Trained on 2021–23 → tested on 2024, trained on 2021–24 → tested on 2025, averaged. Nobody beats ~64% straight-up and nobody beats the spread (~50%) — the market is efficient. The fancy ensembles don't beat simple regression on 1,359 games.

Expected wins from points scored/allowed (exponent 2.37). Positive luck = winning more than the scoreboard says they should.

Steam = 2+ pt spread move or 5%+ ML implied-prob shift since the Sunday-night open. Edge = consensus margin vs the Vegas line by 3+ pts. No public ticket/handle splits exist, so true "sharp vs public" labels aren't possible.

Wind 15+ mph cut scoring by ~4.9 pts (n=88); domes averaged 3.5 pts more than outdoor games. Cold alone showed no clear effect. Shown as context — the market already prices most of this.

Player status is covered live in the 🏥 Injury Report (every snap of the depth chart). Player tracking — Next Gen Stats speed, separation, cushion — has no free public feed; the closest free proxies (RYOE/att, time to throw, blitz rate, QB-fault sacks) are already in this app's advanced metrics.

Picks are the unadjusted ⚡️ model, locked at the first data refresh inside 45 minutes of kickoff — using the freshest injury report, depth charts and lines available at that moment. Never rewritten afterward. SU = picked the winner; ATS = the model's lean covered the spread. Pushes excluded from W-L. Tracking began Week 5, 2026.

Grades: A+ = dominant win, A = convincing, B+ = solid, B = close/ugly win, C = lucky win; F = blowout loss, D = bad loss, C = loss, C+ = close loss playing well, B- = "good" loss. Closing lines tracked from Week 5 (hourly snapshots); Weeks 1-4 predate tracking.

ML units = profit at the logged moneyline price, 1 unit staked per play (ML wins mirror SU; units reward underdog prices). vs ⚡️ model = head-to-head record on graded games where the capper and the model picked different sides (agreements in parentheses). Only plays published publicly at least 5 minutes before kickoff are logged, with the source noted on each pick — blank means no public play was found. Tracking began Week 5, 2026.

Hover/tap a pick for its source. The ⚡️ column is the model's pick for that game.

Picks save on this device only. To join the Cappers leaderboard under the same rules as everyone else (5-minute lock, graded SU/ATS/ML), tap 📋 Copy my picks and paste the text to Torque in chat before kickoff — he'll log them officially. Picks for games already kicked off are locked out here.

Tracking line movement from the first reading each week.

Rating ≈ expected margin vs an average team on neutral ground. A matchup's projected margin ≈ home rating − away rating + home field. See Methodology.

Morning official NFL report plus a live ESPN feed pulled on every open — Out, Doubtful, Questionable, IR. Organized by matchup: each injury shows the next man up from the depth chart and a 🎯 read on how it affects that specific game. = fresh from the live feed.