State of MMA Forecasting

Name: Blueprint MMA Q3 2026 UFC Forecast Benchmark
Creator: Blueprint MMA

Q3 2026 · Blueprint MMA public benchmark

See the complete settled sample behind Blueprint’s headline results: every official pick, win, miss, published win probability, and known limitation for the period.

Current as of Oct 27, 2026, 9:31 PM UTC · Source revision ac96ebae7189009a

Download PDF

Settled forecasts

257

Winner accuracy

58.8%

Wins

151

Misses

106

Win-chance score · lower is better

0.2427

Did the published percentages match the results?

Over a large enough sample, picks published near 70% should win about seven times in ten. Each range stays visible, including ranges with only a few fights.

Probability	Expected	Observed	N	Flag
50–60%	55%	56.3%	160	within band
60–70%	65%	61.8%	76	within band
70–80%	75%	72.2%	18	within band
80–90%	85%	33.3%	3	collecting

How results are counted

The prediction must be published before the fight and remain bound to the settled fight identity. When an event lock ran, the graded pick is the one published before the lock.
Rows with grading_basis legacy_attested_publication or legacy_attested_article (fights from May 30 to July 19, 2026) predate the forecast proof ledger. They are graded from the forecast or article that was public before the fight, have no proof record, and some of those cards had no event lock (lock_executed is false).
The July 25, 2026 card (grading_basis prefight_publication_disclosed_regrade) is graded from the forecasts that were public before its scheduled lock, which never ran. See /corrections#receipt-track-record-regrade.
Wins and losses are graded against the published winner pick; draws, no contests and voids are excluded.
Corrections create an auditable canonical revision and trigger benchmark regeneration.

Check every result See how it works Correction history

Scope and limitations

Official, published and settled UFC winner predictions with a finite pre-fight win probability.

Accuracy and Brier score describe winner-call quality and probability calibration; they do not establish betting profitability.
The sample can change as fights settle or verified corrections are applied.
Engine-versus-market comparisons are reported only when a timestamped, identity-matched market cohort is available.
Prices are American odds with the bookmaker margin included. publish_odds is the price saved with the pick; close_odds is the median of at least two books captured within five minutes of the start; clv_points is the move of the de-vigged pick-side probability between them. Picks without a saved price or a two-book close leave these fields empty.

Download the data

Use the PDF for a readable copy, the CSV for every result row, or the JSON for structured metrics.

Q3 report JSON Row-level benchmark CSV Frozen Q3 PDF

Get the weekly card update.

Official picks, important changes, and settled results, each linked to its public page.

By subscribing you agree to receive marketing emails. Unsubscribe anytime. See our Privacy Policy.