Our model gave Javier Reyes a 31% win probability and picked him anyway. He knocked Douglas Silva de Andrade out in the first round, settling the main event before the judges mattered and handing our desk a correct call on the fight it expected to lose more often than not. That was the headliner. Underneath it sat the cleaner story: a 5-for-5 night on every pick we flagged, and 11 of 13 on the card overall.

The Reyes finish was part of a pattern. Beyond the headliner, the Feb. 28 card produced first-round stoppages from Damian Pinas (KO), Francis Marshall (submission) and Ryan Gandra (KO). Gandra's early finish carried extra weight for our desk, and we will get to why in the autopsy.

The model autopsy

This is the part of the recap where we grade our own work in public, misses included. On Feb. 28, there were not many to own.

The model went 11 for 13 on the card, an accuracy of 85%. That sits above our canonical live mark, which matters less than what happened on the picks we actually stood behind.

We flag a subset of fights as higher-conviction calls, the LOCK and HIGH tier. On this card we flagged five, and all five hit. Two of those were LOCKs, our top tier:

Ryan Gandra at 81% landed. He won by first-round knockout, the kind of decisive result a LOCK is supposed to predict.

Santiago Luna at 84% landed as well.

No LOCK missed. That is the cleanest possible night for the flagged tier, and on a free product where we publish the grade either way, a 5-for-5 night is worth stating plainly rather than dressing up.

The headliner is the more interesting line on the card sheet. Our model picked Reyes to win at a 31% probability, which is the model telling you it expected him to lose more often than not while still siding with him. He won, by first-round KO, exactly the outcome the medium-confidence tag left room for. We mark that one correct. We do not mark it as a confident call, because it was not, and the honest version of a model autopsy distinguishes between a pick that was right and a pick that was sure. This one was right.

The two misses on the card belong to the ledger too. We do not have the specific fights broken out in the verified record, so we will not name them or invent a reason they went sideways. Two of 13 went against us. That is the number, and it stands without a story attached.

One note on what this card does not show. Method accuracy, whether we called the finish versus the decision, is not something we can grade here, so we are not citing a method-correct figure. The 85% is straight pick accuracy.

One honest takeaway

A 5-for-5 flagged night and an 85% card is a good night. It is also a single card, and single cards are exactly where a model can look smarter than it is.

The number that keeps us honest is the long run. Across 2026, our live record is 67.9% on all picks. On the LOCK and HIGH tier, the same flagged group that went perfect here, the canonical figure is 81.4% over 64 calls. Feb. 28 sits above both lines. It does not move them, and we are not going to blend one strong card into the season figure to make the season look better than it is.

The useful read is narrower. Nights like this are why we flag picks at all. The flagged tier is supposed to be where the model earns trust, and on this card it did, hitting all five while the medium-confidence headliner came in as a bonus the model had already half-conceded. The model behaved the way it is built to: confident where the evidence backs it, hedged where the fight is close.

Reyes provided the headline finish. The 5-for-5 flagged slate is the real story underneath it, and the two misses are the reminder that we publish the whole card, not just the highlights.