Record

How every official forecast did. Forecasts are locked before each event and graded against official results; nothing is edited afterwards.

Current model

Very small sample

Record5-2Since Tue 27 Oct 2026

Accuracy71.4%N=7 scored fights

Average log loss0.533Lower is better; 0.693 = coin flip

Brier score0.177Lower is better; 0.25 = coin flip

Counts official forecasts locked before each event under the current winner model. Excluded bouts, draws, no contests and bouts removed after locking are not scored.

Does a strong forecast win more often?

When the model gave the favourite at least this probability, how often did the favourite win? With so few fights, these numbers move a lot from event to event.

Current model — favourites by forecast strength
Forecast strength	Fights	Favourite won	Rate
55% or more	5	4	80.0%
60% or more	5	4	80.0%
65% or more	3	3	100.0%
70% or more	2	2	100.0%
75% or more	2	2	100.0%
80% or more	1	1	100.0%

Event by event

Official locked forecasts by event (all models)
Event	Date	Model	Record	Accuracy	Not scored
UFC Fight Night: Moicano vs Nolan	Sat 31 Oct 2026	Current model	5-2	71.4%	5
UFC 333: Volkanovski vs Evloev	Sat 24 Oct 2026	Previous model (V5)	2-6	25.0%	1
UFC Fight Night: Buckley vs Malott	Sat 17 Oct 2026	Previous model (V5)	3-4	42.9%	6
UFC Fight Night: Allen vs Duncan	Sat 10 Oct 2026	Previous model (V5)	5-6	45.5%	1
UFC Fight Night: Rosas Jr. vs Barcelos	Sat 26 Sep 2026	Previous model (V5)	7-2	77.8%	3

Shown separately

These events are not official pre-event snapshots, so they never count toward any headline record.

Non-official records
Event	Date	Type	Record
UFC 332: Silva vs Wang	Sat 3 Oct 2026	Reported backfill	11-2

Method and distance components

Small samples

Separate estimates of whether a fight goes the distance and how it ends. They are tracked on their own and never mixed into the winner record.

Component records (locked forecasts)
Component	Fights	Accuracy	Log loss	Brier
Goes the distance?	12	66.7%	0.625	0.217
Method of victory	12	50.0%	0.935	0.571

Limited-data estimates

Experimental

Bouts where a fighter has fewer than 3 detailed UFC fights get no official forecast. An experimental estimate is tracked separately: ?-? (33.3% of 6 scored). It is not part of the official record.

Archive: previous winner model (17-18)

The previous winner model (V5) was retired on Tue 27 Oct 2026 with a final live record of 17-18 (48.6% of 35; log loss 0.736, Brier 0.27). It remains in use only as a backup for bouts the current model cannot cover.

Previous model — official locked forecasts
Event	Date	Record	Accuracy
UFC 333: Volkanovski vs Evloev	Sat 24 Oct 2026	2-6	25.0%
UFC Fight Night: Buckley vs Malott	Sat 17 Oct 2026	3-4	42.9%
UFC Fight Night: Allen vs Duncan	Sat 10 Oct 2026	5-6	45.5%
UFC Fight Night: Rosas Jr. vs Barcelos	Sat 26 Sep 2026	7-2	77.8%

Research and shadow models are evaluated internally and are not shown here.