HISTORICAL MEMORY 16 YEARSMATCHES IN DATABASE 124,000+ODDS RECORDS 393,458LEAGUES TRACKED 22BETFAIR CLOSING ODDS LOADEDPINNACLE CLOSING ODDS LOADEDPOLYMARKET MARKETS LIVEPAPER TRADES 1 SETTLEDAGENT STATUS ACTIVEHYPOTHESES TESTED 7 · 0 PROMOTEDIN-PLAY MONITOR LIVEPREDICTION MARKETS BETFAIR · POLYMARKETHISTORICAL MEMORY 16 YEARSMATCHES IN DATABASE 124,000+ODDS RECORDS 393,458LEAGUES TRACKED 22BETFAIR CLOSING ODDS LOADEDPINNACLE CLOSING ODDS LOADEDPOLYMARKET MARKETS LIVEPAPER TRADES 1 SETTLEDAGENT STATUS ACTIVEHYPOTHESES TESTED 7 · 0 PROMOTEDIN-PLAY MONITOR LIVEPREDICTION MARKETS BETFAIR · POLYMARKET
● LIVE EXPERIMENT · AI RESEARCH · PREDICTION MARKETS

CAN AN AI AGENT
BEAT
PREDICTION MARKETS?_

Testing the limits of market efficiency. We deployed a recursive AI agent to find Alpha in the world's sharpest sports prediction markets — using matches and 16 years of historical memory. Every hypothesis, every failure, every position: logged publicly.

--
MATCHES IN DATABASE
--
ODDS RECORDS
--
LEAGUES TRACKED
2010-2026
DATA RANGE
↓ SCROLL
THE EXPERIMENT
WHY PREDICTION MARKETS?

Betfair Exchange is the most efficient sports prediction market in the world. Prices are set by sharp money competing against each other — not a bookmaker with a built-in margin. If a signal survives here, it survives everywhere. Beat prediction markets here and nothing else is a problem.

CLV: THE ONLY METRIC THAT MATTERS

Closing Line Value measures whether the agent found Alpha — a better price than the market settled at. Positive ROI with negative CLV is luck. Positive CLV with negative ROI (short-term) is a real edge. The agent measures both — and won't promote a signal without beating the closing line.

HOW IT WORKS
01
RECURSIVE HYPOTHESIS
The agent generates non-obvious signals to test — no human preconceptions. Each failure feeds the next hypothesis.
02
MASSIVE BACKTESTING
Python runs all statistics against 16 years of historical data. Walk-forward validation only. Zero lookahead bias.
03
ADVERSARIAL CRITIC
A second AI sub-agent tries to destroy every hypothesis. Hunts for p-hacking, small samples, out-of-sample decay.
04
PUBLIC EXECUTION
Survivors are executed live on Betfair Exchange. Every position posted pre-kickoff. 200 minimum before any verdict.
HYPOTHESIZEBACKTESTFALSIFYEXECUTE
THE DATA FOUNDATION
WHAT'S LOADED
Match results + half-time scores
Betfair Exchange closing odds
Pinnacle closing odds (sharp benchmark)
Bet365, B365, market odds (9 books)
Match stats: shots, corners, cards
xG data — Big 5 leagues (FBref)
Betfair historical opening pricesCOMING
Polymarket / Kalshi prediction markets
27 LEAGUES · 20102026
Premier League
Championship
League One
League Two
La Liga
Serie A
Bundesliga
Ligue 1
Segunda División
Serie B
2. Bundesliga
Ligue 2
Eredivisie
Liga Portugal
Süper Lig
Pro League
Scottish Prem
Super League GR
Champions League
Europa League
World Cup
Euros
+ 5 more
NON-NEGOTIABLE RULES
1
ZERO LOOKAHEAD BIAS
Every feature must be available before kickoff. No post-hoc data. No exceptions.
2
CLV IS KING
Beat the closing line or it's luck. Positive ROI with negative CLV gets rejected.
3
200 SELECTIONS MIN
No verdict before 200 positions. A hot streak is not Alpha. Sample size is everything.
4
CRITIC MUST FAIL
The adversarial agent's job is to destroy every hypothesis before it gets executed.
5
ALL SIGNALS PRE-KICKOFF
Every position is posted publicly before the match starts. No retroactive claims. Ever.
6
GRAVEYARD IS PUBLIC
Every rejected hypothesis is logged. The failures are as valuable as the wins.
JOIN THE EXPERIMENT

FOLLOW EVERY POSITION.
EVERY FAILURE. EVERY EDGE.

Weekly digest: all live positions with pre-kickoff reasoning, new hypotheses, Alpha summary, and everything the agent got wrong. The graveyard is as public as the leaderboard.

No spam. Unsubscribe anytime. All picks posted before kickoff.