Methodology

Knockout MMA runs on an outcome-distribution model: for every UFC fight it estimates a full probability distribution over who wins, how it ends (KO/TKO, submission, decision), and when (round by round) — not just a winner. Everything free on this site is a view of that one distribution: the win-probability bars, the method-by-round grid, the survival curve, the finish forecast.

Predictions are the product, and predictions can't be front-run — which is why they're free.

Opening prices at retail sportsbooks move slower than the market's true consensus. We hold two independent estimates of every fight's fair price: our model's calibrated probability, and the no-vig fair implied by the sharpest books. When both agree that a posted price is materially wrong, that's a play.

The edge is as much about execution and timing as prediction — it exists because some prices are slow, not because we know something nobody else does. For exactly that reason, the selection rules themselves (which prices, how mispriced, at which moments) stay private: an edge that depends on slow prices dies when you teach the market to speed up.

Walk-forward: train on the past, bet forward on fights the model had never seen, roll the window ahead, repeat — across 2023–2026. No look-ahead: every input is timestamped before the fight, and every bet is graded at a price that actually existed. Staking in the backtest is flat, by the same rules the live ledger uses.

Across 644 bets that's +14.2% ROI, +91.6 units — +$9,160 at a flat $100 a bet, scaled to whatever your unit is — with a 95% confidence interval on ROI of +5.8% to +23%. The interval is wide on purpose — 644 bets is a real sample, not an infinite one, and we'd rather show you the uncertainty than hide it.

Every official play is staked flat — one or two units, by a fixed rule set in advance. No Kelly, no doubling after losses, no gut-feel overrides, no retroactive resizing. The backtest assumed these rules, the live ledger grades with them, and this page documents them: one sizing story everywhere.

Variance is part of the deal — at the backtest's 56% win rate, losing streaks of seven or eight in a row are statistically expected over a full year of betting. A system you can't sit through is a system you don't have.

Official plays come only from markets where the system is validated end-to-end. Everything else on this site — every other edge readout, model number, and market row — is model reference: interesting, public, and explicitly not a bet. If it isn't in the ledger, it isn't a play, and it isn't in the track record.

Before every card, the official plays are frozen into a file and that file's SHA-256 fingerprint is published before the first fight. After the card, the full picks and their grades are revealed — anyone can re-hash the revealed file and check it against the pre-fight commitment. Picks can't be edited, back-dated, or quietly forgotten.

The running ledger lives on the track record page.

A backtest is evidence, not a guarantee. Live results will be streakier than any aggregate suggests, some cards will be ugly, and no honest model can promise otherwise. We will never claim proven profit — we publish the record, the method, and the losing streaks, and let you judge. Nothing on this site is betting advice.