Five times on this card the model told readers it had separation, flagging a pick at LOCK or HIGH confidence. All five landed. The lone LOCK, Umar Nurmagomedov at 83%, won, and the headliner went the model's way too.
Waldo Cortes Acosta finished Derrick Lewis by knockout in the second round to headline UFC Fight Night on Jan. 24. Acosta walked in as the model's pick, a HIGH-confidence call at 72%, and the result followed the read.
The undercard supplied the violence early. Three fights ended inside the opening round by knockout: Josh Hokit, Alex Perez and Ty Miller each closed the show in the first. First-round finishes tend to scramble any night's predictions because they leave no time for the better fighter to assert control, but here the early stoppages did not dent the model's stronger calls.
The model autopsy
The card graded 8 of 11, or 73%. That is the headline number, and it sits a touch above where our long-run accuracy lives, so no complaints on the top line.
The number that matters more is the flagged tier. The model carried five picks at LOCK or HIGH confidence on this card, and all five hit. Five for five. When we flag a pick, we are telling readers we have separation, and on this night the separation held across the board.
The single LOCK was Umar Nurmagomedov at 83%. He won. That is the cleanest kind of result for us: high stated confidence, correct outcome, no asterisk.
The headliner was a HIGH call rather than a LOCK. The model picked Acosta at 72% over Lewis, and Acosta won by second-round knockout. Correct, and on the firmer end of the confidence range.
No LOCK miss to own here, and no headliner miss. We publish these autopsies whether the night runs clean or ugly, so it is worth being plain that this one ran clean. The three misses among the 11 picks came outside the flagged tier, where the model is openly telling readers it has less conviction. That is the system working as designed: the picks we stood behind hardest were the ones that landed, and the slips came on the fights we never marked as separations in the first place.
One note on method. We are not citing a method-correct rate for this card because that figure is not available in the verified record, and we would rather omit a number than reconstruct one. The winner, the knockout finish and the second-round timing on the headliner are confirmed; the round-by-round method accuracy across the full slate is not, so it stays out.
One honest takeaway
A 5-for-5 flagged night is a good night. It is not the typical one. Our canonical live 2026 record sits at 67.9% across all picks and 81.4% on the LOCK and HIGH tier across 70 such calls. That second number is the one we point to, because it reflects the picks we actually flag for readers, and it is where the model earns its keep.
This Jan. 24 card ran above both marks. That is what a clean night looks like, and it is also why we do not blend single-card results into the running record or let a perfect flagged slate stand in for the season figure. One card of five flagged hits does not move 81.4%, and it should not. The plain version of the story is the model called its strong fights correctly here, the long-run flagged number is 81.4%, and the gap between a great night and the average night is exactly why we keep grading every card the same way.
No bets, no projections off this. Just the result, the grade and where it fits.