Our model does not say "lock" often. At UFC 328 on May 9 it said it about Khamzat Chimaev, an 84 percent call and the most confident pick on the card. Sean Strickland spent 25 minutes proving it wrong, out-working Chimaev to a decision in the main event and handing our highest-conviction tier its loudest loss of the spring. This is the result this column exists to own.

The night in brief

Sean Strickland beat Khamzat Chimaev by decision after five rounds in the UFC 328 main event, a result our own system did not see coming.

The undercard offered a cleaner read than the headliner. Ateba Gautier and Baisangur Susurkaev both won, two of the night's surer outcomes that landed as projected. Those wins did most of the work in keeping the card grade respectable.

But the story of the evening sat at the top of the marquee, where Strickland carried the fight to a decision over 25 minutes against a fighter the model favored heavily.

The model autopsy

Our system went 9 for 12 on UFC 328, a 75% hit rate. That is a serviceable night by most standards and below our season line, and the gap traces almost entirely to one fight.

On flagged picks, the LOCK and HIGH calls we stake the most confidence on, the model went 4 for 5. Gautier (87%) and Susurkaev (81%) both cashed. The miss in that group was the one that stings, because it was also the night's centerpiece.

The model picked Chimaev to beat Strickland and tagged it a LOCK at 84%. Strickland won the decision. We were wrong. A LOCK is the system saying it has seen enough signal to bet its reputation, and an 84% probability on the wrong fighter is a read that missed, not a coin flip that broke badly. The headliner was our most confident call on the card and our most public error.

There is no spin available here. When a model fails on the fight everyone is watching, that failure is the fight everyone remembers. The other 11 picks going 9 for 11 is the kind of context that sounds like an excuse, so we will state it plainly instead: the system nailed the supporting cast and whiffed on the headline.

What went wrong is harder to answer with the facts in hand, and we are not going to invent a clean narrative for it. Strickland's decision win over a fighter the model favored that heavily suggests the system overweighted a projected early advantage and underweighted a five-round grind. That is a hypothesis, not a verified post-mortem, and we will treat it as one.

One honest takeaway

A single missed LOCK does not move the season. Across 2026 the model sits at 67.9% on all picks and 81.4% on its LOCK and HIGH calls over 70 graded fights. UFC 328's 4-of-5 flagged record folds into that LOCK-and-HIGH line without flattering it or breaking it.

The honest point is that 81.4% on high-confidence picks means roughly one in six of them still loses, and on this card the one that lost was the main event. That is the cost of publishing a confidence tier instead of hiding behind a flat average. We told you Chimaev was a LOCK before the fight, so we are telling you now that the LOCK lost.

We grade ourselves in public for exactly this reason. The 9-for-12 night and the 81.4% season figure are both true, and neither one excuses the other. UFC 328 was a good card for the model on volume and a bad one on the fight that mattered most, and both of those things can sit in the same recap without being blended into a single comfortable number.