Our edge: KC 42 · BUF 58, favored Buffalo. Vegas: KC −2.5, favored Kansas City. Bills won outright as home dogs — we caught the upset Vegas didn't.

Missed It

ARI vs DAL

27–17

Our edge: ARI 63 · DAL 77, favored Dallas. Vegas: DAL −3.5, agreed. Arizona won outright on the road — both models missed this one.

Called It

SF vs NYG

34–24

Our edge: SF 63 · NYG 58, favored 49ers. Vegas: SF −2.5, agreed. SF covered and won as called.

Design notes: teaser cards intentionally show only kickoff time, a one-line edge lean, and a single "hook" stat — enough to signal there's a real mismatch worth knowing about, without giving away the full breakdown. Full four-stat detail lives one click away.

On the Vegas line: the "Vegas" number is a real closing spread, pulled free from the same nflverse dataset already powering the schedules and head-to-head history — no paid odds API needed. It's shown next to our own edge score so a visitor can see where we agree with the market and where we don't. Odds get their own green badge on purpose, separate from the blue "Full Breakdown" CTA — this is the module where a real sportsbook affiliate link would eventually live, since a visitor looking at a spread is a warmer lead than one just reading stats. That's also the part of the site that would need 21+ gating and a responsible-gambling disclosure once it's live money, not a demo.

On the grading badge: "Called It" just means the higher-edge team won the game outright — a low bar, since the pre-game favorite wins most NFL games anyway. All three cards above are real graded results, not examples. A production version should grade against the closing spread (covered/didn't cover) instead of a straight win/loss, and would need a full season of graded calls before "accuracy" means anything to a visitor.