The model's best call was its quietest one

Alberto Montes was not the favorite against Ricky Turcios on March 7. Our model gave him a 37% chance, a medium-confidence read that landed on the underdog side of the headliner. Montes won anyway, finishing Turcios by submission in the second round. The system got the winner right while the broader card slipped below its usual line.

That split is the story of the night: a correct, against-the-grain headliner call wrapped inside one of the model's softer card grades of the year.

The night in brief

Montes closed the main event with a second-round submission over Turcios. The model had named Montes as its pick despite the lower implied probability, so the headliner went in the win column.

Two fights ended almost before they began. Rodolfo Bellato scored a first-round knockout, and Gregory Rodrigues did the same, both finishes coming inside the opening five minutes. Two first-round knockouts on one card is the kind of result that leaves little for a prediction model to work with.

The model autopsy

This is where we grade ourselves in public, misses included.

Across the full card, the model went 7/12 for 58% accuracy. That is below our canonical live 2026 mark, and we are not going to dress it up. On a 12-fight slate, missing five is a soft night, and a couple of those first-round finishes are the kind of early, low-information outcomes that swing a card grade quickly.

The flagged picks tell a steadier story. On the LOCK and HIGH confidence tier, the model went 2/3. Those are the picks we stand behind most firmly, and hitting two of three keeps the high-confidence band roughly in line with where it should sit even on an off night for the card as a whole. The one miss in that group is the honest cost of the evening: a flagged pick that did not land, folded into the wider 7/12 result.

The headliner is the bright spot, and it is worth being precise about why. The model picked Montes at a 37% implied probability with medium confidence. That is the model disagreeing with the market lean and being right. It was not a lock and it was not a favorite, and it still came in. We log that as a correct call rather than a lucky one, because the pick was on record before the result, but we also will not inflate a 37% read into something it was not. There was no called LOCK on this fight, and no upset flagged elsewhere on the card that we can claim credit for. The medium-confidence headliner is the cleanest thing the model did all night.

No LOCK miss to own here, because there was no LOCK on the card. The accountability that matters is the card grade itself: 58% is below our standard, and the five misses are real.

One honest takeaway

A 58% night does not match the numbers we lead with, and it should not. Our canonical live 2026 record sits at 67.9% across all picks and 81.4% on the LOCK and HIGH tier over 64 graded calls. The Turcios vs Montes card came in under the all-picks line, and the small flagged sample here, 2/3, is too thin to move that 81.4% one way or the other.

The point of grading every card is that the soft ones count too. We do not average a weak night out of existence or quietly drop it from the ledger. The model read the headliner correctly against the odds and held its flagged picks at 2/3, while the rest of the card got away from it. Both of those things are true, and both go on the record at full weight.

The honest version of this recap is that the system earned its headliner call and lost the card. On a free, accountable model, that is exactly the kind of night we publish in full.