Our model went into the main event of UFC Fight Night: Leavitt vs. del Valle on Feb. 21 with a contrarian read, and the result went against it.\n\n### The night in brief\n\nJordan Leavitt beat Yadier del Valle by decision over three rounds in the headliner. No finish, no late drama in the result line, just 15 minutes that went Leavitt's way on the cards.\n\nThe undercard supplied the violence. Three fights ended in the first round by knockout: Uros Medic, Melquizael Costa and Jacobe Smith each closed the show before the opening five minutes were up. For a card that finished with a decision on top, the early stoppages did the heavy lifting on the highlight reel.\n\n### The model autopsy\n\nWe grade our own predictions in public, and this card was a middling night that got dragged down by where the miss landed.\n\nThe model went 9-for-14 on the full card, 64 percent. That is below our season line and below where we want to be, but the single number that matters most here is the one at the top of the marquee.\n\nOn the flagged picks, the calls we tag LOCK or HIGH because the model is most confident, the card went 4-for-6. Two of those higher-conviction picks missed. One of them was the main event.\n\nThe model picked Yadier del Valle to beat Jordan Leavitt. It was wrong. Worse, it was a HIGH-confidence flag, and the model had Leavitt at just 27 percent. That is not a coin flip we lost. The model backed the less likely outcome, and the more likely outcome happened on the cards. When you assign a fighter a 27 percent chance and he wins a three-round decision, the honest reading is that the model misjudged the matchup, not that variance bit us.\n\nWe log it as a wrong HIGH pick on the headliner, the most expensive kind of miss we track, because it is the fight most readers actually watched. Owning it is the entire point of doing this in the open.\n\nThe other side of the 4-for-6 flagged record is that the model still cleared two-thirds of its confident calls, and the broader 9-of-14 means the back half of the card carried more weight than the top. The first-round knockouts from Medic, Costa and Smith were finishes the model does not need to call perfectly to score; what it needed, and did not get, was the main event.\n\n### One honest takeaway\n\nA 64 percent card with a busted HIGH pick on top is a reminder of what our headline numbers do and do not promise.\n\nAcross live 2026 picks, the model sits at 67.5 percent on all picks and 82.8 percent on its LOCK and HIGH flagged calls over 64 graded predictions. The leak-free backtest lands near 70 percent overall and 84.3 percent on LOCK. Those are the canon numbers, and this card sits under both of them.\n\nThat is how a confidence band works. It is a rate, not a guarantee. A flagged record of roughly 83 percent means roughly one in six confident calls misses, and on Feb. 21 one of those misses was the most visible fight on the card. The Feb. 21 result does not move those season numbers on its own, and we are not blending it into a rosier figure to make the night look better than it was.\n\nIf the flagged hit rate is going to mean anything, the misses have to be reported with the same ink as the hits. This one was a 27 percent underdog pick on the headliner that the favorite settled over three rounds. We called it, we got it wrong, and the grade stands.