Served fights with a full model price

70 / 98

board of 21 September 2026

MMA: what we price, and what we refuse to price

Across the 98 fights on our MMA board on 21 September 2026, 70 carried a price from the full model, 12 carried a degraded price, and 16 carried no model price at all. The board says which is which, fight by fight.

Coverage figure Not a performance figure
Full model price, both fighters known70 · 71 %
Degraded, one known, median stats12 · 12 %
No model price16 · 16 %

The 16 we do not price are the point

We could fill them with a regional Elo rating instead. We measured what that would be worth:

Regional Elo, Brier score0.2414
Declared chance, for comparison0.2500
Share of the signal the market captures~22 %

A price that weak, sitting next to prices built on full fight statistics and looking exactly like them, is worse than no price. So we publish the gap instead of covering it.

What this page does not claim

Our MMA engine has not been judged against its own results. There is no out-of-sample accuracy, no calibration error and no closing line value for MMA on this site, because we have not measured them.

Nothing here says our MMA prices are accurate. It says how often we produce one, and how often we decline to. When the engine is measured, it goes on the track record page with its intervals, and not before.

Why the model stops at the UFC

Fights in the model's corpus8,869
Fighters in the corpus2,712
Upcoming fights, 26 organisations307
Of those, both fighters known26 · 8 %
Across 22 regional organisations0

The model is fitted on per-fight statistics: strikes attempted and landed, takedowns, control time. UFCStats is the only source that publishes them. Wikipedia and Tapology publish cards, who fights whom, and when, not statistics. A fighter from KSW, Cage Warriors or Shooto is therefore absent from the corpus, and collecting more cards would not change that: shooto 43 fights, none known; lfa 34, none; cagewarriors 30, none; oktagon 25, none; and ares, hexagone, fury, rizin, ksw, aca, dwcs and the rest all at zero without exception.

Method

The three counts come from one pass over the fights served on the board on 21 September 2026, classified by the scorer's own definition of a known fighter: both fighters present in the model's fighter table. One known fighter means median statistics are substituted for the other, and the fight is labelled degraded. Neither known means no price.

These are a snapshot of one day's board, not an average over the season, and we do not present them as one. A week heavy with regional shows scores lower; a UFC fight week scores higher.

Citing this page

MySyndicate MMA coverage, 21 September 2026: 70 of 98 served fights (71%) with a full model price, 12 (12%) degraded, 16 (16%) unpriced. A coverage figure, not a performance figure, the MMA engine has not been judged against its own results. https://www.mysyndicate.io/mma-coverage