The most useful number from UFC Fight Night: Petrino vs. Asplund on March 14 sits well below the headliner. It is an 18% prediction that came in. Our model had Kevin Vallejos at 18% to win and flagged him as a lock anyway; he scored a first-round knockout. That single call did more to test the system than anything at the top of the card.\n\n### The night in brief\n\nVitor Petrino beat Steven Asplund by decision in the third round to take the main event. The fight went the distance, which is the cleaner read for a model that leaned Petrino at 64% with medium confidence. No drama in the result, no asterisk on the grade.\n\nThe finishes came earlier on the card. Bia Mesquita closed out a first-round submission. Ion Cutelaba forced a tap in the first as well. Vallejos landed the first-round knockout already mentioned, the only one of the three that arrived as a striking finish. Three fighters did not need the judges, and two of them did it on the mat inside five minutes.\n\n### The model autopsy\n\nThe card grade: 12 correct out of 14, or 86%. That is a strong night by any measure, and it holds up under the part that matters more than the raw count, which is how the flagged picks did.\n\nWe separate every slate into picks we are willing to stand behind (LOCK and HIGH confidence) and the rest. On this card the flagged group went 4 for 5. Four of the five calls we put real weight on came through. One did not. We are not going to bury which one, because the whole point of publishing a grade is owning the miss as plainly as the hits.\n\nThe Vallejos call is the one to sit with, and it cuts in our favor in a way that deserves scrutiny rather than a victory lap. The model assigned him an 18% chance to win and still flagged him. He won by first-round knockout. A flagged pick on an 18% line is the kind of result that looks brilliant in hindsight and should make a forecaster nervous. It should not make one proud. Either the flag logic caught something real that the win probability undersold, or the probability was simply wrong and the flag got lucky. One card cannot tell you which. We log it and watch whether the pattern repeats, and we do not pretend a single right answer settles the question.\n\nThe headliner is the easier honesty test, and it passed. The model picked Petrino at 64%, medium confidence, and Petrino won. No miss to own there. When the main event goes the way the number pointed, we say so without inflating a 64% read into a sure thing it never was.\n\nSo the two misses on the card sit outside the headliner and outside most of the flagged group. The flagged-pick line reads 4 for 5; the overall line reads 12 for 14. Both are good. Neither is perfect, and the gap between those two records is where the real information lives.\n\n### One honest takeaway\n\nA single 86% card is a data point, not a trend, and it would be dishonest to wave it around as proof of anything. Our canonical live record across 2026 sits at 67.9% on all picks. On the flagged tier, the LOCK and HIGH calls we actually stand behind, it sits at 81.4% across 70 graded selections. Those are the numbers that describe what the model does over time.\n\nThis card ran ahead of both lines. That happens, the same way a sub-70% card happens, and we publish those too. The honest frame is that one strong night neither raises nor lowers the standing record; it just adds to it. The Vallejos hit is the kind of outlier that flatters a single-card grade and tells you almost nothing about the next one.\n\nWe grade in public because the average is the product, not the highlight. March 14 was a clean night with a correctly called main event, a flagged group that went 4 for 5, and one finish that should keep us skeptical of our own win probabilities. That last part is the one worth carrying forward.