added).
close_sourcestringconsensus.
devigstringmultiplicative.
betsintegerDecided bets with a fair probability.
bets_without_probintegerDecided bets without one.
left_outobjectBets left out.
pushintegerPushes.
halfintegerHalf results (quarter lines).
lines_convertedintegerBets whose close was carried to the bet's line.
expected_unitsnumbernullableSum of stake x (q x (price - 1) - (1 - q)).
actual_unitsnumbernullableUnits the bets made.
luck_unitsnumbernullableactual_units less expected_units.
luck_se_unitsnumbernullableluck's standard error given the probabilities.
expected_winsnumbernullableSum of q.
actual_winsintegernullableWins.
luck_percentilenumbernullableWhere the wins sit among the market's (Poisson-binomial, mid-p): 0.95 is luckier than (95) of (100) identical paths.
expected_roinumbernullableexpected_units over units risked.
actual_roinumbernullableactual_units over units risked.
thetanumbernullablePer-bet units regressed on expected units (30 bets or more): near 1 when the close is calibrated on these bets.
by_seasonarray of objectEach season, in calendar order; luck_units add up to the total.
seasonstringSeason.
betsintegerDecided bets with a probability.
expected_unitsnumberUnits the market expected.
actual_unitsnumberUnits made.
luck_unitsnumberactual_units less expected_units.
capacityobjectnullableF8: the limits Kelly measured at the bet moment (ADR-K783); evidence beside the verdict, never in it. No book publishes its limits, so only measured ones are shown.
availablebooleanWhether measured limits are shown for this market and moment.
marketstringnullableThe market as line history names it (spreads, totals, h2h).
before_startstringnullableThe time before the start the bet moment falls in.
One of: 0_1h, 1_12h, 12_36h, 36_72h, 72h_plus (open: new values may be added).
booksarray of objectPer book, highest median first.
bookstringnullableThe book as the vendor names it.
vendorstringnullableWho measured it.
currencystringnullableThe stake currency, null when the vendor doesn't say.
mediannumbernullableThe median limit.
p25numbernullableIts 25th percentile.
p75numbernullableIts 75th percentile.
minnumbernullableThe smallest.
maxnumbernullableThe largest.
snapshotsintegerLimit readings behind it.
gamesintegerGames behind it.
first_seenstringnullableThe first reading's date.
last_seenstringnullableThe last reading's date.
max_stake_per_betobjectnullableThe highest book median: the most one book typically took on one bet.
bookstringnullableThe book.
amountnumbernullableThe median limit.
currencystringnullableIts currency.
per_season_at_limitnumbernullablemax_stake_per_bet times the rule's bets a season.
bets_per_seasonnumbernullableThe rule's bets a season.
limits_heldintegerLimit readings Kelly holds for this market and moment.
reasonstringnullableWhy none are shown, when available is false.
stressobjectThe same bets graded harder.
half_point_worseobjectnullableEvery line half a point worse (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
flat_110objectnullableEvery bet at -110 (none for moneylines).
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
other_sideobjectThe other side of every bet.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
break_evenobjectnullableF7: how much worse the price or line can get before the profit is gone.
price_multipliernumbernullableThe share of each payout (d - 1) that breaks even (under 1: the room to spare).
american_avgnumbernullableThe flat American price that breaks even.
decimal_avgnumbernullableThe same, decimal.
line_pointsnumbernullableThe most points worse (in halves, up to 3) the line could be with a profit left; 0 when none; null for moneylines.
gridarray of objectnullableF7: the bets at -105 to -120 with the line 0 to 1 point worse (moneylines: payouts cut 0-6%); left out at detail summary.
vigintegernullableEvery bet at this American price.
line_shiftnumbernullablePoints the line moved (negative: worse).
price_multipliernumbernullableMoneylines: the share of each payout kept.
unitsnumbernullableUnits.
roinumbernullableROI.
slippage_mcobjectnullableF7: each bet's line worsened by a draw from the market's own Monday-to-close moves, 1,000 seeded runs; null without such moves.
runsintegerRuns.
roi_p05numbernullable5th percentile ROI.
roi_p50numbernullableMedian ROI.
moves_fromintegerLine moves the draws come from.
move_mediannumbernullableTheir median size, points.
price_worseobjectnullableTeasers: every ticket ten cents worse (price_american). Racing: every bet's winnings 5% smaller.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
legsobjectnullableTeasers: every teased leg, judged against the per-leg break-even (1/D)^(1/n).
recordstringLegs won-lost(-pushed).
decisionsintegerLegs won plus lost.
cover_ratenumbernullableLegs won over legs decided.
break_even_per_legnumberThe cover rate a leg needs at the ticket's price.
significanceobjectnullableThe legs' record test against the per-leg break-even.
sample_neededobjectnullableLegs needed to tell this cover rate from the break-even.
random_baselineobjectnullableF18: the rule against random bets of the same count on the same games, matched by season and price bucket.
simsintegerRandom rules drawn.
betsintegerThe rule's bets.
poolintegerBets the market offered.
percentilenumbernullableShare of random rules with a lower ROI.
beats_per_100integerRandom rules out of 100 the rule beats.
p_valuenumbernullable(1 + random rules at or above) / (sims + 1).
analytic_pnumbernullableThe finite-population normal cross-check.
random_roi_mediannumbernullableMedian random ROI (about minus the vig).
random_roi_p05numbernullable5th percentile random ROI.
random_roi_p95numbernullable95th percentile.
best_of_testsobjectWould the best of M random rules (M = rules tried) do this well?
testsintegerM.
p_valuenumbernullable1 - F(ROI)^M.
histogramobjectnullableRandom ROIs in 30 bins (edges, counts); full detail only.
decayobjectnullableF24: is the rule still making money? Rolling ROI and CLV, one unknown break (sup-F, block permutation p), a CUSUM alarm, a known-date test, the trend and the decay ratio.
betsintegerGraded bets.
windowintegernullableRolling window, bets.
rollingarray of objectnullableUp to 200 points; full detail only.
breakobjectnullableThe most likely date the edge changed.
atstringDate.
bet_indexintegerBets before it.
sup_fnumbernullablesup-F.
p_valuenumbernullableBlock permutation p-value.
at_edgebooleanWithin the first or last 15%: not a confirmed break.
roi_beforenumbernullableROI before.
roi_afternumbernullableROI after.
bets_afterintegerBets after.
cusumobjectnullablePage's lower CUSUM alarm.
known_breakobjectnullableA Chow test at break_date.
trend_per_100_betsnumbernullableProfit per bet's change per 100 bets (1 unit risked).
decay_rationumbernullableROI after the break (or the newest 30%) over ROI before.
clv_breakobjectnullableThe break test on CLV.
roi_newest_30numbernullableROI of the newest 30% of bets.
ablationobjectnullableF27: each top-level AND condition's Shapley share of the units beyond the base's ROI.
baseobjectEvery game with a line: decisions and ROI.
totalnumbernullableThe rule's units beyond the base's ROI (the Shapley values add up to it).
clausesarray of objectOne a condition, in the order written.
conditionstringThe condition.
shapley_unitsnumbernullableIts share of the extra units.
labelstringdead, harmful, load_bearing or helps.
One of: dead, harmful, load_bearing, helps (open: new values may be added).
fixobjectnullableRemove it: the rule without it.
pairsarray of objectPairs (up to 6 conditions).
methodstringexact or sampled.
One of: exact, sampled (open: new values may be added).
nearbyobjectnullableF28: every number in the rule moved two steps each way (steps fixed by the field).
numbersarray of objectThe rule's numbers.
variantsintegerVariants in the grid.
distinct_bet_setsintegerVariants with their own bets.
roi_mediannumbernullableMedian ROI of the variants.
roi_p05numbernullable5th percentile.
roi_p95numbernullable95th percentile.
spike_rationumbernullable(ROI - median) / IQR of the variants.
readingstringnullableplateau, slope or spike.
nastringnullableWhy there are no numbers to move.
monte_carloobjectnullableF31: drawdowns and losing runs the same bets give in other orders.
simsintegerSimulations.
modestringblock, iid or permutation.
max_drawdown_unitsobjectMax drawdown, units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
longest_losing_runobjectLongest losing run, bets.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
final_unitsobjectFinal units.
p05numbernullable5th percentile.
p50numbernullableMedian.
p95numbernullable95th percentile.
actualnumbernullableThe real order's figure.
prob_below_zeronumbernullableShare of runs finishing below zero.
worst_100_bets_units_p05numbernullable5th percentile of the worst 100-bet stretch.
noiseobjectnullableF32: jittered inputs, lines moved half a point either way, and an earlier bet moment.
inputsobjectnullableThe rule's compared numbers jittered.
runsintegerRuns.
profitable_runsintegerRuns with a profit.
roi_mediannumbernullableMedian ROI.
roi_p05numbernullable5th percentile ROI.
overlap_mediannumbernullableMedian Jaccard overlap of the bets with the rule's.
linesobjectnullableLines (moneylines: payouts) moved at random.
delayobjectnullableThe rule bet at the earlier moment.
concentrationobjectnullableF34: the profit without the best season, team, opponent, month and weekday.
byarray of objectGroupings in a fixed order.
groupstringseason, team, opponent, month or weekday.
bestobjectnullableThe best group (3+ bets).
without_bestobjectnullableThe rule without it.
effective_groupsnumbernullable1 / Herfindahl of the profitable groups' shares.
p_more_concentratednumbernullablePermutation p-value (teams and opponents).
trim_top_0_05objectnullableWithout the top 5% of bets (moneylines) or bet-days.
split_dimensionsarray of objectnullableF35: the rule by fixed dimensions with shrunk ROI and intervals.
dimensionstringThe dimension.
heterogeneity_pnumbernullableChi-square p-value of win rates across its levels.
heterogeneity_qnumbernullableBenjamini-Hochberg q across the dimensions.
real_differencebooleanq < 0.10.
where_profitstringnullableThe level with the most shrunk extra units, when real.
cellsarray of objectnullableLevels in a fixed order; full detail only.
robustnessobjectThe suite's run: the verdict v2 checks it feeds and its seed.
checksarray of objecthalf_point_worse, nearby_holds, not_concentrated and still_working.
keystringrequiredThe check.
One of: half_point_worse, nearby_holds, not_concentrated, still_working (open: new values may be added).
labelstringrequiredThe check, in words.
statusstringrequiredpass, fail or na.
One of: pass, fail, na (open: new values may be added).
valuestringrequiredThe figure behind it, in words.
statisticnumbernullableThe figure the status compares.
thresholdnumbernullableWhat it is compared with.
fast_pathbooleanThe rule's numbers were re-bet from its compiled columns.
seedintegerThe seed (a hash of the rule and its options).
evaluationsintegerFormula evaluations the suite ran.
offarray of stringBlocks switched off.
pathobjectThe units path in date order: drawdown and longest losing run; at detail full its series (F47).
max_drawdown_unitsnumberThe deepest fall from a peak, units.
longest_losing_runintegerMost losses in a row (pushes don't break it).
seriesobjectnullableBet-indexed lines of the graded bets (detail full): index the bet numbers kept (past 600 bets thinned by LTTB, the deepest drawdown always kept), dates, units, drawdown, base_units (the market base's ROI at the same stakes, null unless measured), band_p025 and band_p975 (the no-skill range: 1,000 seeded runs of the same bets at their fair chances; null under 20 bets or for racing), rolling_roi_100 with its no-skill range rolling_p025 and rolling_p975, and markers (season starts, the holdout).
xstringAlways bet_index.
One of: bet_index (open: new values may be added).
indexarray of integerBet numbers (1 is the first graded bet).
datesarray of stringEach kept bet's date (YYYY-MM-DD, UTC).
unitsarray of numberUnits to date.
drawdownarray of numberUnits below the running peak (0 or below).
base_unitsarray of numbernullableThe market base at the same stakes.
band_p025array of numbernullableNo-skill 2.5th percentile.
band_p975array of numbernullableNo-skill 97.5th percentile.
rolling_roi_100array of numberROI over the last 100 bets (null before).
rolling_p025array of numberIts no-skill 2.5th percentile.
rolling_p975array of numberIts no-skill 97.5th percentile.
markersarray of objectWhere each season and the holdout start.
typestringseason or holdout.
One of: season, holdout (open: new values may be added).
atintegerThe bet number it starts at.
labelstringnullableThe season, or newest 30%.
downsampledbooleanThe series were thinned to 600 points.
expected_max_drawdown_zero_edgenumbernullableThe median deepest drawdown of the no-skill runs, units.
holdoutobjectnullableThe latest 30% of bets against the earlier 70%.
seasonsarray of objectSeasons in calendar order.
splitsarray of objectnullableHome, away, favorite, underdog and line bands, in a fixed order; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
withoutarray of objectnullableOne row a top-level AND condition: the rule's record without it; null at detail summary.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
p_value_vs_rulenumbernullableThe rule against the bets the condition removes (F19).
p_value_vs_rule_adjustednumbernullablep_value_vs_rule, Holm-adjusted over the conditions shown.
betsarray of objectEvery bet in date order (left out at detail summary).
keystring{game_id}:{team_id} (racing: {race_id}:{runner_id}).
starts_atstringKickoff (RFC 3339).
format date-time
seasonstringThe season.
labelstringDate, team and opponent.
priceobjectnullableThe price at the bet moment, decimal.
linenumbernullableThe bet's line as graded (the team's spread, the total).
resultstringnullablewin, loss or push (soccer quarter lines: half_win or half_loss); null until played.
close_probnumbernullableThe fair closing probability of the same side, for CLV.
units_to_datenumberUnits after this bet, in date order.
how_gradedstringnullableHow the bet was graded, in words.
unitsnumbernullableUnits the bet won or lost at the run's staking (racing: its exact profit or loss with dead heats, each way, Rule 4 and commission); null until played.
drawdown_to_datenumbernullableUnits to date less their running peak (0 or below); null until played.
sidestringnullablehome or away (spread, moneyline), over, under or draw.
marketstringThe market graded (the other total when fading a total).
bet_momentstringWhen the bet was taken: edition, day_before or close.
clv_probnumbernullableThe fair closing chance less the fair chance at the bet moment, when Kelly holds both.
samplestringnullableholdout for the latest 30% of graded bets by time, else in_sample; null with no holdout or until played.
One of: in_sample, holdout (open: new values may be added).
whyarray of objectWhy it matched: up to six conditions (F46).
clausestringThe condition as written.
valuenumber | boolean | stringnullableThe compared expression's number for a comparison (rest_days - opponent.rest_days for rest_days - opponent.rest_days >= 3), else true or false.
settlementarray of stringThe settlement rules that decided the bet (IDs in settlement.rules).
leak_testobjectThe leak test (F3): the rule and each part of it evaluated again on a sample of rows with the data cut at the bet moment. A leak refuses the test (422 rule_not_testable, problem leak_found), so a report always says clean or not_run.
statusstringrequiredclean, leak_found or not_run.
One of: clean, leak_found, not_run (open: new values may be added).
rows_checkedintegerrequiredRows rebuilt at the bet moment.
matched_checkedintegerOf them, rows the rule matched.
matches_flippedintegerRows whose match changed with the data cut.
partsarray of objectrequiredParts that changed, in the formula's order (empty when clean).
namestringThe part, as written (rule is the whole rule).
rows_changedintegerRows where it changed.
max_changenumbernullableThe largest change in a number.
examplestringOne changed row, {game_id}:{team_id}.
warm_upobjectThe warm-up check (F4): bets taken before the rule's windows had the earlier games they read (one season at a time; windows never reach into last season).
policystringThe policy graded in the report.
One of: as_written, require_full (open: new values may be added).
games_neededintegerEarlier games every window needs (0: the rule has no window).
windowsarray of objectEach window in the formula's order.
exprstringThe window, as written.
games_neededintegernullableEarlier games it reads (null: season to date).
short_rowsintegerRows in scope with fewer earlier games this season.
short_betsintegerBets the rule took with fewer.
short_betsintegerBets taken with a window short.
by_policyarray of objectnullableas_written, then require_full; null when the rule has no window.
recordstringWins-losses(-pushes).
betsintegerBets graded or not.
decisionsintegerWins plus losses.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
unitsnumbernullableUnits won or lost at the staking asked for.
roinumbernullableUnits over units risked.
p_valuenumbernullableThe record's own p-value against no skill at its prices.
p_adjustednumbernullableOn splits and condition-removed rows: p_value adjusted (Holm) for every such row the report shows.
shown_rowsintegerOn splits and condition-removed rows: the rows p_adjusted counts.
warningbooleanShort bets are over 10% of the bets, or leaving them out moves ROI by more than half its 95% interval's half-width.
settlementobjectHow bets were settled (F9): Kelly's own grading rules as data, and each rule's bets.
profilestringThe rules profile (kelly_standard).
rules_versionstringThe rules table's date.
appliedarray of objectRules that decided at least one bet, in table order.
rulestringRule ID.
betsintegerBets.
rulesarray of objectThe rules for this league and market.
idstringRule ID.
textstringThe rule in words.
effective_fromstringThe date Kelly's grader took it up.
coverageobjectCoverage (F10): from every game in scope to the bets graded, each season's priced share, blank inputs and whether the rows left out for no line differ from the bets kept.
flowarray of objectThe steps in order; rule_true rows = decisions + every row out after it.
stepstringin_scope, rule_true, one_bet_a_game, full_windows, priced, graded or decisions.
gamesintegerGames (in_scope only).
rowsintegerRows.
outobjectRows that left at this step, by reason.
by_seasonarray of objectEvery season asked for, in order.
seasonstringSeason.
gamesintegerGames in scope.
rowsintegerRows.
rule_trueintegerRows the rule matched.
betsintegerBets graded.
inputs_missingarray of objectInputs read bare on the row, with blanks.
attributestringThe input.
rows_blankintegerRows with no value.
left_out_differsobjectRows left out for no line against the bets kept: standardized mean difference over home, neutral site and rest days; flagged over 0.1 with 10 or more left out.
flagbooleanThey differ.
max_smdnumbernullableThe largest difference.
covariatestringnullableWhere it is.
left_outintegerRows.
keptintegerRows.
by_covariatearray of objectEach covariate's difference.
momentobjectThe bet moment's clock (F5): its target time, the prices' snapshot age at it and where the lines came from.
bet_momentstringedition, day_before or close.
targetstringnullableThe moment's time for a game, in words.
max_stale_minutesintegerOldest price kept, in minutes before the target.
floor_minutes_before_oct_2022integerThe limit's floor for games before October 2022 (the Odds API's 10-minute snapshots).
rows_pricedintegerRows the rule took that had a price.
rows_agedintegerRows whose price carries its snapshot time and was kept.
median_snapshot_age_minutesnumbernullableMedian minutes from snapshot to target, kept rows.
max_snapshot_age_minutesnumbernullableOldest kept snapshot, in minutes before the target.
stale_priceintegerRows left out: price older than max_stale_minutes.
after_momentintegerRows left out: price seen after the target.
age_unknownintegerRows kept whose price has no snapshot time (Kelly's own frozen lines).
line_sourcesobjectKept rows by line source (sheet_line, odds_api_early, odds_api_close).
left_outobjectGames the rule matched but didn't bet.
no_lineintegerNo Kelly line at the bet moment.
both_sidesintegerThe rule matched both teams.
stale_priceintegerThe price was older than max_stale_minutes at the bet moment, or seen after it (F5).
unpairedintegerTeasers: legs left off a ticket (a week with too few, or the leftover in chronological pairing).
no_priceintegerRacing: runners with no price to settle at.
non_runnersintegerRacing: non-runners (void).
void_racesintegerRacing: abandoned or void races.
riskobjectRisk of ruin (F36, kmath.backtest.rule_ruin): the chance of losing ruin_share of the bankroll within horizon_bets bets betting like the record, with its win rate taken as exact and allowing for its sample (a Beta posterior draw a path); the Lundberg bound; the bankroll curve.
bankroll_unitsnumberBankroll in units.
horizon_betsintegerBets ahead.
ruin_if_rate_exactnumbernullableRuin chance if the record's win rate were exact.
ruin_allowing_errornumbernullableRuin chance allowing for how little the record's sample tells.
lose_all_evernumbernullableLundberg bound on ever losing the whole bankroll at the exact rate.
lose_all_ever_allowing_errornumbernullableThe same, averaged over the win rate's posterior.
drawdown_p95_unitsnumbernullable95th percentile of the worst drop over the horizon, units.
down_after_horizonnumbernullableChance of being down after the horizon.
bankroll_for_ruin_005_unitsnumbernullableSmallest bankroll whose ruin chance is at most 0.05.
curvearray of objectnullableRuin chance by bankroll size; null at detail summary.
statusstringok, or too_few (under 30 decisions: closed form only).
One of: ok, too_few (open: new values may be added).
staking_plansobjectStaking plans replayed over the real bet order (F37, kmath.backtest.staking_replay). The Kelly criterion's fractions and the kelly plan are stake mode: on askkelly.ai only, for a signed-in, attested person; elsewhere null with withheld (D34).
kellyobjectThe Kelly criterion at the record's win rate and mean price, the Baker-McHale multiple for its sample size and the fraction bet.
full_fractionnumbernullableFull Kelly criterion (stake mode; null off askkelly.ai).
shrinknumbernullableThe shrink for the sample size.
honest_fractionnumbernullableThe fraction bet: shrink × full, capped; 0 unless the verdict is promising or holds up (stake mode; null off askkelly.ai).
withheldstringWhy the stake-mode figures are null.
One of: stake_first_party_only (open: new values may be added).
plansarray of objectEach plan asked for, in the order asked.
planstringflat, to_win, percent or kelly.
One of: flat, to_win, percent, kelly (open: new values may be added).
risk_adjustedobjectRisk-adjusted stats per unit risked (F40, kmath.backtest.risk_adjusted), with 95% intervals from a day-block bootstrap; at detail summary only sharpe_per_bet and profit_factor.
sharpe_per_betnumbernullableMean return over its standard deviation, a bet.
profit_factornumbernullableUnits won over units lost (null with no losing bet).
capsobjectnullableExposure caps applied in time order (F39); the record and checks are on the capped ledger.
skippedintegerBets the caps skipped.
by_reasonobjectSkipped bets by cap.
uncappedobjectThe record without the caps.
random_priority_mean_roinumbernullableMean ROI over random tie-breaks.
ruleobjectThe rule and the options it was tested with.
rule_hashstringThe rule's identity: SHA-256 of its canonical formula (spacing, case and the order of AND and OR conditions don't change it) with the competition, market, side, bet moment, seasons, season type, stake and price. Detail and tests are not part of it.
pattern ^sha256:[0-9a-f]{64}$
translation_idstringThe rules:translate result this formula came from (F43), when the request named one.
pattern ^tx_[0-9a-hjkmnp-tv-z]{20}$
readsarray of objectEvery name the rule reads, once.
namestringThe attribute, with its relation (opponent.rest_days).
knownstringIts moment: static, pregame, edition, close or final (a window's input is read before the game).
windowedbooleanA backward window reads it (earlier games only).
formula_versionstringThe formula language version the rule was read with (F45).
catalogue_versionstringnullableThe stat catalogue version its names were bound against.
reportobjectA shared report's own facts (get_rule_report only, F54).
report_idstringThe report.
report_urlstringIts page.
format uri
visibilitystringunlisted or public.
One of: unlisted, public (open: new values may be added).
author_rules_triedintegerThe author's rules tried on this market when shared.
frozenbooleanAlways true.
rerunobjectThe run's request, to test it again on today's data.
report_urlstringnullableA frozen report of this run anyone with the link can open (F54): only when the request asked (share) and the call has an account; else null. share_rule_run makes or changes one later.
format uri
chartsarray of objectOnly when the request asked (charts) at detail full: equity against the no-skill band, the drawdown and the rolling ROI as Kelly Chart Specs (F47).
transferobjectnullableOther markets and sports (F33): the same rule on its other market and on other leagues, numbers mapped to the same meaning, pooled with DerSimonian-Laird random effects. Evidence beside the verdict, never part of it; these runs don't count toward rules tried. Present only when the request asks for transfer.
statusstringmeasured or na.
One of: measured, na (open: new values may be added).
whystringnullableWhy nothing ran, when nothing did.
optionsobjectThe leagues, markets and thresholds asked for.
cellsarray of objectEach market and league in a fixed order (the other market first, then sibling leagues), never sorted by a result; up to 12.
competitionstringThe league.
marketstringThe market.
thresholdsstringliteral (the rule's own numbers) or quantile (each number moved to the same percentile of its field in this league).
One of: literal, quantile (open: new values may be added).
formulastringThe formula this cell ran (quantile cells carry the moved numbers).
seasonsarray of stringSeasons read, as this league labels them.
mappedarray of objectnullableQuantile cells only.
numberstringThe number as written.
fieldstringWhat it is compared with.
percentilenumbernullableIts percentile in the rule's own league (a fraction).
valuenumbernullableThe number at the same percentile here.
decisionsintegerWins plus losses.
betsintegerGraded bets.
recordstringWins-losses(-pushes).
unitsnumbernullableUnits won or lost.
roinumbernullableUnits over units risked.
roi_ci95array of numbernullableROI's 95% score interval.
senumbernullableROI's standard error at 1 unit risked a bet.
p_valuenumbernullableThe cell's own one-sided p against no skill at its prices.
p_holmnumbernullableHolm's adjusted p across the cells.
pooledbooleanThe cell is in the pool (another league, one a league, 10+ decisions).
pooledobjectnullableThe pool of the other leagues' cells.
roinumbernullableRandom-effects ROI.
roi_ci95array of numbernullableIts 95% interval.
senumbernullableIts standard error.
p_valuenumbernullableTwo-sided p of the pooled ROI.
i2numbernullableI²: the share of the spread across leagues beyond chance (a fraction).
tau2numbernullableτ²: the between-league variance of true ROI.
qnumbernullableCochran's Q.
cellsintegerCells pooled.
same_directionobjectnullablePooled cells whose ROI has the rule's own sign.
agreeintegerCells the same way.
cellsintegerCells.
p_valuenumbernullableTwo-sided sign-test p.
textstring"4 of 5".
not_testablearray of objectLeagues and markets left out, with why.
competitionstringThe league.
marketstringnullableThe market.
missingarray of stringFields that league lacks.
whystringWhy.
notestringWhat the panel is and isn't.
run_idstringnullableThis run, kept on the account (run_…): the same request on the same data is the same run. GET /v1/rules/runs/{run_id} reads it back. Null without an account.
pattern ^run_[0-9a-hjkmnp-tv-z]{20}$
data_versionstringThe data the run read: Kelly's lake snapshot (or the data's as-of time). A new snapshot is a new run of the same rule.
manifestobjectnullableWhat this run read and ran, so the same request can be run again and checked against it (ADR-K885). No prices, lines or rows of source data.
versionintegerThe manifest's own layout version.
codeobjectThe code that graded the run.
engineintegerThe tester's revision; it rises with any fix that can move a result on the same data.
releasestringnullableThe deployed commit.
formula_versionstringnullableThe formula language's version.
catalogue_versionstringnullableThe stat catalogue's version.
attributes_sha256stringnullableA hash of the attribute registry: what each input means and when it is known.
settlement_rulesstringnullableThe settlement rules' version.
dataobjectThe data the run read.
versionstringThe same value as data_version.
snapshotsarray of stringThe lake snapshots read, in order.
one_snapshotbooleanWhether every season was read on one snapshot.
as_ofstringnullableWhen the read was made (RFC 3339).
format date-time
data_throughobjectHow far Kelly's history went, by league.
latest_resultstringnullableThe newest game read that had a result.
rows_built_atstringnullableWhen the prepared rows were built, where the run read them.
format date-time
queriesobjectEach reviewed lake query the tester reads, with a hash of its text.
requestobjectThe request.
request_hashstringThe request's hash.
rule_hashstringThe rule's hash.
seedintegerThe seed every simulation in the run used.
sourcesobjectWhere the run's inputs came from.
pricesobjectBets priced, by the source of the line.
inputsarray of objectEach input the rule read, with the moment it is known.
namestringThe input.
knownstringnullableThe moment it is known.
seasonsarray of objectEach season read.
seasonstringThe season.
rowsintegerTeam-games in scope.
left_outobjectMatched games that were not bet, by reason.
sha256stringThe hash of this block. Two runs with the same hash read the same data with the same code.
since_runobjectnullableOnly on a run read back: what has changed in Kelly's tester since the run was kept.
statusstringcurrent: nothing that can move its numbers changed. superseded: something did. no_manifest: the run was kept before manifests.
One of: current, superseded, no_manifest (open: new values may be added).
engine_changedbooleannullableWhether the tester's revision has risen since.
changedarray of objectThe parts that differ, each with its value then (from) and now (to).
partstringengine, release, formula_language, catalogue, attributes, settlement_rules or queries.
tested_atstringWhen the run was kept (RFC 3339); only on a run read back.
format date-time
rule_versionobjectnullableThe saved rule this test was kept under (save, or rule_id): its rule_id, name, version and versions_tried. Null when the test wasn't saved.
rule_idstringrequiredThe saved rule (rl_…).
pattern ^rl_[0-9a-hjkmnp-tv-z]{20}$
namestringIts name.
versionintegerrequiredThis rule's version under it: the same rule again keeps its version.
min 1
new_versionbooleanThis test made a new version.
versions_triedintegerrequiredVersions saved under it, this one included.
min 1
since_savedobjectnullableThe saved version's record since it was saved (backtester T6, F50), when the test names its rule_id and repeats that version: verdict v2's since_saved check reads it. Null for a new version (its clock starts now) or an unsaved test.
saved_atstringnullableWhen the clock started: the version's save, or the registration (RFC 3339).
format date-time
recordstringrequiredWins-losses(-pushes) since then.
winsintegerWins.
lossesintegerLosses.
pushesintegerPushes.
decisionsintegerrequiredWins plus losses.
betsintegerGraded bets.
unitsnumberrequiredUnits won or lost.
roinumbernullableUnits over units risked, a fraction.
clv_meannumbernullableMean closing-line value of the bets (null at the close: none by definition).
clv_betsintegerBets with a closing-line value.
e_valuenumbernullableAnytime-valid e-value of the forward bets against no edge at the prices taken (safe to read every day).
max_e_valuenumbernullableIts running maximum.
e_thresholdnumberThe e-value that counts as evidence at alpha 0.05 (20).
vs_backtestobjectnullableIs live consistent with the backtest?
win_rate_backtestnumberThe backtest's win rate.
win_rate_forwardnumbernullableThe win rate since the save.
p_value_differencenumbernullableTwo-sided p-value that the two win rates differ.
methodstringnullablefisher_exact (under 30 forward decisions) or two_proportion_z.
e_value_worsenumberE-value that live wins less often than the backtest; 20 or more fails the check.
pendingintegerBets recorded at their bet moment, not yet final.
no_lineintegerMatches Kelly held no line for at the bet moment (kept, never skipped).
voidintegerPostponed games past the window.
statusstringrequiredThe since_saved check: na until min_decisions, then pass or fail.
One of: na, pass, fail (open: new values may be added).
min_decisionsintegerForward decisions before the check is judged (30).
last_bet_atstringnullableThe latest bet moment recorded.
format date-time
tests_on_accountobjectnullableRules this account has tried on this competition and market, this one included (the rules-tried ledger: every distinct rule tested on the site, the API or Kelly MCP, kept for good). The not-luck check adjusts for the larger of this count and the request's own tests. Null when the call has no account.
competitionstringrequiredThe competition.
marketstringrequiredThe market.
countintegerrequiredDistinct rules tried.
min 1
shown_countedintegerVariants an earlier report showed (its condition-removed forms, nearby numbers and home or away splits) that joined the count because this rule is one of them (audit F6).
stakingstringHow each bet was staked.
matchedintegerBets taken.
ungradedintegerBets not yet played.
noticestringThis tests a rule on past games; it isn't a pick.
calibrationobjectnullableoutput: probability only. The formula's chances against the market's no-vig close of the same side (moneyline: the team wins; spread: it covers the bet's line), on the newest 30% of the scored team-games (whole days), and the bets it makes at the edge threshold.
formulastringThe probability formula as sent.
bet_formulastringThe rule it bets: the chance at the bet's own price worth more than the edge threshold (the rule tested, counted and kept).
edge_thresholdnumberExpected return a bet needs, p × decimal price − 1 (0.02 by default).
basisstringnewest_30: scored on the holdout only.
One of: newest_30 (open: new values may be added).
gamesintegerGames scored.
rowsintegerTeam-games scored (both teams of a game).
in_sample_rowsintegerTeam-games before the holdout, not scored.
holdout_fromstringnullableThe holdout's first day.
left_outobjectTeam-games not scored.
pushesintegerPushes and half results.
outside_0_1integerValues below 0 or above 1.
no_valueintegerRows the formula gave no number for.
clippedintegerValues of 0 or 1 clipped to 1e-6 from the edge.
briernumbernullableBrier score of the formula.
brier_marketnumbernullableBrier score of the close.
brier_skillnumbernullable1 − formula / close (above 0: the formula scores better).
log_lossnumbernullableLog loss of the formula.
log_loss_marketnumbernullableLog loss of the close.
log_loss_skillnumbernullable1 − formula / close.
dm_p_valuenumbernullableDiebold–Mariano p on the per-game log-loss difference (two-sided).
dm_statisticnumbernullableThe HLN-corrected statistic (positive: the formula's loss is higher).
mcbnumbernullableCORP miscalibration (Brier).
dscnumbernullableCORP discrimination.
uncnumbernullableUncertainty; brier = mcb − dsc + unc.
mcb_p_valuenumbernullableThe MCB test: the chance a calibrated formula looks this miscalibrated (999 resamples).
reliabilityarray of objectnullableThe CORP reliability curve at up to 20 forecast values.
pnumbernullableThe formula's probability.
observednumbernullableHow often it happened (PAV fit).
lonumbernullable90% consistency band, low.
hinumbernullable90% consistency band, high.
ecenumbernullableExpected calibration error over ten equal-width bins (bin-dependent; for familiarity).
distinct_forecastsintegerDistinct values the formula gave.
slopenumbernullableCalibration slope b in y ~ a + b·logit p (1 is ideal; under 1 overconfident).
slope_senumbernullableIts game-clustered standard error.
interceptnumbernullableCalibration intercept a (0 is ideal).
encompassingobjectnullabley ~ a + b_q·logit q + b_p·logit p, game-clustered: does the formula know something the close doesn't?
b_modelnumbernullableb_p.
b_marketnumbernullableb_q.
se_modelnumbernullableb_p's standard error.
p_valuenumbernullableTwo-sided p of b_p.
edge_binsarray of objectnullableRealized results against the formula's edge.
edgenumbernullableMean of p − q in the bin.
realized_minus_marketnumbernullableMean of y − q in the bin.
rowsintegerTeam-games in the bin.
statusstringThe calibrated check: pass, fail or na (under 200 games, or one value).
One of: pass, fail, na (open: new values may be added).
valuestringThe check's line.
statisticnumbernullableThe encompassing p.
thresholdnumbernullable0.05.
summarystringrequiredOne or two sentences with the key figure, written by a server template; when the request could mean two things, the question to ask.
assumptionsarray of stringrequiredEvery default applied (regular season, the qualifier used, ...).
has_morebooleanrequiredMore rows follow.
next_cursorstringnullablePass it back as cursor for the next page.
as_ofstringrequiredWhen the value was true; for a derived value, its newest input's time.
format date-time
sourcestringrequiredWho produced the value, as shown to readers.
min length 1
attributionstringnullableAttribution the source's terms require.
request_idstringrequirednullableThe request that produced this response.
live_topicsarray of stringKelly Wire topics that update this resource (API and cards only).
rg_linestringnullableThe cost line (COMP-1, D19) on every resource and result that touches wagering: one engine-computed cost fact, or the short form, plus 21+ and the helpline. Show it with the value. Null on resources that don't touch wagering.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · rule_not_testable · query_too_broad · snapshot_expired · upstream_timeout
POST https://api.askkelly.ai/v1/trends:audit
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "=home",
"competition": "nba",
"seasons": [
"2025-26"
],
"claimed_record": "sixty and forty"
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "claimed_record: 'sixty and forty' does not match '^\\\\s*\\\\d{1,6}\\\\s*-\\\\s*\\\\d{1,6}(\\\\s*-\\\\s*\\\\d{1,6})?\\\\s*$'",
"param": "claimed_record",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_78dd0ee167e1a78b243b966e",
"retryable": false,
"display_message": "Something in that request was not understood."
}POST https://api.askkelly.ai/v1/trends:audit
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "=AND(home, rest_days >= 2)",
"competition": "nba",
"seasons": [
"2025-26"
],
"claimed_record": "60-40",
"published_on": "2025-11-01"
}
{
"object": "trend_audit",
"status": "tested",
"claim": {
"text": "=AND(home, rest_days >= 2)",
"claimed_record": "60-40",
"claimed_roi": null,
"published_on": "2025-11-01"
},
"translation": {
"formula": "=AND(home, rest_days >= 2)",
"readback": "Bets the team against the spread when it is at home, and its rest in days is at least 2.",
"translation_id": "tx_xbj639zqrcq2kg653rsf"
},
"formula": "=AND(home, rest_days >= 2)",
"claim_check": {
"claimed_record": "60-40",
"claimed_roi": 0.14545454545454548,
"claimed_roi_basis": "the claimed record at -110",
"kelly_record": "0-0",
"kelly_roi": null,
"claim_inflation_roi": null
},
"since_published": {
"from": "2025-11-02",
"first_game": null,
"record": "0-0",
"decisions": 0,
"units": 0,
"roi": null
},
"coverage": {
"seasons_tested": [
"2025-26"
],
"held_from": "2020-06-01",
"note": "Kelly holds closes from June 2020; seasons a claim counts before that can't be checked."
},
"test": {
"staking": "flat, 1 unit risked a bet",
"matched": 0,
"ungraded": 0,
"overall": {
"wins": 0,
"losses": 0,
"pushes": 0,
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"win_rate": null,
"break_even": null,
"significance": null,
"few_decisions": true,
"small_sample": true,
"deflated": null
},
"clv": {
"bets_with_close": 0,
"bets_without_close": 0,
"reference": "consensus",
"devig": "multiplicative",
"lines_converted": 0,
"coverage": null,
"closes_stale": 0,
"points_coverage": null,
"points_test": null,
"basis": "probability",
"mean": null,
"ci95": null,
"p_value": null,
"beat_close_share": null,
"beat_rate": null,
"beat_rate_ci95": null,
"sign_test_p": null,
"mean_prob": null,
"mean_points": null,
"points_bets": 0,
"profit_slope": null,
"profit_slope_se": null,
"sample_needed": null,
"sample_needed_roi": null,
"significance": null,
"note": "Bets at the close have no closing-line value by definition."
},
"seasons": [],
"holdout": null,
"dependence": null,
"honest": null,
"luck": {
"reading": "versus_the_market",
"close_source": "consensus",
"devig": "multiplicative",
"bets_without_prob": 0,
"left_out": {
"push": 0,
"half": 0
},
"lines_converted": 0,
"bets": 0,
"expected_units": null,
"actual_units": null,
"luck_units": null,
"luck_se_units": null,
"expected_wins": null,
"actual_wins": null,
"luck_percentile": null,
"expected_roi": null,
"actual_roi": null,
"theta": null,
"by_season": []
},
"base": {
"formula": "=home",
"source": "auto",
"decisions": 1,
"status": "na",
"why": "the base is under twice the rule's bets"
},
"stress": {
"half_point_worse": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"flat_110": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"other_side": {
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null
},
"break_even": null,
"slippage_mc": null,
"grid": [
{
"vig": -105,
"line_shift": 0,
"units": 0,
"roi": null
},
{
"vig": -105,
"line_shift": -0.5,
"units": 0,
"roi": null
},
{
"vig": -105,
"line_shift": -1,
"units": 0,
"roi": null
},
{
"vig": -110,
"line_shift": 0,
"units": 0,
"roi": null
},
{
"vig": -110,
"line_shift": -0.5,
"units": 0,
"roi": null
},
{
"vig": -110,
"line_shift": -1,
"units": 0,
"roi": null
},
{
"vig": -115,
"line_shift": 0,
"units": 0,
"roi": null
},
{
"vig": -115,
"line_shift": -0.5,
"units": 0,
"roi": null
},
{
"vig": -115,
"line_shift": -1,
"units": 0,
"roi": null
},
{
"vig": -120,
"line_shift": 0,
"units": 0,
"roi": null
},
{
"vig": -120,
"line_shift": -0.5,
"units": 0,
"roi": null
},
{
"vig": -120,
"line_shift": -1,
"units": 0,
"roi": null
}
]
},
"leak_test": {
"status": "clean",
"rows_checked": 6,
"matched_checked": 1,
"matches_flipped": 0,
"parts": []
},
"warm_up": {
"policy": "as_written",
"games_needed": 0,
"windows": [],
"short_bets": 0,
"by_policy": null,
"warning": false
},
"settlement": {
"profile": "kelly_standard",
"rules_version": "2026-11-08",
"applied": [],
"rules": [
{
"id": "final_score",
"text": "Every market grades on the final score as Kelly's score source records it, overtime and extra innings included.",
"effective_from": "2026-11-06"
},
{
"id": "spread_push",
"text": "A spread bet whose margin lands exactly on the line pushes: the stake comes back.",
"effective_from": "2026-11-06"
},
{
"id": "postponed_void",
"text": "A game with no final score 48 hours after its scheduled start is void: the bet is left out.",
"effective_from": "2026-11-07"
},
{
"id": "not_played_yet",
"text": "A game still to be played is listed with no result and left out of every figure.",
"effective_from": "2026-11-06"
}
]
},
"moment": {
"bet_moment": "close",
"target": "the start",
"max_stale_minutes": 60,
"floor_minutes_before_oct_2022": 10,
"rows_priced": 0,
"rows_aged": 0,
"median_snapshot_age_minutes": null,
"max_snapshot_age_minutes": null,
"stale_price": 0,
"after_moment": 0,
"age_unknown": 0,
"line_sources": {}
},
"coverage": {
"flow": [
{
"step": "in_scope",
"games": 3,
"rows": 6
},
{
"step": "rule_true",
"rows": 1
},
{
"step": "one_bet_a_game",
"rows": 1,
"out": {
"both_sides": 0,
"same_game_total": 0
}
},
{
"step": "priced",
"rows": 0,
"out": {
"no_line": 1
}
},
{
"step": "graded",
"rows": 0,
"out": {
"void": 0,
"not_played_yet": 0
}
},
{
"step": "decisions",
"rows": 0,
"out": {
"push": 0
}
}
],
"by_season": [
{
"season": "2025-26",
"games": 3,
"rows": 6,
"rule_true": 1,
"bets": 0,
"priced_share": 0.3333
}
],
"inputs_missing": [
{
"attribute": "home",
"rows": 6,
"rows_blank": 0,
"share": 0,
"seasons_blank": []
},
{
"attribute": "rest_days",
"rows": 6,
"rows_blank": 2,
"share": 0.3333,
"seasons_blank": []
}
],
"left_out_differs": {
"left_out": 1,
"kept": 0,
"flag": false,
"max_smd": null,
"covariate": null,
"by_covariate": []
}
},
"random_baseline": null,
"decay": {
"bets": 0,
"window": null,
"rolling": [],
"break": null,
"cusum": null,
"known_break": null,
"trend_per_100_bets": null,
"decay_ratio": null,
"clv_break": null,
"roi_newest_30": null,
"status_reason": "no graded bets"
},
"ablation": {
"base": {
"decisions": 2,
"roi": -0.04500000000000004
},
"value": "excess_units",
"total": 0,
"rule": {
"decisions": 0,
"units": 0
},
"clauses": [
{
"condition": "home",
"shapley_units": 0.4775,
"share": null,
"alone": {
"decisions": 1,
"units": 0.9099999999999999,
"roi": 0.9099999999999999
},
"without": {
"decisions": 0,
"units": 0,
"roi": null,
"changed_share": 0,
"formula": "=rest_days >= 2"
},
"label": "dead",
"implied_by": [],
"fix": {
"label": "Remove it",
"formula": "=rest_days >= 2"
},
"sampled_se": null
},
{
"condition": "rest_days >= 2",
"shapley_units": -0.4775,
"share": null,
"alone": {
"decisions": 0,
"units": 0,
"roi": null
},
"without": {
"decisions": 1,
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"changed_share": 1,
"formula": "=home"
},
"label": "dead",
"implied_by": [],
"fix": {
"label": "Remove it",
"formula": "=home"
},
"sampled_se": null
}
],
"pairs": [
{
"conditions": [
"home",
"rest_days >= 2"
],
"interaction_units": -0.955
}
],
"method": "exact",
"evaluations": 3
},
"nearby": {
"numbers": [],
"variants": 0,
"distinct_bet_sets": 0,
"grid": null,
"roi_median": null,
"roi_p05": null,
"roi_p95": null,
"plateau_share": null,
"spike_ratio": null,
"reading": null,
"na": "the rule compares no numbers"
},
"monte_carlo": null,
"noise": {
"seed": 3125498839,
"inputs": {
"runs": 200,
"profitable_runs": 0,
"profitable_share": 0,
"roi_median": null,
"roi_p05": null,
"input_scale": 0.1,
"overlap_median": 1,
"numbers": 1
},
"inputs_na": null,
"lines": null,
"delay": {
"bet_moment": "day_before",
"na": "day-before lines for this league are still loading"
},
"low_power": true
},
"concentration": {
"bets": 0,
"units": 0,
"by": [],
"trim_top_0_05": null
},
"split_dimensions": [],
"robustness": {
"fast_path": true,
"seed": 2248428446,
"evaluations": 1,
"off": [],
"checks": [
{
"key": "half_point_worse",
"label": "Holds ½ point worse",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "nearby_holds",
"label": "Holds with nearby numbers",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "not_concentrated",
"label": "Not one team or season",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
},
{
"key": "still_working",
"label": "Still working",
"status": "na",
"value": "fewer than (30) decided bets",
"statistic": null,
"threshold": null
}
]
},
"verdict": {
"version": 2,
"headline": "No bets",
"level": "too_few",
"passed": 0,
"applicable": 0,
"gate": {
"key": "no_look_ahead",
"status": "pass",
"value": "every value the rule reads is known at the bet moment",
"leak_test": "clean"
},
"checks": []
},
"risk": {
"staking": "flat",
"stake_fraction": null,
"bankroll_units": 100,
"ruin_share": 0.5,
"horizon_bets": 500,
"paths": 5000,
"seed": 4440280713539600,
"lose_all_ever": null,
"lose_all_ever_allowing_error": null,
"ruin_ever_if_rate_exact": null,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"estimation_premium": null,
"drawdown_p95_units": null,
"drawdown_p95": null,
"down_after_horizon": null,
"bankroll_for_ruin_005_units": null,
"curve": [],
"status": "too_few",
"method": "lundberg+posterior_monte_carlo",
"warnings": [
"too_few_decisions"
],
"percent": {
"stake_fraction": 0.02,
"ruin_if_rate_exact": null,
"ruin_allowing_error": null,
"drawdown_p95": null,
"status": "too_few"
}
},
"staking_plans": {
"bankroll_units": 100,
"percent": 0.02,
"kelly_cap": 0.05,
"kelly": {
"win_rate": null,
"push_rate": 0,
"full_fraction": null,
"shrink": null,
"multiplier": null,
"honest_fraction": null,
"capped": null,
"simultaneous": false,
"basis": "record",
"multiplier_rule": "honest",
"withheld": "stake_first_party_only"
},
"plans": [
{
"plan": "flat",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "to_win",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "percent",
"status": "ok",
"stopped_at_bet": null,
"bets": 0,
"end_bankroll_units": 100,
"units": 0,
"max_drawdown": 0,
"growth_per_bet": null,
"roi": null
},
{
"plan": "kelly",
"status": "withheld",
"withheld": "stake_first_party_only"
}
],
"warnings": [
"small_sample"
]
},
"risk_adjusted": {
"bets": 0,
"warnings": [
"no_bets"
]
},
"caps": null,
"capacity": {
"available": false,
"market": "spreads",
"before_start": "0_1h",
"books": [],
"max_stake_per_bet": null,
"per_season_at_limit": null,
"bets_per_season": null,
"limits_held": 0,
"reason": "Kelly holds no measured spreads limits for nba at this moment yet."
},
"splits": [],
"without": [
{
"condition": "home",
"formula": "=rest_days >= 2",
"bets": 0,
"decisions": 0,
"record": "0-0",
"units": 0,
"roi": null,
"p_value": null,
"p_value_vs_rule": null,
"p_adjusted": null,
"shown_rows": 1,
"p_value_vs_rule_adjusted": null
},
{
"condition": "rest_days >= 2",
"formula": "=home",
"bets": 1,
"decisions": 1,
"record": "1-0",
"units": 0.9099999999999999,
"roi": 0.9099999999999999,
"p_value": 0.5235602094240838,
"p_value_vs_rule": null,
"p_adjusted": 0.5235602094240838,
"shown_rows": 1,
"p_value_vs_rule_adjusted": null
}
],
"rule": {
"formula": "=AND(home, rest_days >= 2)",
"competition": "nba",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2025-26"
],
"season_type": "regular",
"stake": "risk",
"price": "consensus",
"base_formula": null,
"reads": [
{
"name": "home",
"known": "pregame",
"windowed": false
},
{
"name": "rest_days",
"known": "pregame",
"windowed": false
}
],
"rule_hash": "sha256:1e9e3c96193481744619102207b3b7c22de187b64f1fdd3c062d04c31cc8d451",
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"translation_id": "tx_xbj639zqrcq2kg653rsf"
},
"left_out": {
"no_line": 1,
"both_sides": 0,
"stale_price": 0
},
"matched_without_line": 1,
"matched_both_sides": 0,
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (1).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"as_of": "2026-11-06T00:00:00Z",
"notice": "This tests your rule on past games. It isn't a pick and doesn't rank selections.",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"stale": false,
"since_saved": null,
"tests_on_account": null,
"data_version": "snap_52",
"manifest": {
"version": 1,
"code": {
"engine": 8,
"release": null,
"formula_version": "1.0.0",
"catalogue_version": "1:0",
"attributes_sha256": "0000000000000000",
"settlement_rules": "2026-11-08"
},
"data": {
"version": "snap_52",
"snapshots": [
"snap_52"
],
"one_snapshot": true,
"as_of": "2026-11-06T00:00:00Z",
"data_through": {},
"latest_result": "2025-10-25T23:00:00Z",
"rows_built_at": "2026-11-06T00:00:00Z",
"queries": {
"rule_test_games": "0000000000000000"
}
},
"request": {
"request_hash": "sha256:e65dcd26df2fdf9aaf92df60ff1159b2a2f11b28898d1af882341bd9a877ad58",
"rule_hash": "sha256:1e9e3c96193481744619102207b3b7c22de187b64f1fdd3c062d04c31cc8d451",
"seed": 16551235696159332000
},
"sources": {
"prices": {},
"inputs": [
{
"name": "home",
"known": "pregame"
},
{
"name": "rest_days",
"known": "pregame"
}
]
},
"seasons": [
{
"season": "2025-26",
"rows": 6,
"priced_share": 0.3333
}
],
"left_out": {
"both_sides": 0,
"no_line": 1,
"stale_price": 0
},
"sha256": "sha256:a0e4e4125043a139c2c30ac28dcf4a0105b68d55996eeaacf85d176c80f81a16"
},
"run_id": null,
"rule_version": null,
"report_url": null
},
"summary": "Bets the team against the spread when it is at home, and its rest in days is at least 2. Kelly's record: 0-0. Since it was published: 0-0. Verdict: No bets.",
"assumptions": [
"Each bet is taken at Kelly's consensus close price; games Kelly holds no close line for are left out (1).",
"Games where the rule matched both teams are left out (0).",
"One flat unit risked a bet."
],
"has_more": false,
"next_cursor": null,
"as_of": "2026-11-06T00:00:00Z",
"source": "kelly",
"attribution": null,
"request_id": "req_3dd95de735d1f1e0b4319c7f",
"live_topics": [],
"rg_line": null
}POST /v1/rules:import
scope write:formulas · 40 credits, not charged during the beta · MCP tool import_rule · status live
Reads a betting system written for another tool (SDQL, a filter list from Bet Labs, BetIQ, TrendsCenter, BettingPros or Covrr, a formula in Excel, SQL, pandas and other modes) or in plain words, and maps each of its conditions to Kelly's formula language: exact, close (with how it differs), ask (a question with choices) or cant_map (with why). The draft carries the combined formula, the look-ahead check and the exact POST /v1/rules:test body. dry_run (the default) keeps nothing; dry_run false keeps the draft for 30 days. Kelly never fetches a link or logs in anywhere (I2). Needs your own live API key with write:formulas; an AI app imports through Kelly MCP, which asks you first.
Body, application/json (required):
textstringThe system: SDQL, a filter list, a formula, or plain English.
max length 20000
rowsarray of objectFilter rows a reader split, instead of text.
max items 60
toolstringrequiredThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
labelstringrequiredThe choice in words.
max length 200
opanyrequiredThe comparison.
One of: =, !=, <, <=, >, >=, between, in, is (open: new values may be added).
valueanyrequiredAs shown: text, a number, a list for in; the min for between.
value2anyThe max for between (null when open-ended).
source_textstringrequiredThe condition in the source's own words.
max length 2000
confidenceanyThe vision reader's confidence; low maps to status ask.
One of: high, low (open: new values may be added).
pageintegernullableThe image or page the row was read from.
min 0
formatanyWhat the text is; auto (default) tries every reader.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
toolstringWhere it came from (TOOLS, or kelly_template:<id>).
pattern ^[a-z0-9_.:-]{1,64}$
formula_modeanyFor format formula: the mode it is written in.
One of: excel, sheets, sql, polars, pandas, dplyr, dax, tableau, pine (open: new values may be added).
source_kindanySet by the entry point; default paste (api on the API, mcp on MCP).
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
raw_refstringWhere it came from (a URL, file or message id). Kept, never fetched (I2).
max length 500
viaanyThe entry point's channel.
One of: site, api, mcp, chat, extension, bookmarklet, share_target, email, zapier, make, link (open: new values may be added).
filesarray of objectFiles for the file and vision readers (T9c), uploaded first.
max items 20
file_idstringrequiredThe file's ID.
max length 100
namestringThe name.
max length 255
content_typestringThe file's media type.
max length 100
sizeintegerThe file's size in bytes.
min 0
namestringThe name.
max length 120
competitionstringThe league.
max length 32
marketanyspread, moneyline, over or under.
One of: spread, moneyline, over, under, teaser, draw, win, place, each_way, lay, lay_place (open: new values may be added).
sideanyrule or fade.
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of string1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
from_seasonstringFirst season.
to_seasonstringLast season.
season_typeanyregular, postseason or all.
One of: regular, postseason, all (open: new values may be added).
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
versions_triedintegerHow many versions the user tried in the other tool; default 20 (I1).
min 1 · max 1000 · default 20
dry_runbooleanDefault true: return the draft without keeping it (id null). False keeps it for 30 days under an imp_ id.
default true
200 A dry run: the draft, kept nowhere (id null).
objectconstrequiredAlways rule_import.
Always "rule_import".
idstringrequirednullableThe import's ID (imp_…); null on a dry run.
pattern ^imp_[0-9a-hjkmnp-tv-z]{20}$
stateanyrequireddraft, confirmed, undone or expired.
One of: draft, confirmed, undone, expired (open: new values may be added).
sourceobjectrequiredWhere the system came from and its text.
toolstringnullableThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
tool_labelstringnullableThe tool's name in words.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
formatanyrequiredWhat the text is: sdql, filter_list, formula, plain or auto.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
viaone of several shapesThe entry point that passed it on.
formula_modeone of several shapesThe formula's mode (excel, sql, pandas...).
raw_refstringnullableWhere it came from (a URL, file or message), kept and never fetched.
captured_atstringnullableWhen Kelly received it.
format date-time
textstringnullableThe text as received (kept with the draft and the saved rule, for the proof screen; the owner sees it).
text_sha256stringnullableThe text's SHA-256, hex.
pattern ^[0-9a-f]{64}$
systemobjectrequiredThe bet the system makes, as the rule tester names it. A reader fills what the source says; the caller's hints win.
namestringnullableThe name.
max length 120
competitionstringnullablenfl, ncaaf, nba, wnba, ncaab, nhl, mlb ...
max length 32
marketone of several shapesspread, moneyline, over or under.
sideanyrule (bet the matched team or over/under) or fade (the other side).
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of stringnullable1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
season_typeone of several shapesregular, postseason or all.
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
conditionsarray of objectrequiredOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
source_textstringrequiredThe condition as the source wrote it.
max length 2000
statusanyrequiredexact, close, ask or cant_map.
One of: exact, close, ask, cant_map (open: new values may be added).
formulastringnullableKelly formula (Excel mode, starts with =); null for ask and cant_map.
max length 2000
reasonstringnullableWhy close or can't map; how it differs from the source.
questionstringnullableFor ask: the question for the user.
choicesarray of objectThe readings to choose from, each with its formula.
labelstringrequiredThe choice in words.
formulastringrequiredThe Kelly formula (TRUE or FALSE on each team-game).
data_gapobjectnullableWhat Kelly lacks for a condition it can't import.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: no_attribute, no_history, not_loaded, out_of_grain, licence, edge, look_ahead (open: new values may be added).
detailstringrequiredThe detail, in words.
attributesarray of stringKelly attribute names the formula reads.
methodanydeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
resolutionone of several shapesremoved or live_only, for a condition left out.
formulastringrequirednullableAND of the exact and close conditions.
matchanyrequiredexact, close_match, needs_answers or cant_import.
One of: exact, close_match, needs_answers, cant_import (open: new values may be added).
countsobjectrequiredConditions by status.
exactintegerrequiredConditions mapped exactly.
min 0
closeintegerrequiredConditions mapped closely.
min 0
askintegerrequiredConditions waiting on an answer.
min 0
cant_mapintegerrequiredConditions Kelly can't import.
min 0
checkobjectrequiredkformula.check and the look-ahead check on the combined formula, at system.bet_moment. The leak test runs with the rule tester at confirm.
testablebooleanrequiredWhether the rule tester can test the formula.
problemsarray of objectrequiredEach reason it can't be tested yet.
codestringrequiredThe reason's code.
pathstringnullableThe attribute it concerns.
messagestringrequiredThe reason in words.
fixesarray of objectWays to fix it.
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
ledgerobjectrequiredHow the import counts on the rules-tried ledger at confirm (I1).
labelconstrequiredThe choice in words.
Always "tuned elsewhere".
triesintegerrequiredRules counted as tried.
min 1 · max 1000
tries_fromanyrequireddefault (20) or user.
One of: default, user (open: new values may be added).
labelsarray of object"tuned elsewhere", and "close match" when match is close_match.
readerobjectrequiredThe reader that read it.
namestringrequiredThe name.
methodanyrequireddeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
attemptsintegerThe AI reader's calls (1 to 3).
min 1
warningsarray of objectrequiredNotes on how Kelly read it.
codestringrequiredThe reason's code.
messagestringrequiredThe reason in words.
test_requestobjectnullableThe POST /v1/rules:test body this draft would run; for competition exchange, the POST /v1/rules:test_exchange body (test_exchange_strategy). Null when match is needs_answers or cant_import.
ruleobjectnullableAfter confirm: the saved rule (T0).
rule_idstringnullableThe saved rule's ID (rl_…).
namestringThe name.
versionintegernullableThe saved version.
new_versionbooleannullableWhether confirm made a new version.
versions_triedintegernullableVersions tried, as the rule tester counts.
run_idstringnullableThe rule test kept for it (run_…).
testobjectnullableConfirm only: the rule tester's report on the saved rule (test_rule's facts without the bets, and its summary).
othersarray of stringWhen the source held several systems: the other drafts' ids.
created_atstringnullableWhen the draft was kept.
format date-time
expires_atstringnullableWhen an unconfirmed draft expires (30 days).
format date-time
confirmed_atstringnullableWhen it was confirmed.
format date-time
undo_untilstringnullableUndo is open until then (7 days).
format date-time
contract_versionconstrequiredThe import contract's version.
Always "1.0.0".
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · import_unreadable
POST https://api.askkelly.ai/v1/rules:import
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "H and line<=-3 and p:L",
"tool": "sdql",
"competition": "nfl",
"market": "spread",
"from_season": "2023",
"to_season": "2024"
}
{
"object": "rule_import",
"id": null,
"state": "draft",
"source": {
"tool": "sdql",
"tool_label": "SDQL (KillerSports / SportsDataBase)",
"kind": "api",
"format": "sdql",
"via": null,
"formula_mode": null,
"raw_ref": null,
"captured_at": "2026-10-04T17:30:00Z",
"text": "H and line<=-3 and p:L",
"text_sha256": "ac5e42601cf5209bb381275989ee4652f463b74fa25190e705dfa422f87f9f96"
},
"system": {
"name": null,
"competition": "nfl",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2023",
"2024"
],
"season_type": "regular",
"stake": "risk"
},
"conditions": [
{
"index": 0,
"source_text": "H",
"status": "exact",
"formula": "=AND(home, NOT(neutral))",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"home",
"neutral"
],
"method": "deterministic",
"resolution": null
},
{
"index": 1,
"source_text": "line<=-3",
"status": "exact",
"formula": "=spread_close <= -3",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"spread_close"
],
"method": "deterministic",
"resolution": null
},
{
"index": 2,
"source_text": "p:L",
"status": "exact",
"formula": "=LAG(result = \"L\")",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"result"
],
"method": "deterministic",
"resolution": null
}
],
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"match": "exact",
"counts": {
"exact": 3,
"close": 0,
"ask": 0,
"cant_map": 0
},
"check": {
"testable": true,
"problems": []
},
"source_record": null,
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"labels": [
"tuned elsewhere"
],
"reader": {
"name": "sdql",
"method": "deterministic",
"attempts": 1
},
"warnings": [
{
"code": "reader",
"message": "Kelly reads earlier games (p:, P:) within the season tested, so a team's first game of a season has no previous game."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
}
],
"test_request": {
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"competition": "nfl",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2023",
"2024"
],
"season_type": "regular",
"stake": "risk"
},
"rule": null,
"created_at": "2026-10-04T17:30:00Z",
"expires_at": null,
"confirmed_at": null,
"undo_until": null,
"contract_version": "1.0.0"
}POST https://api.askkelly.ai/v1/rules:import
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "Home/Away: Home\nFavorite/Underdog: Favorite\nRest Advantage: 2+ days\nPublic Betting %: under 40%",
"competition": "nfl",
"dry_run": false
}
{
"object": "rule_import",
"id": "imp_7h7p7w2tvm42z7xg6fpy",
"state": "draft",
"source": {
"tool": "bet_labs",
"tool_label": "Bet Labs (Action Network)",
"kind": "api",
"format": "filter_list",
"via": null,
"formula_mode": null,
"raw_ref": null,
"captured_at": "2026-10-04T17:30:00Z",
"text": "Home/Away: Home\nFavorite/Underdog: Favorite\nRest Advantage: 2+ days\nPublic Betting %: under 40%",
"text_sha256": "076d17982bcb14c5005dc77136b1d087fad4b9e64ee54cd3917480fd10d22cbd"
},
"system": {
"name": null,
"competition": "nfl",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2020",
"2021",
"2022",
"2023",
"2024",
"2025",
"2026"
],
"season_type": "regular",
"stake": "risk"
},
"conditions": [
{
"index": 0,
"source_text": "Home/Away: Home",
"status": "exact",
"formula": "=AND(home, NOT(neutral))",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"home",
"neutral"
],
"method": "deterministic",
"resolution": null
},
{
"index": 1,
"source_text": "Favorite/Underdog: Favorite",
"status": "exact",
"formula": "=spread_close < 0",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"spread_close"
],
"method": "deterministic",
"resolution": null
},
{
"index": 2,
"source_text": "Rest Advantage: 2+ days",
"status": "ask",
"formula": null,
"reason": null,
"question": "Rest Advantage: what range did you set? Kelly read '2+ days'.",
"choices": [],
"data_gap": null,
"attributes": [],
"method": "deterministic",
"resolution": null
},
{
"index": 3,
"source_text": "Public Betting %: under 40%",
"status": "cant_map",
"formula": null,
"reason": "Bets %: Kelly holds betting splits only from Nov 2026 (S8).",
"question": null,
"choices": [],
"data_gap": {
"kind": "no_history",
"detail": "Kelly holds betting splits only from Nov 2026 (S8)"
},
"attributes": [],
"method": "deterministic",
"resolution": null
}
],
"formula": "=AND(AND(home, NOT(neutral)), spread_close < 0)",
"match": "needs_answers",
"counts": {
"exact": 2,
"close": 0,
"ask": 1,
"cant_map": 1
},
"check": {
"testable": true,
"problems": []
},
"source_record": null,
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"labels": [
"tuned elsewhere"
],
"reader": {
"name": "filters:bet_labs",
"method": "deterministic",
"attempts": 1
},
"warnings": [
{
"code": "reader",
"message": "Kelly's names for bet_labs's filters come from its help pages, not a real export; check each condition against the source."
},
{
"code": "needs_market",
"message": "Say which bet the system makes: spread, moneyline, over or under. Kelly assumed the spread."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
},
{
"code": "seasons",
"message": "The source names no seasons; Kelly tests 2020 to 2026, every season it holds lines for."
}
],
"test_request": null,
"rule": null,
"created_at": "2026-10-04T17:30:00Z",
"expires_at": "2026-11-03T17:30:00Z",
"confirmed_at": null,
"undo_until": null,
"contract_version": "1.0.0"
}POST https://api.askkelly.ai/v1/rules:import
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "H and line<=-3 and p:L",
"tool": "sdql",
"competition": "nfl",
"market": "parlay",
"from_season": "2023",
"to_season": "2024"
}
{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "market: 'parlay' is not one of ['spread', 'moneyline', 'over', 'under', 'teaser', 'draw', 'win', 'place', 'each_way', 'lay', 'lay_place']",
"param": "market",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_607628b862cc4d353365ed7b",
"retryable": false,
"display_message": "Something in that request was not understood."
}POST https://api.askkelly.ai/v1/rules:import
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"text": "H and line<=-3 and p:L",
"tool": "sdql",
"competition": "nfl",
"market": "spread",
"from_season": "2023",
"to_season": "2024",
"dry_run": false
}
{
"object": "rule_import",
"id": "imp_rm4wkv9vhwevdmh05d1z",
"state": "draft",
"source": {
"tool": "sdql",
"tool_label": "SDQL (KillerSports / SportsDataBase)",
"kind": "api",
"format": "sdql",
"via": null,
"formula_mode": null,
"raw_ref": null,
"captured_at": "2026-10-04T17:30:00Z",
"text": "H and line<=-3 and p:L",
"text_sha256": "ac5e42601cf5209bb381275989ee4652f463b74fa25190e705dfa422f87f9f96"
},
"system": {
"name": null,
"competition": "nfl",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2023",
"2024"
],
"season_type": "regular",
"stake": "risk"
},
"conditions": [
{
"index": 0,
"source_text": "H",
"status": "exact",
"formula": "=AND(home, NOT(neutral))",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"home",
"neutral"
],
"method": "deterministic",
"resolution": null
},
{
"index": 1,
"source_text": "line<=-3",
"status": "exact",
"formula": "=spread_close <= -3",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"spread_close"
],
"method": "deterministic",
"resolution": null
},
{
"index": 2,
"source_text": "p:L",
"status": "exact",
"formula": "=LAG(result = \"L\")",
"reason": null,
"question": null,
"choices": [],
"data_gap": null,
"attributes": [
"result"
],
"method": "deterministic",
"resolution": null
}
],
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"match": "exact",
"counts": {
"exact": 3,
"close": 0,
"ask": 0,
"cant_map": 0
},
"check": {
"testable": true,
"problems": []
},
"source_record": null,
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"labels": [
"tuned elsewhere"
],
"reader": {
"name": "sdql",
"method": "deterministic",
"attempts": 1
},
"warnings": [
{
"code": "reader",
"message": "Kelly reads earlier games (p:, P:) within the season tested, so a team's first game of a season has no previous game."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
}
],
"test_request": {
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"competition": "nfl",
"market": "spread",
"side": "rule",
"bet_moment": "close",
"seasons": [
"2023",
"2024"
],
"season_type": "regular",
"stake": "risk"
},
"rule": null,
"created_at": "2026-10-04T17:30:00Z",
"expires_at": "2026-11-03T17:30:00Z",
"confirmed_at": null,
"undo_until": null,
"contract_version": "1.0.0"
}GET /v1/rules/imports/{import_id}
scope read:formulas · 0 credits, not charged during the beta · MCP tool get_rule_import · status live
One of your kept imports: the draft with each condition's mapping, or the saved rule it became. Anyone else's answers 404.
import_idstring · pathrequiredThe import's ID (imp_…).
pattern ^imp_[0123456789abcdefghjkmnpqrstvwxyz]{20}$
200 The import: a draft, or the rule it was confirmed into.
objectconstrequiredAlways rule_import.
Always "rule_import".
idstringrequirednullableThe import's ID (imp_…); null on a dry run.
pattern ^imp_[0-9a-hjkmnp-tv-z]{20}$
stateanyrequireddraft, confirmed, undone or expired.
One of: draft, confirmed, undone, expired (open: new values may be added).
sourceobjectrequiredWhere the system came from and its text.
toolstringnullableThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
tool_labelstringnullableThe tool's name in words.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
formatanyrequiredWhat the text is: sdql, filter_list, formula, plain or auto.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
viaone of several shapesThe entry point that passed it on.
formula_modeone of several shapesThe formula's mode (excel, sql, pandas...).
raw_refstringnullableWhere it came from (a URL, file or message), kept and never fetched.
captured_atstringnullableWhen Kelly received it.
format date-time
textstringnullableThe text as received (kept with the draft and the saved rule, for the proof screen; the owner sees it).
text_sha256stringnullableThe text's SHA-256, hex.
pattern ^[0-9a-f]{64}$
systemobjectrequiredThe bet the system makes, as the rule tester names it. A reader fills what the source says; the caller's hints win.
namestringnullableThe name.
max length 120
competitionstringnullablenfl, ncaaf, nba, wnba, ncaab, nhl, mlb ...
max length 32
marketone of several shapesspread, moneyline, over or under.
sideanyrule (bet the matched team or over/under) or fade (the other side).
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of stringnullable1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
season_typeone of several shapesregular, postseason or all.
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
conditionsarray of objectrequiredOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
source_textstringrequiredThe condition as the source wrote it.
max length 2000
statusanyrequiredexact, close, ask or cant_map.
One of: exact, close, ask, cant_map (open: new values may be added).
formulastringnullableKelly formula (Excel mode, starts with =); null for ask and cant_map.
max length 2000
reasonstringnullableWhy close or can't map; how it differs from the source.
questionstringnullableFor ask: the question for the user.
choicesarray of objectThe readings to choose from, each with its formula.
labelstringrequiredThe choice in words.
formulastringrequiredThe Kelly formula (TRUE or FALSE on each team-game).
data_gapobjectnullableWhat Kelly lacks for a condition it can't import.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: no_attribute, no_history, not_loaded, out_of_grain, licence, edge, look_ahead (open: new values may be added).
detailstringrequiredThe detail, in words.
attributesarray of stringKelly attribute names the formula reads.
methodanydeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
resolutionone of several shapesremoved or live_only, for a condition left out.
formulastringrequirednullableAND of the exact and close conditions.
matchanyrequiredexact, close_match, needs_answers or cant_import.
One of: exact, close_match, needs_answers, cant_import (open: new values may be added).
countsobjectrequiredConditions by status.
exactintegerrequiredConditions mapped exactly.
min 0
closeintegerrequiredConditions mapped closely.
min 0
askintegerrequiredConditions waiting on an answer.
min 0
cant_mapintegerrequiredConditions Kelly can't import.
min 0
checkobjectrequiredkformula.check and the look-ahead check on the combined formula, at system.bet_moment. The leak test runs with the rule tester at confirm.
testablebooleanrequiredWhether the rule tester can test the formula.
problemsarray of objectrequiredEach reason it can't be tested yet.
codestringrequiredThe reason's code.
pathstringnullableThe attribute it concerns.
messagestringrequiredThe reason in words.
fixesarray of objectWays to fix it.
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
ledgerobjectrequiredHow the import counts on the rules-tried ledger at confirm (I1).
labelconstrequiredThe choice in words.
Always "tuned elsewhere".
triesintegerrequiredRules counted as tried.
min 1 · max 1000
tries_fromanyrequireddefault (20) or user.
One of: default, user (open: new values may be added).
labelsarray of object"tuned elsewhere", and "close match" when match is close_match.
readerobjectrequiredThe reader that read it.
namestringrequiredThe name.
methodanyrequireddeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
attemptsintegerThe AI reader's calls (1 to 3).
min 1
warningsarray of objectrequiredNotes on how Kelly read it.
codestringrequiredThe reason's code.
messagestringrequiredThe reason in words.
test_requestobjectnullableThe POST /v1/rules:test body this draft would run; for competition exchange, the POST /v1/rules:test_exchange body (test_exchange_strategy). Null when match is needs_answers or cant_import.
ruleobjectnullableAfter confirm: the saved rule (T0).
rule_idstringnullableThe saved rule's ID (rl_…).
namestringThe name.
versionintegernullableThe saved version.
new_versionbooleannullableWhether confirm made a new version.
versions_triedintegernullableVersions tried, as the rule tester counts.
run_idstringnullableThe rule test kept for it (run_…).
testobjectnullableConfirm only: the rule tester's report on the saved rule (test_rule's facts without the bets, and its summary).
othersarray of stringWhen the source held several systems: the other drafts' ids.
created_atstringnullableWhen the draft was kept.
format date-time
expires_atstringnullableWhen an unconfirmed draft expires (30 days).
format date-time
confirmed_atstringnullableWhen it was confirmed.
format date-time
undo_untilstringnullableUndo is open until then (7 days).
format date-time
contract_versionconstrequiredThe import contract's version.
Always "1.0.0".
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/imports/imp_rm4wkv9vhwevdmh05d1z
Authorization: Bearer $KELLY_API_KEY{
"id": "imp_rm4wkv9vhwevdmh05d1z",
"rule": null,
"check": {
"problems": [],
"testable": true
},
"match": "exact",
"state": "draft",
"counts": {
"ask": 0,
"close": 0,
"exact": 3,
"cant_map": 0
},
"labels": [
"tuned elsewhere"
],
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"object": "rule_import",
"reader": {
"name": "sdql",
"method": "deterministic",
"attempts": 1
},
"source": {
"via": null,
"kind": "api",
"text": "H and line<=-3 and p:L",
"tool": "sdql",
"format": "sdql",
"raw_ref": null,
"tool_label": "SDQL (KillerSports / SportsDataBase)",
"captured_at": "2026-10-04T17:30:00Z",
"text_sha256": "ac5e42601cf5209bb381275989ee4652f463b74fa25190e705dfa422f87f9f96",
"formula_mode": null
},
"system": {
"name": null,
"side": "rule",
"stake": "risk",
"market": "spread",
"seasons": [
"2023",
"2024"
],
"bet_moment": "close",
"competition": "nfl",
"season_type": "regular"
},
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"warnings": [
{
"code": "reader",
"message": "Kelly reads earlier games (p:, P:) within the season tested, so a team's first game of a season has no previous game."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
}
],
"conditions": [
{
"index": 0,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=AND(home, NOT(neutral))",
"data_gap": null,
"question": null,
"attributes": [
"home",
"neutral"
],
"resolution": null,
"source_text": "H"
},
{
"index": 1,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=spread_close <= -3",
"data_gap": null,
"question": null,
"attributes": [
"spread_close"
],
"resolution": null,
"source_text": "line<=-3"
},
{
"index": 2,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=LAG(result = \"L\")",
"data_gap": null,
"question": null,
"attributes": [
"result"
],
"resolution": null,
"source_text": "p:L"
}
],
"created_at": "2026-10-04T17:30:00Z",
"expires_at": "2026-11-03T17:30:00Z",
"undo_until": null,
"confirmed_at": null,
"test_request": {
"side": "rule",
"stake": "risk",
"market": "spread",
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"seasons": [
"2023",
"2024"
],
"bet_moment": "close",
"competition": "nfl",
"season_type": "regular"
},
"source_record": null,
"contract_version": "1.0.0"
}GET https://api.askkelly.ai/v1/rules/imports/imp_zzzzzzzzzzzzzzzzzzzz
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/not_found",
"title": "Not found",
"status": 404,
"code": "not_found",
"detail": "No import with that ID on this account.",
"param": "import_id",
"doc_url": "https://docs.askkelly.ai/problems/not_found",
"request_id": "req_303b0fcd43914312dba33164",
"retryable": false,
"display_message": "That could not be found."
}POST /v1/rules:import/{import_id}:confirm
scope write:formulas · 1 credit, not charged during the beta · MCP tool confirm_rule_import · status live
Applies your answers and edits (conditions), tests the rule with the rule tester and saves it as a saved rule labelled tuned elsewhere, counted as the versions you tried elsewhere (versions_tried, default 20) on your rules-tried ledger. A close match needs accept_close: true and is labelled close match; a condition Kelly can't import is left out, never dropped silently. Needs your own live API key with write:formulas; an AI app imports through Kelly MCP, which asks you first.
import_idstring · pathrequiredThe import's ID (imp_…).
pattern ^imp_[0123456789abcdefghjkmnpqrstvwxyz]{20}$
Body, application/json (required):
namestringThe saved rule's name (unique per account); default system.name.
max length 120
conditionsarray of objectOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
formulastringThe formula to use (an answer's choice or the user's own).
statusanyDefault: exact for a choice the mapper offered, close for the user's own formula.
One of: exact, close, cant_map (open: new values may be added).
resolutionanyFor a cant_map condition: removed or live_only.
One of: removed, live_only (open: new values may be added).
versions_triedintegerHow many versions the user tried in the other tool (I1).
min 1 · max 1000 · default 20
accept_closebooleanRequired true when match is close_match (I4). Every cant_map condition needs a resolution; with accept_close, one left unresolved is removed.
default false
competitionstringThe league.
marketanyspread, moneyline, over or under.
One of: spread, moneyline, over, under, teaser, draw, win, place, each_way, lay, lay_place (open: new values may be added).
sideanyrule or fade.
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of string1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
detailanyThe rule tester report's detail (default summary).
One of: summary, full (open: new values may be added).
200 The import, confirmed: the saved rule under rule and the rule tester's report under test.
objectconstrequiredAlways rule_import.
Always "rule_import".
idstringrequirednullableThe import's ID (imp_…); null on a dry run.
pattern ^imp_[0-9a-hjkmnp-tv-z]{20}$
stateanyrequireddraft, confirmed, undone or expired.
One of: draft, confirmed, undone, expired (open: new values may be added).
sourceobjectrequiredWhere the system came from and its text.
toolstringnullableThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
tool_labelstringnullableThe tool's name in words.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
formatanyrequiredWhat the text is: sdql, filter_list, formula, plain or auto.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
viaone of several shapesThe entry point that passed it on.
formula_modeone of several shapesThe formula's mode (excel, sql, pandas...).
raw_refstringnullableWhere it came from (a URL, file or message), kept and never fetched.
captured_atstringnullableWhen Kelly received it.
format date-time
textstringnullableThe text as received (kept with the draft and the saved rule, for the proof screen; the owner sees it).
text_sha256stringnullableThe text's SHA-256, hex.
pattern ^[0-9a-f]{64}$
systemobjectrequiredThe bet the system makes, as the rule tester names it. A reader fills what the source says; the caller's hints win.
namestringnullableThe name.
max length 120
competitionstringnullablenfl, ncaaf, nba, wnba, ncaab, nhl, mlb ...
max length 32
marketone of several shapesspread, moneyline, over or under.
sideanyrule (bet the matched team or over/under) or fade (the other side).
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of stringnullable1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
season_typeone of several shapesregular, postseason or all.
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
conditionsarray of objectrequiredOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
source_textstringrequiredThe condition as the source wrote it.
max length 2000
statusanyrequiredexact, close, ask or cant_map.
One of: exact, close, ask, cant_map (open: new values may be added).
formulastringnullableKelly formula (Excel mode, starts with =); null for ask and cant_map.
max length 2000
reasonstringnullableWhy close or can't map; how it differs from the source.
questionstringnullableFor ask: the question for the user.
choicesarray of objectThe readings to choose from, each with its formula.
labelstringrequiredThe choice in words.
formulastringrequiredThe Kelly formula (TRUE or FALSE on each team-game).
data_gapobjectnullableWhat Kelly lacks for a condition it can't import.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: no_attribute, no_history, not_loaded, out_of_grain, licence, edge, look_ahead (open: new values may be added).
detailstringrequiredThe detail, in words.
attributesarray of stringKelly attribute names the formula reads.
methodanydeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
resolutionone of several shapesremoved or live_only, for a condition left out.
formulastringrequirednullableAND of the exact and close conditions.
matchanyrequiredexact, close_match, needs_answers or cant_import.
One of: exact, close_match, needs_answers, cant_import (open: new values may be added).
countsobjectrequiredConditions by status.
exactintegerrequiredConditions mapped exactly.
min 0
closeintegerrequiredConditions mapped closely.
min 0
askintegerrequiredConditions waiting on an answer.
min 0
cant_mapintegerrequiredConditions Kelly can't import.
min 0
checkobjectrequiredkformula.check and the look-ahead check on the combined formula, at system.bet_moment. The leak test runs with the rule tester at confirm.
testablebooleanrequiredWhether the rule tester can test the formula.
problemsarray of objectrequiredEach reason it can't be tested yet.
codestringrequiredThe reason's code.
pathstringnullableThe attribute it concerns.
messagestringrequiredThe reason in words.
fixesarray of objectWays to fix it.
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
ledgerobjectrequiredHow the import counts on the rules-tried ledger at confirm (I1).
labelconstrequiredThe choice in words.
Always "tuned elsewhere".
triesintegerrequiredRules counted as tried.
min 1 · max 1000
tries_fromanyrequireddefault (20) or user.
One of: default, user (open: new values may be added).
labelsarray of object"tuned elsewhere", and "close match" when match is close_match.
readerobjectrequiredThe reader that read it.
namestringrequiredThe name.
methodanyrequireddeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
attemptsintegerThe AI reader's calls (1 to 3).
min 1
warningsarray of objectrequiredNotes on how Kelly read it.
codestringrequiredThe reason's code.
messagestringrequiredThe reason in words.
test_requestobjectnullableThe POST /v1/rules:test body this draft would run; for competition exchange, the POST /v1/rules:test_exchange body (test_exchange_strategy). Null when match is needs_answers or cant_import.
ruleobjectnullableAfter confirm: the saved rule (T0).
rule_idstringnullableThe saved rule's ID (rl_…).
namestringThe name.
versionintegernullableThe saved version.
new_versionbooleannullableWhether confirm made a new version.
versions_triedintegernullableVersions tried, as the rule tester counts.
run_idstringnullableThe rule test kept for it (run_…).
testobjectnullableConfirm only: the rule tester's report on the saved rule (test_rule's facts without the bets, and its summary).
othersarray of stringWhen the source held several systems: the other drafts' ids.
created_atstringnullableWhen the draft was kept.
format date-time
expires_atstringnullableWhen an unconfirmed draft expires (30 days).
format date-time
confirmed_atstringnullableWhen it was confirmed.
format date-time
undo_untilstringnullableUndo is open until then (7 days).
format date-time
contract_versionconstrequiredThe import contract's version.
Always "1.0.0".
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · payload_too_large · not_found · import_not_ready · import_confirmed · import_expired · rule_not_testable · conflict
POST https://api.askkelly.ai/v1/rules:import/imp_7h7p7w2tvm42z7xg6fpy:confirm
Authorization: Bearer $KELLY_API_KEY
Content-Type: application/json{
"name": "Rested home favourites"
}
{
"type": "https://docs.askkelly.ai/problems/import_not_ready",
"title": "Import not ready",
"status": 422,
"code": "import_not_ready",
"detail": "Answer condition 2 first (conditions[].formula from its choices).",
"param": "conditions",
"doc_url": "https://docs.askkelly.ai/problems/import_not_ready",
"request_id": "req_d085c4c4f9bb07331b3d9eaa",
"retryable": false,
"display_message": "That import needs your answers before it can be saved.",
"match": "needs_answers",
"problems": []
}POST /v1/rules:import/{import_id}:undo
scope write:formulas · 0 credits, not charged during the beta · MCP tool undo_rule_import · status live
Hides the rule an import saved, once, within 7 days of confirming. The rules-tried ledger keeps its count (S6). Needs your own live API key with write:formulas; an AI app imports through Kelly MCP, which asks you first.
import_idstring · pathrequiredThe import's ID (imp_…).
pattern ^imp_[0123456789abcdefghjkmnpqrstvwxyz]{20}$
200 The import, undone: its saved rule hidden.
objectconstrequiredAlways rule_import.
Always "rule_import".
idstringrequirednullableThe import's ID (imp_…); null on a dry run.
pattern ^imp_[0-9a-hjkmnp-tv-z]{20}$
stateanyrequireddraft, confirmed, undone or expired.
One of: draft, confirmed, undone, expired (open: new values may be added).
sourceobjectrequiredWhere the system came from and its text.
toolstringnullableThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
tool_labelstringnullableThe tool's name in words.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
formatanyrequiredWhat the text is: sdql, filter_list, formula, plain or auto.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
viaone of several shapesThe entry point that passed it on.
formula_modeone of several shapesThe formula's mode (excel, sql, pandas...).
raw_refstringnullableWhere it came from (a URL, file or message), kept and never fetched.
captured_atstringnullableWhen Kelly received it.
format date-time
textstringnullableThe text as received (kept with the draft and the saved rule, for the proof screen; the owner sees it).
text_sha256stringnullableThe text's SHA-256, hex.
pattern ^[0-9a-f]{64}$
systemobjectrequiredThe bet the system makes, as the rule tester names it. A reader fills what the source says; the caller's hints win.
namestringnullableThe name.
max length 120
competitionstringnullablenfl, ncaaf, nba, wnba, ncaab, nhl, mlb ...
max length 32
marketone of several shapesspread, moneyline, over or under.
sideanyrule (bet the matched team or over/under) or fade (the other side).
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of stringnullable1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
season_typeone of several shapesregular, postseason or all.
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
conditionsarray of objectrequiredOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
source_textstringrequiredThe condition as the source wrote it.
max length 2000
statusanyrequiredexact, close, ask or cant_map.
One of: exact, close, ask, cant_map (open: new values may be added).
formulastringnullableKelly formula (Excel mode, starts with =); null for ask and cant_map.
max length 2000
reasonstringnullableWhy close or can't map; how it differs from the source.
questionstringnullableFor ask: the question for the user.
choicesarray of objectThe readings to choose from, each with its formula.
labelstringrequiredThe choice in words.
formulastringrequiredThe Kelly formula (TRUE or FALSE on each team-game).
data_gapobjectnullableWhat Kelly lacks for a condition it can't import.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: no_attribute, no_history, not_loaded, out_of_grain, licence, edge, look_ahead (open: new values may be added).
detailstringrequiredThe detail, in words.
attributesarray of stringKelly attribute names the formula reads.
methodanydeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
resolutionone of several shapesremoved or live_only, for a condition left out.
formulastringrequirednullableAND of the exact and close conditions.
matchanyrequiredexact, close_match, needs_answers or cant_import.
One of: exact, close_match, needs_answers, cant_import (open: new values may be added).
countsobjectrequiredConditions by status.
exactintegerrequiredConditions mapped exactly.
min 0
closeintegerrequiredConditions mapped closely.
min 0
askintegerrequiredConditions waiting on an answer.
min 0
cant_mapintegerrequiredConditions Kelly can't import.
min 0
checkobjectrequiredkformula.check and the look-ahead check on the combined formula, at system.bet_moment. The leak test runs with the rule tester at confirm.
testablebooleanrequiredWhether the rule tester can test the formula.
problemsarray of objectrequiredEach reason it can't be tested yet.
codestringrequiredThe reason's code.
pathstringnullableThe attribute it concerns.
messagestringrequiredThe reason in words.
fixesarray of objectWays to fix it.
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
ledgerobjectrequiredHow the import counts on the rules-tried ledger at confirm (I1).
labelconstrequiredThe choice in words.
Always "tuned elsewhere".
triesintegerrequiredRules counted as tried.
min 1 · max 1000
tries_fromanyrequireddefault (20) or user.
One of: default, user (open: new values may be added).
labelsarray of object"tuned elsewhere", and "close match" when match is close_match.
readerobjectrequiredThe reader that read it.
namestringrequiredThe name.
methodanyrequireddeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
attemptsintegerThe AI reader's calls (1 to 3).
min 1
warningsarray of objectrequiredNotes on how Kelly read it.
codestringrequiredThe reason's code.
messagestringrequiredThe reason in words.
test_requestobjectnullableThe POST /v1/rules:test body this draft would run; for competition exchange, the POST /v1/rules:test_exchange body (test_exchange_strategy). Null when match is needs_answers or cant_import.
ruleobjectnullableAfter confirm: the saved rule (T0).
rule_idstringnullableThe saved rule's ID (rl_…).
namestringThe name.
versionintegernullableThe saved version.
new_versionbooleannullableWhether confirm made a new version.
versions_triedintegernullableVersions tried, as the rule tester counts.
run_idstringnullableThe rule test kept for it (run_…).
testobjectnullableConfirm only: the rule tester's report on the saved rule (test_rule's facts without the bets, and its summary).
othersarray of stringWhen the source held several systems: the other drafts' ids.
created_atstringnullableWhen the draft was kept.
format date-time
expires_atstringnullableWhen an unconfirmed draft expires (30 days).
format date-time
confirmed_atstringnullableWhen it was confirmed.
format date-time
undo_untilstringnullableUndo is open until then (7 days).
format date-time
contract_versionconstrequiredThe import contract's version.
Always "1.0.0".
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found · undo_closed
POST https://api.askkelly.ai/v1/rules:import/imp_rm4wkv9vhwevdmh05d1z:undo
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/undo_closed",
"title": "Undo closed",
"status": 409,
"code": "undo_closed",
"detail": "Only a confirmed import can be undone.",
"param": "import_id",
"doc_url": "https://docs.askkelly.ai/problems/undo_closed",
"request_id": "req_845d62b82ce36f74ebf78092",
"retryable": false,
"display_message": "That import can't be undone any more."
}GET /v1/rules/imports
scope read:formulas · 0 credits, not charged during the beta · MCP tool list_rule_imports · status live
Your kept imports, newest first: drafts waiting on answers, the ones you confirmed (with the saved rule's name and when undo closes), undone and expired drafts. state keeps one state; a dry run is never kept, so it never appears.
statestring · queryOnly imports in this state (expired: a draft past its 30 days).
One of: draft, confirmed, undone, expired (open: new values may be added).
limitinteger · queryImports per page, 1 to 100.
min 1 · max 100 · default 20
cursorstring · queryThe next_cursor of the page before.
max length 200
200 A page of your kept imports, newest first.
objectconstrequiredAlways list.
Always "list".
dataarray of objectrequiredThe page's imports, newest first.
objectconstrequiredAlways rule_import.
Always "rule_import".
idstringrequirednullableThe import's ID (imp_…); null on a dry run.
pattern ^imp_[0-9a-hjkmnp-tv-z]{20}$
stateanyrequireddraft, confirmed, undone or expired.
One of: draft, confirmed, undone, expired (open: new values may be added).
sourceobjectrequiredWhere the system came from and its text.
toolstringnullableThe tool it came from (sdql, bet_labs, betiq...).
pattern ^[a-z0-9_.:-]{1,64}$
tool_labelstringnullableThe tool's name in words.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: paste, url, file, image, email, extension, api, mcp, chat, bets, template (open: new values may be added).
formatanyrequiredWhat the text is: sdql, filter_list, formula, plain or auto.
One of: auto, sdql, filter_list, formula, plain, game_list (open: new values may be added).
viaone of several shapesThe entry point that passed it on.
formula_modeone of several shapesThe formula's mode (excel, sql, pandas...).
raw_refstringnullableWhere it came from (a URL, file or message), kept and never fetched.
captured_atstringnullableWhen Kelly received it.
format date-time
textstringnullableThe text as received (kept with the draft and the saved rule, for the proof screen; the owner sees it).
text_sha256stringnullableThe text's SHA-256, hex.
pattern ^[0-9a-f]{64}$
systemobjectrequiredThe bet the system makes, as the rule tester names it. A reader fills what the source says; the caller's hints win.
namestringnullableThe name.
max length 120
competitionstringnullablenfl, ncaaf, nba, wnba, ncaab, nhl, mlb ...
max length 32
marketone of several shapesspread, moneyline, over or under.
sideanyrule (bet the matched team or over/under) or fade (the other side).
One of: rule, fade (open: new values may be added).
bet_momentanyclose or edition.
One of: open, edition, close (open: new values may be added).
seasonsarray of stringnullable1-25 season labels ("2024" or "2024-25"); or from_season and to_season.
min items 1 · max items 25
season_typeone of several shapesregular, postseason or all.
stakeanyrisk or to_win.
One of: risk, to_win (open: new values may be added).
strategyobjectnullableCompetition exchange: the trading strategy's actions besides entry_when (exchange.contract.ACTIONS); the test runs them with the draft's formula as entry_when.
entry_sidestringback (bet it to win) or lay (bet it to lose).
One of: back, lay (open: new values may be added).
entry_orderstringtake: match now at the best price on the other side (back at back_price), walking down the ladder up to max_slippage_ticks. join: offer at the best price on your own side (back at lay_price) and wait in the queue, entry_offset_ticks better or worse. limit: offer at entry_price.
One of: take, join, limit (open: new values may be added).
entry_pricenumberlimit only: the price to offer at (a back takes this or longer, a lay this or shorter).
entry_offset_ticksintegerjoin only: ticks from the best own-side price; positive asks for a better price (a back offered longer), negative a worse one.
max_slippage_ticksintegertake only: how many ticks past the best price a take may match at; what is left waits at the last price taken.
stakenumberThe entry stake in the market's currency (a lay's backer's stake unless stake_type is liability).
stake_typestringlay only: stake is the backer's stake you accept; liability is what you can lose.
One of: stake, liability (open: new values may be added).
max_entriesintegerEntries per runner per market (Bet Angel's 'place bet triggers on the selection'); the next waits until the last is closed.
fill_or_kill_snumberSeconds an unmatched entry may wait before it is cancelled (0 cancels what doesn't match at once). Blank: it waits until the market turns in play.
keep_in_playbooleanTRUE keeps unmatched orders when the market turns in play (Betfair's persist); FALSE (default) cancels them at the off (lapse).
take_profit_ticksintegerOnce the entry is matched, offer the green-up exit this many ticks in profit and wait for it to match.
stop_loss_ticksintegerWhen the position is this many ticks against (position_ticks <= -n), cancel the take-profit and green up now at the best price (a take).
trailing_stopbooleanThe stop loss follows the best position_ticks so far (it trails n ticks behind).
exit_whenstringA formula on the runner's ticks that greens up now at the best price when true (may read position_* columns).
exit_before_off_snumberGreen up at the best price this many seconds before the scheduled off (0 is at the off). Blank: no time exit.
green_upbooleanTRUE (default): exits are sized to make the same profit on every outcome. FALSE: no exit, the matched entry is held to the result.
commissionnumberThe exchange's commission on a market's net winnings, as a share (0.05 is 5%).
min_stakenumberThe exchange's smallest stake: a green-up smaller than this can't be placed, and the position is held to the result.
fill_modelstringqueue (default): a waiting order matches only after the money queued ahead of it at that price has traded. touch: it matches as soon as its price trades (optimistic).
One of: queue, touch (open: new values may be added).
conditionsarray of objectrequiredOne entry per condition the source states, in its order.
max items 60
indexintegerrequiredThe condition's place in the source, from 0.
min 0
source_textstringrequiredThe condition as the source wrote it.
max length 2000
statusanyrequiredexact, close, ask or cant_map.
One of: exact, close, ask, cant_map (open: new values may be added).
formulastringnullableKelly formula (Excel mode, starts with =); null for ask and cant_map.
max length 2000
reasonstringnullableWhy close or can't map; how it differs from the source.
questionstringnullableFor ask: the question for the user.
choicesarray of objectThe readings to choose from, each with its formula.
labelstringrequiredThe choice in words.
formulastringrequiredThe Kelly formula (TRUE or FALSE on each team-game).
data_gapobjectnullableWhat Kelly lacks for a condition it can't import.
kindanyrequiredHow it reached Kelly: paste, url, file, image, email, extension, api, mcp, chat, bets or template.
One of: no_attribute, no_history, not_loaded, out_of_grain, licence, edge, look_ahead (open: new values may be added).
detailstringrequiredThe detail, in words.
attributesarray of stringKelly attribute names the formula reads.
methodanydeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
resolutionone of several shapesremoved or live_only, for a condition left out.
formulastringrequirednullableAND of the exact and close conditions.
matchanyrequiredexact, close_match, needs_answers or cant_import.
One of: exact, close_match, needs_answers, cant_import (open: new values may be added).
countsobjectrequiredConditions by status.
exactintegerrequiredConditions mapped exactly.
min 0
closeintegerrequiredConditions mapped closely.
min 0
askintegerrequiredConditions waiting on an answer.
min 0
cant_mapintegerrequiredConditions Kelly can't import.
min 0
checkobjectrequiredkformula.check and the look-ahead check on the combined formula, at system.bet_moment. The leak test runs with the rule tester at confirm.
testablebooleanrequiredWhether the rule tester can test the formula.
problemsarray of objectrequiredEach reason it can't be tested yet.
codestringrequiredThe reason's code.
pathstringnullableThe attribute it concerns.
messagestringrequiredThe reason in words.
fixesarray of objectWays to fix it.
source_recordobjectnullableThe source's own record, as reported; never Kelly's numbers.
winsintegernullableWins as the source wrote them.
min 0
lossesintegernullableLosses as the source wrote them.
min 0
pushesintegernullablePushes as the source wrote them.
min 0
unitsnumbernullableUnits as the source wrote them.
roinumbernullableA fraction (0.078).
sincestringnullableThe first season or date the source counted from.
marketstringnullableThe bet the record is on, as the source said it.
as_reportedbooleanAlways true: the source's claim, copied (T9f's pre-pass).
gamesarray of objectThe source's own list of games.
max items 5000
datestringThe game's date.
teamstringThe team.
opponentstringnullableThe opponent.
homebooleannullableWhether the team was at home.
linenumbernullableThe line taken.
pricenumbernullableAmerican odds as listed.
resultstringnullableThe result as the source wrote it.
unitsnumbernullableUnits as the source wrote them.
textstringnullableThe row as the source wrote it.
ledgerobjectrequiredHow the import counts on the rules-tried ledger at confirm (I1).
labelconstrequiredThe choice in words.
Always "tuned elsewhere".
triesintegerrequiredRules counted as tried.
min 1 · max 1000
tries_fromanyrequireddefault (20) or user.
One of: default, user (open: new values may be added).
labelsarray of object"tuned elsewhere", and "close match" when match is close_match.
readerobjectrequiredThe reader that read it.
namestringrequiredThe name.
methodanyrequireddeterministic, ai or user.
One of: deterministic, ai, user (open: new values may be added).
attemptsintegerThe AI reader's calls (1 to 3).
min 1
warningsarray of objectrequiredNotes on how Kelly read it.
codestringrequiredThe reason's code.
messagestringrequiredThe reason in words.
test_requestobjectnullableThe POST /v1/rules:test body this draft would run; for competition exchange, the POST /v1/rules:test_exchange body (test_exchange_strategy). Null when match is needs_answers or cant_import.
ruleobjectnullableAfter confirm: the saved rule (T0).
rule_idstringnullableThe saved rule's ID (rl_…).
namestringThe name.
versionintegernullableThe saved version.
new_versionbooleannullableWhether confirm made a new version.
versions_triedintegernullableVersions tried, as the rule tester counts.
run_idstringnullableThe rule test kept for it (run_…).
testobjectnullableConfirm only: the rule tester's report on the saved rule (test_rule's facts without the bets, and its summary).
othersarray of stringWhen the source held several systems: the other drafts' ids.
created_atstringnullableWhen the draft was kept.
format date-time
expires_atstringnullableWhen an unconfirmed draft expires (30 days).
format date-time
confirmed_atstringnullableWhen it was confirmed.
format date-time
undo_untilstringnullableUndo is open until then (7 days).
format date-time
contract_versionconstrequiredThe import contract's version.
Always "1.0.0".
has_morebooleanrequiredWhether another page follows.
next_cursorstringrequirednullablePass as cursor for the next page.
invalid_request · credential_in_url · unsupported_api_version · unauthenticated · method_not_allowed · rate_limited · internal_error · service_unavailable · quota_exhausted · insufficient_scope · not_entitled · not_found
GET https://api.askkelly.ai/v1/rules/imports?limit=2
Authorization: Bearer $KELLY_API_KEY{
"object": "list",
"data": [
{
"id": "imp_rm4wkv9vhwevdmh05d1z",
"rule": null,
"check": {
"problems": [],
"testable": true
},
"match": "exact",
"state": "draft",
"counts": {
"ask": 0,
"close": 0,
"exact": 3,
"cant_map": 0
},
"labels": [
"tuned elsewhere"
],
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"object": "rule_import",
"reader": {
"name": "sdql",
"method": "deterministic",
"attempts": 1
},
"source": {
"via": null,
"kind": "api",
"text": "H and line<=-3 and p:L",
"tool": "sdql",
"format": "sdql",
"raw_ref": null,
"tool_label": "SDQL (KillerSports / SportsDataBase)",
"captured_at": "2026-10-04T17:30:00Z",
"text_sha256": "ac5e42601cf5209bb381275989ee4652f463b74fa25190e705dfa422f87f9f96",
"formula_mode": null
},
"system": {
"name": null,
"side": "rule",
"stake": "risk",
"market": "spread",
"seasons": [
"2023",
"2024"
],
"bet_moment": "close",
"competition": "nfl",
"season_type": "regular"
},
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"warnings": [
{
"code": "reader",
"message": "Kelly reads earlier games (p:, P:) within the season tested, so a team's first game of a season has no previous game."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
}
],
"conditions": [
{
"index": 0,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=AND(home, NOT(neutral))",
"data_gap": null,
"question": null,
"attributes": [
"home",
"neutral"
],
"resolution": null,
"source_text": "H"
},
{
"index": 1,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=spread_close <= -3",
"data_gap": null,
"question": null,
"attributes": [
"spread_close"
],
"resolution": null,
"source_text": "line<=-3"
},
{
"index": 2,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=LAG(result = \"L\")",
"data_gap": null,
"question": null,
"attributes": [
"result"
],
"resolution": null,
"source_text": "p:L"
}
],
"created_at": "2026-10-04T17:30:00Z",
"expires_at": "2026-11-03T17:30:00Z",
"undo_until": null,
"confirmed_at": null,
"test_request": {
"side": "rule",
"stake": "risk",
"market": "spread",
"formula": "=AND(AND(home, NOT(neutral)), spread_close <= -3, LAG(result = \"L\"))",
"seasons": [
"2023",
"2024"
],
"bet_moment": "close",
"competition": "nfl",
"season_type": "regular"
},
"source_record": null,
"contract_version": "1.0.0"
},
{
"id": "imp_7h7p7w2tvm42z7xg6fpy",
"rule": null,
"check": {
"problems": [],
"testable": true
},
"match": "needs_answers",
"state": "draft",
"counts": {
"ask": 1,
"close": 0,
"exact": 2,
"cant_map": 1
},
"labels": [
"tuned elsewhere"
],
"ledger": {
"label": "tuned elsewhere",
"tries": 20,
"tries_from": "default"
},
"object": "rule_import",
"reader": {
"name": "filters:bet_labs",
"method": "deterministic",
"attempts": 1
},
"source": {
"via": null,
"kind": "api",
"text": "Home/Away: Home\nFavorite/Underdog: Favorite\nRest Advantage: 2+ days\nPublic Betting %: under 40%",
"tool": "bet_labs",
"format": "filter_list",
"raw_ref": null,
"tool_label": "Bet Labs (Action Network)",
"captured_at": "2026-10-04T17:30:00Z",
"text_sha256": "076d17982bcb14c5005dc77136b1d087fad4b9e64ee54cd3917480fd10d22cbd",
"formula_mode": null
},
"system": {
"name": null,
"side": "rule",
"stake": "risk",
"market": "spread",
"seasons": [
"2020",
"2021",
"2022",
"2023",
"2024",
"2025",
"2026"
],
"bet_moment": "close",
"competition": "nfl",
"season_type": "regular"
},
"formula": "=AND(AND(home, NOT(neutral)), spread_close < 0)",
"warnings": [
{
"code": "reader",
"message": "Kelly's names for bet_labs's filters come from its help pages, not a real export; check each condition against the source."
},
{
"code": "needs_market",
"message": "Say which bet the system makes: spread, moneyline, over or under. Kelly assumed the spread."
},
{
"code": "line_source",
"message": "Kelly tests at its own consensus line (closing lines from June 2020), which can differ from the source's line by a half point; the proof step compares the two game by game."
},
{
"code": "seasons",
"message": "The source names no seasons; Kelly tests 2020 to 2026, every season it holds lines for."
}
],
"conditions": [
{
"index": 0,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=AND(home, NOT(neutral))",
"data_gap": null,
"question": null,
"attributes": [
"home",
"neutral"
],
"resolution": null,
"source_text": "Home/Away: Home"
},
{
"index": 1,
"method": "deterministic",
"reason": null,
"status": "exact",
"choices": [],
"formula": "=spread_close < 0",
"data_gap": null,
"question": null,
"attributes": [
"spread_close"
],
"resolution": null,
"source_text": "Favorite/Underdog: Favorite"
},
{
"index": 2,
"method": "deterministic",
"reason": null,
"status": "ask",
"choices": [],
"formula": null,
"data_gap": null,
"question": "Rest Advantage: what range did you set? Kelly read '2+ days'.",
"attributes": [],
"resolution": null,
"source_text": "Rest Advantage: 2+ days"
},
{
"index": 3,
"method": "deterministic",
"reason": "Bets %: Kelly holds betting splits only from Nov 2026 (S8).",
"status": "cant_map",
"choices": [],
"formula": null,
"data_gap": {
"kind": "no_history",
"detail": "Kelly holds betting splits only from Nov 2026 (S8)"
},
"question": null,
"attributes": [],
"resolution": null,
"source_text": "Public Betting %: under 40%"
}
],
"created_at": "2026-10-04T17:30:00Z",
"expires_at": "2026-11-03T17:30:00Z",
"undo_until": null,
"confirmed_at": null,
"test_request": null,
"source_record": null,
"contract_version": "1.0.0"
}
],
"has_more": false,
"next_cursor": null
}GET https://api.askkelly.ai/v1/rules/imports?cursor=bm90LWEtY3Vyc29y
Authorization: Bearer $KELLY_API_KEY{
"type": "https://docs.askkelly.ai/problems/invalid_request",
"title": "Invalid request",
"status": 400,
"code": "invalid_request",
"detail": "That cursor isn't one Kelly gave.",
"param": "cursor",
"doc_url": "https://docs.askkelly.ai/problems/invalid_request",
"request_id": "req_1ca1bd7a5f343108e1e25b49",
"retryable": false,
"display_message": "Something in that request was not understood."
}Generated from Kelly API 2026-10-05.
Support: support@askkelly.ai · Security and disclosure