The numbers, worn by the Bills

The Bills went 5-1, but a record is a result, not an explanation. The explanation starts with EPA/play: Buffalo-style, every snap either moved the expected score or it didn’t. The Bills offense averaged +0.210 — 1st of 32, meaning a typical snap gained about 0.21 points of expected score. Their defense allowed +0.148 per play (31st — remember, lower is better on defense).

Success rate tells you the texture: 49.2% of snaps kept them on schedule against a league average near 44%. Pair that with pace and you get identity: at 29.0 seconds a snap they ran the 16th-fastest offense of 32 — middle of the league in tempo.

Raw EPA flatters teams with soft schedules, so the net rating re-solves it against opponents: +0.065est.. Multiply by the fitted 15.1 points per unit and the Bills are worth about +1.0 points against a league-average team on a neutral field — that number, plus era-fitted home field and the QB adjustment, is the whole model margin on every game card.

Playoff odds come from 10,000 simulated seasons — full scores, because tiebreakers need points. The Bills made the field in 84% of them, won the division in 65%.

Settle every game against its closing spread and you get the ATS record: 4-2, beating their number by +0.5 points per game on average. As favorites they went 3-2; as dogs 1-0. Their games went 5-1 to the over. Descriptive, not predictive — 52.38% is break-even at standard vig, and last season’s cover rate does not persist.

A concrete game: Chargers at Bills, week 3, closed at BUF -7. That line sits on a key number, so the model prices a 4.9%est. push — and ½-point at risk of 4.9%, which is why buying the half point matters here and not at 5½.

Where the model margin and the market disagree, remember the disagreement entry: model-preferred sides have gone 49.93% ATS. The market is the sharper instrument; the model is the explainer.