Fight Outcome Model + Optimal Bet Sizing
A probability model that stakes real bankroll on its own calibration - and publishes the scoreboard either way. The harshest eval a model can face, sold as an engine for your own outcomes.
FIGHT CARD - SATURDAY · MODEL vs MARKET
Model output - logged pre-fight
Main event · Silva vs Kowalski
logged Thu 18:02 · 46h before close
MODEL 0.62 [0.57-0.67] vs MARKET 0.54 @ 1.85
+8 pp EDGE → BETCo-main · Blake vs Souza
logged Thu 18:02
MODEL 0.51 [0.45-0.57] vs MARKET 0.50 @ 2.00
NO EDGE → NO BET · discipline is the productStake sizing - modern bankroll theory
full Kelly ·········· 12.4% (theoretical)
fractional 0.25× ····· 3.1%
drawdown constraint ✓ · 2.5%
correlation check ✓ ·· single exposure
SEASON CALIBRATION
Brier 0.19 · CLV +2.3%
no-bet rate 61% · all public
How it works
Anyone can post winners after the fight. This model writes its probability down before the odds close, bets only when it disagrees with the market, and lets the scoreboard judge.
Features that matter
Reach and stance matchups, takedown defense, age curves, camp quality, short-notice flags, judge tendencies - engineered from fight-level data, not vibes.
Probabilities, not picks
The output is a calibrated probability with an interval - and the calibration is tracked publicly, so 62% has to mean 62% over the season.
Edge or no bet
The model's number is compared to the market's. No meaningful gap - no bet. Most fights are no-bets; the discipline is the strategy.
Size to survive
Fractional Kelly with drawdown constraints and correlation checks - modern bankroll theory. Growth is a side effect of not being ruined.
The Four Guarantees™ - turned on ourselves
Measured value
Closing-line value and Brier score, published every event. CLV is the one metric that can't be faked by a lucky streak.
Defensible
Every pick timestamped before odds close; features and model version logged; losing runs stay on the record. Nothing is deleted, ever.
Self-correcting
Recalibrated after every event; drift on the market's sharpness watched; when the edge decays, the honest output is "no bet" - not a stretch.
Yours & everywhere
One container, your cloud, full source handover - the same engine pointed at your domain's outcomes, with MCP tools your agents query directly.
The scoreboard, sized honestly
Illustrative season figures - the real ledger is published as it happens, misses included. No bankroll advice is offered or implied.
Three ways to own it
| Tier | What you get | Price | |
|---|---|---|---|
| Scaffolding ★ | The full repo - probability model, calibration tracking, fractional-Kelly sizing engine, eval gate, audit log, MCP server. Trained on public fight data. You own the source. | €2,990 | |
| PoC ★★RECOMMENDED | The engine trained and applied on your domain's outcomes - calibration (Brier, reliability) and decision-sizing quality measured against your historical data. Full code and the quality report are yours. No performance guarantee at this stage, by design. | from €9,000 | |
| Implementation ★★★ | Production: wired to your data feeds, decision sizing live under your risk limits, monitoring + drift alerts, recalibration every cycle, monthly scoreboard - the agreed number (conservative) attaches here. | from €27,000 |
★ = engagement depth. PoC is the recommended path: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.
Born in the lab, proven on the harshest market
What this is not
Ready to see your own number?
Request the build: within 48h you get a personal reply with the value sized to your volume.
No commitment · reply within 48h · your data stays in the EU