Public model accountability

Model scorecard

Every Combat Edge prediction is generated before fight night, frozen, and graded here once results land — hits and misses alike. Explore the complete record by event, promotion, model version, confidence and time period, with the same fights scored against an Elo-only baseline. Analytics, not betting advice.

62%Model accuracy
356Graded fights
63%Elo-only baseline
v0.7-2026-08-28Current frozen version
Explore the record356 of 356 graded fights
Combat Edge model62%222 of 356 fights called correctlyBrier 0.228 · Log-loss 0.647
Elo-only baseline63%225 of 356 fights called correctlyBrier 0.234 · Log-loss 0.660
50/50 probability baselineEven oddsThe same probability for every fighterBrier 0.250 · Log-loss 0.693

Evaluation scope all stored prediction records

555Stored predictions
122Upcoming / result window
433Evaluation candidates
356Graded
77Excluded

A record becomes an evaluation candidate six hours after its event starts. It is graded only when it was generated before that start time, maps to the stored bout, has valid probabilities, and the result names one of the two fighters as winner.

Why records were excluded all ungraded evaluation candidates

ReasonRecords
No recorded winner (pending, draw, NC or cancelled)57
Generated at or after event start20

Live calibration confidence should match observed win rate

A calibrated model’s confidence and observed win rate should stay close. Sample size remains visible because small bands move sharply.
View exact calibration data
ConfidenceFightsAverageObservedGap
50–59%14754.0%54.4%+0.5 pts
60–69%13663.8%61.8%-2 pts
70–79%5574.1%83.6%+9.5 pts
80–89%1684.1%62.5%-21.6 pts
90–100%292.3%100.0%+7.7 pts

By model version forward-only · frozen predictions

VersionGradedAccuracyLog-lossBrierPeriod
v0.7-2026-08-28 Current12560%0.7000.251Aug 28, 2026 – Sep 15, 2026
v0.5-2026-07-0119762%0.6230.218Jul 10, 2026 – Aug 28, 2026
v3-2026-06-103471%0.5880.199Jun 20, 2026 – Jun 27, 2026
Event reports

Model scorecards by event

12 of 36 events

Open an event to see every graded pick, the Elo comparison and the model’s most confident hit and miss. Event summaries use only the active filters above.

UFC · Sep 15, 2026CONTENDER SERIES 2026 : WEEK 63/560% accuracyElo +1
ZUFFA · Sep 12, 2026ZUFFA BOXING : GARCIA VS BENN7/978% accuracyModel tied Elo
UFC · Sep 12, 2026NOCHE UFC : SILVA VS DELGADO9/1369% accuracyModel +2 vs Elo
OKTAGON · Sep 12, 2026OKTAGON 93 : ROUSAL VS MAGARD6/967% accuracyModel +1 vs Elo
ACA · Sep 12, 2026ACA 207 : GONCHAROV VS ALMEIDA5/1436% accuracyElo +2
LFA · Sep 11, 2026LFA 241 : PIRES VS PEREIRA8/1457% accuracyElo +1
UFC · Sep 8, 2026CONTENDER SERIES 2026 : WEEK 52/450% accuracyElo +1
UFC · Sep 5, 2026UFC FIGHT NIGHT : HOOKER VS PARNASSE8/1457% accuracyElo +3
SAMOURAI · Sep 4, 2026SAMOURAI MMA 21 : DUFORT VS CHAPUT5/863% accuracyElo +1
CFFC · Sep 4, 2026CFFC 160 : KOSTYUCHENKO VS VEAL6/875% accuracyModel tied Elo
UFC · Sep 1, 2026CONTENDER SERIES 2026 : WEEK 43/560% accuracyElo +1
UFC · Aug 29, 2026UFC FIGHT NIGHT : NURMAGOMEDOV VS SONG8/1362% accuracyModel +3 vs Elo

How grading works

  • Frozen pre-fight. Predictions are written by an automated job while an event is upcoming and stop updating once it starts. Only records generated before the stored event time are graded — nothing is back-filled.
  • Eligibility is explicit. Every valid frozen prediction with a recorded winner counts: favorites, underdogs and prelims. Missing results, invalid mappings and late records stay out of the score and remain visible in the exclusion audit.
  • Accuracy is the share of fights where the fighter the model gave ≥50% actually won.
  • Brier score and log-loss grade the probability, not just the pick. Lower is better; confident misses are penalized more. A 50/50 forecast has Brier 0.250 and log-loss 0.693 on any set of binary results.
  • Live calibration groups the model’s chosen fighter by confidence, then compares average confidence with how often those fighters actually won.
  • Simple baselines. The same graded fights are scored using Elo expected outcomes. The fixed 50/50 line is theoretical and does not claim an accuracy because it does not choose a side.
  • Current model: v0.7-2026-08-28, trained on historical fights with point-in-time features only — no data from after each fight leaks in.