Every pick we flag is a pick we have to answer for, so start with the three that mattered most: at UFC: Kim vs Fan on Jan. 31, our model put three picks in the LOCK+HIGH bucket and went 3 for 3. On the same card it also flagged Tallison Teixeira as a winner at a 19% model probability, and Teixeira delivered. That is the kind of low-number call that either looks sharp or looks foolish by the end of the night.

The night in brief

Dom Mar Fan beat Sangwook Kim by decision, the bout going the full three rounds before the judges turned in the result. There was no finish here, and the model did not read it as a runaway either, which matters for the autopsy below.

The card produced at least one fast finish. Quillan Salkilld closed out his fight by submission in the first round, the sort of early result that rewards a model willing to back a finisher and punishes one that hedges. Beyond the headliner and that opener, the verified ledger for this card is thin, so we are not going to dress up fights we cannot confirm.

The model autopsy

The headline grade: 10 correct out of 13, 77% on the card. That is a clean night by any reasonable bar, and the flagged picks carried it. The three LOCK+HIGH selections, the ones we tell readers to weight most heavily, all hit. When the model flags a pick as confident, this is the result it is supposed to produce.

Own the headliner read honestly, because it is the most interesting line in the grade. The model picked Dom Mar Fan, and Dom Mar Fan won. Correct call. But it was a LOW-confidence pick at a 40% model probability, which is the model effectively saying it had no strong lean and slightly favored the other man on paper. So this is a win, not a brag. A 40% pick landing on the right side is a coin that came up heads, and we would be lying to you if we framed a near-tossup as a confident headliner read. It went in the win column. It does not go in the "we saw it coming" column.

The Teixeira call is the more genuine signal from this card. At a 19% probability the model was not predicting a Teixeira win in any confident sense, yet it flagged him and he came through. One result is not a trend, and we will not pretend a single low-probability hit proves anything about edge. It is logged, it counts, and it is the kind of outcome we track over many cards rather than celebrate on one.

The misses are baked into the 10-of-13: three picks went the wrong way. The verified facts for this card do not itemize which three fell, and we are not going to invent the losers to make the autopsy read cleaner. What we can say plainly is that none of the three misses came from the flagged bucket. The model's confident picks were perfect; the damage was in the lower-conviction calls, which is where you want your errors to live if you are going to have them.

One honest takeaway

A 77% card with a clean flagged sweep is a strong night, and a strong night is exactly the kind of result that tempts a desk to overclaim. We will not. Our canonical 2026 live record sits at 67.9% across all picks, and 81.4% on the LOCK+HIGH tier where this card's 3-for-3 belongs. This event ran ahead of both lines. One card running hot does not move the season number, and one card running cold would not break it either. The reason the flagged tier is the one we point readers to is on display here: the model's confident picks held, its near-tossup headliner happened to land, and its three misses stayed in the bucket where misses are survivable.

No betting advice here, and none implied by a single good night. The point of grading ourselves in public is that you see the 77% and the 40% headliner read in the same breath, and you decide what that is worth.