Pre-fight
The model's read going in
The walk-forward model's call at the time of the bout — persisted out-of-sample, with no hindsight about the result.
Each bar is a feature's push on the model from the center axis. Red pushes toward St-Pierre, blue toward Hughes. Tap a bar for why.
Odds are model-implied (no real prop market is priced for these). The “most” prices are draw-no-bet — a tie (for takedowns, usually 0–0) is a push, so each side is priced off its chance of winning given a result, while the % shown is the raw chance of landing strictly more. Drag a line in half-point steps: a half line (e.g. 2.5) is a clean over/under, while a whole number adds a third outcome — landing exactly on it, a push — so over, exact and under sum to 100%. On a whole number the over/under odds are draw-no-bet (priced off the decisive mass, since the push refunds), so the two sides stay mirror images rather than implying an arbitrage.
Expected fight time 19.9 min. Predicted-output ranges are percentiles of the model’s simulated fights: sig strikes the middle 50% (25th–75th); takedowns the wider 10th–90th, since they’re 0 in most fights so a narrow band would collapse to 0. Win-by-method = win probability × that fighter’s own method split: the model forecasts how the fight ends, and that forecast is divided between the corners by each fighter’s finishing record — UFC wins in full, earlier-career wins at half weight, as of fight night — reconciled so the fighters’ totals still equal their win probabilities and the methods still total the fight-level forecast above. “Equity” is each of those over that fighter’s win % — how they win given that they win — so a column adds to 100%, and it differs by corner. A fighter with no classifiable wins at any level leans on the fight-level split. “How it ends” is the model’s round-of-finish split. “Mkt” is the de-vigged market prop (consensus across books), shown for context, not as a betting signal; rounds & totals slots fill as that data lands. Percentages are apportioned by largest remainder, so each set adds to its total exactly and an individual figure can sit a point off its own rounding.
Market % is the de-vigged (no-vig) consensus across the books that price both fighters; the line is the average across books with the min–max range in parentheses. Gap is the model minus the market in points (+ = the model rates that outcome above the market). Thin or one-sided markets are flagged or left blank rather than shown as a fabricated %. Shown for context, not as a bet — the closing line is the sharpest public price and the model trails it on average. For 5-round bouts the model splits its round-3+ finish mass into rounds 3–5 by the historical within-bucket rate, so the over-3.5 / 4.5 lines carry a model estimate.
- WJosh KoscheckU-DEC R3Close -2864mo before
- LMatt SerraKO/TKO R1Close -13009mo before
- WMatt HughesKO/TKO R2Close +1151y 1mo before
- WBJ PennS-DEC R3Close -1151y 10mo before
- WSean SherkKO/TKO R2Close -3002y 1mo before
- WChris LytleU-DEC R3Close -65010mo before
- LGeorges St-PierreKO/TKO R2Close -1451y 1mo before
- WBJ PennKO/TKO R3Close -2251y 3mo before
- WRoyce GracieKO/TKO R1Close -3001y 7mo before
- WJoe RiggsSUB R1Close -3602y 1mo before
Bar length is fight count (largest bar = full width); shading splits finishes by round and decisions by unanimous vs split — tap a bar for the round-by-round count.
Bar length is fight count (largest bar = full width); shading splits finishes by round and decisions by unanimous vs split — tap a bar for the round-by-round count.
Profiles
Opponents' records on fight night — toggle their full cross-promotional record vs UFC bouts only.
- Head69%(257)div 66%
- Body16%(58)div 20%
- Legs15%(56)div 14%
- Head75%(296)div 66%
- Body23%(90)div 20%
- Legs2%(8)div 14%
Where each fighter lands significant strikes (career). “div” = the weight-class average.
Per-fight: SS & TD shown landed/attempted; Close = consensus closing line.
Per-fight: SS & TD shown landed/attempted; Close = consensus closing line.