Open source  ·  MIT License  ·  v3.30

Sports intelligence
for AI agents

SportMind teaches AI agents how to reason about sports — the commercial, financial, and competitive intelligence the industry runs on.

View on GitHub Explore the library
456
Skill files
126
Calibration records
96%
Direction accuracy
21
Sports validated

The problem

AI agents don't know sports.

A general-purpose LLM given a match to analyse produces generic output. It does not know the things that actually change outcomes. SportMind teaches agents exactly those things.

Cricket
An evening T20 in Mumbai with 78% humidity means the team batting second has a structural advantage. Dew accumulates. Spin bowling fails. Standard models miss this.
MMA
A fighter missing weight signals preparation failure, altered psychology, and physiological compromise. It is categorically different from a team losing a regular game.
Formula 1
A 0.3 second qualifying advantage predicts race outcomes more reliably than season form. But only on certain circuit types. SportMind knows which ones.
Fan tokens
With gamified tokenomics, a WIN prediction is simultaneously a SUPPLY REDUCTION event. Tokens burn on wins, mint on losses. A standard model sees half the signal.

The library

Five layers. One system.

Load in order — macro → market → sport domain → athlete → fan token. Compatible with any LLM. Skills are structured markdown, not API wrappers.

Layer 1
Sport domain
42 sports. Event playbooks, risk variables, signal weights, and agent reasoning prompts.
sports/
Layer 2
Athlete intelligence
29 sports. Form, availability, composite modifier (0.55–1.25×).
athlete/
Layer 3
Fan token commercial
37 skills. Lifecycle phases 1–5e, gamified tokenomics, DeFi, governance, RWA.
fan-token/
Layer 4
Market intelligence
42 documents. Commercial tier, fanbase depth, competition calendar.
market/
Layer 5
Macro intelligence
9 documents. Crypto cycles, regulatory frameworks (MiCA, SEC/CFTC), geopolitical.
macro/

Signal output

What SportMind produces.

Every analysis returns a consistent structured output — direction, confidence score, modifiers applied, and active flags.

sportmind · pre-match signal ⚽ football · ucl 
{
  "direction":           "HOME",
  "adjusted_score":      72.4,
  "sms":                 79,
  "recommended_action":  "ENTER",
  "composite_modifier":  1.10,

  "modifiers_applied": {
    "athlete_modifier":    1.10,
    "macro_modifier":      1.00
  },

  "flags": {
    "lineup_unconfirmed":    false,
    "macro_override_active": false
  }
}
PSG vs Arsenal · UCL · Parc des Princes v3.30

Empirical foundation

Every claim is verifiable.

126 records across 21 sports. All in the repository. Submitted before real matches. Including the 5 wrong ones.

96%
121 of 126 records correct
Direction accuracy 121 / 126
Zero wrong-direction records
qualifying_delta (F1)4/4 ✓
dew_factor (cricket)5/5 ✓
india_pakistan ×2.003/3 ✓
morning_skate (NHL)3/3 ✓
raider_primacy (kabaddi)1/1 ✓
21 sports calibrated
football cricket basketball mma formula1 tennis ice-hockey rugby-union rugby-league afl athletics motogp snooker darts swimming winter-sports kabaddi handball nascar netball rowing

Contribute

Every record moves a modifier forward.

The first 10 external contributors become Founding Calibrators — permanently credited in the library's history.

★
Submit a calibration record
Run SportMind before a real match. Submit the outcome. No coding required. Wrong predictions are as valuable as correct ones.
30 min
★★
Expand a stub sport
14 sports are stubs waiting for domain knowledge — badminton, volleyball, table tennis, and more.
2–3 hrs
★★★★
External recalibration
Submit 10+ records for one modifier and write the analysis. First external voice in SportMind's calibration history.
8+ hrs
Read the First Record Challenge →

Free to use.
Free to build on.

MIT licensed. Compatible with any LLM.
The library is ready.

View on GitHub → Find your starting point