Yards and TDs allowed per game to each position, this season. Rank 1 = gives up the most (best matchup). Slate defenses first.
How a prop gets graded
- Hit rates. Every 2025 and 2026 game for the player is checked against tonight's line: last 5, last 10, this season, and head-to-head versus this opponent. Streak counts consecutive overs (O) or unders (U).
- Matchup. Defense rank compares what the opponent allows per game to that position against the league average this season.
- Stats probability. A recency-weighted average (this season counts most), nudged by the matchup and by the team's Vegas implied total, then turned into an over probability that accounts for yardage being lopsided. For TD props, the last-10 hit rate is blended in 30%.
- Market anchor. The book's two-sided odds are stripped of vig to get a fair price. The final probability moves only 30% of the way from that fair price toward the stats probability. Books already know about injuries and role changes that box scores miss.
- EV and grade. EV is the expected profit per $1 at the posted odds. Strong: EV ≥ 4%, hit rate agrees, 3+ games this season. Lean: EV ≥ 2%. Check: stats and market disagree by more than 22 points, which almost always means a role change. Read the news before betting those.
- Stake. Half-Kelly shown as a percent of bankroll. Treat it as a ceiling, not a target.
Backtest
Walk-forward: every prediction used only games played before that week. Settings were tuned on the 2025 season, then checked on 2026 Weeks 1–5, which the tuner never saw. Free historical prop lines don't exist, so each line is a stand-in: the player's median over his last 8 games. That makes these numbers a test of the stats model, not a real-money record.
Skill = how much lower the error is than a coin-flip baseline (for TDs, the player's season-to-date TD rate). Above 0 means the model adds information. Hit rate counts plays where the model was at least 56/44.
Changes the backtest made
- Yardage is now treated as lopsided. Big games pull the average up, so the old model overrated overs: its "55%" receiving overs hit only 47%.
- Vegas implied team totals now feed every projection. Players on teams expected to score more get more TD and yardage upside.
- Recency weighting was softened, matchup weight cut in half, and hit rates dropped from yardage probabilities (they added noise). Hit rates still show on every card as context.
- TD props are capped at Lean. The TD model only tied a naive baseline on 2026 games.
- Volume markets added and backtested the same way (skill on 2026 games): receptions +5.6%, rush attempts +8.9%, pass attempts +6.9%, completions +5.4%, pass TDs +19.0%, INTs +15.8%, rush + rec yards +10.7%. Low-count markets like pass TDs and INTs look strongest partly because the stand-in line is crude for them; expect less against real books.
- Kickers: field goals made scored +5.0% skill on 2026 games (62% on its stronger plays), so it grades normally. Kicking points (3 per FG + 1 per extra point) scored −0.8%, so it is capped at Lean and marked track-only.
Limits
One-way markets (anytime TD) have no under price, so their fair price assumes an 8% hold. Lines come from one book per refresh; shop Novig, Fliff and Hard Rock before placing. Small 2026 samples mean wide error bars on everything here.