GameScope NFL research

Team ratings (Elo): ARI 1406.6 vs DAL 1474.8. 393 games trained (Oct 30).

Model: nfl-elo-v2 — Elo ratings trained on real ESPN results (2025 regular+postseason and 2026 regular season to date), chronological replay, K=20, home advantage = 65 Elo rating points, applied only at true home venues: games the feed flags neutral-site get no home bonus, in training and in prediction. Totals baseline is the mean total points of the training games (measured, not assumed). Score-model spread/total probabilities use the engine's bivariate-normal assumptions (score std 13.5, correlation 0.3) — documented approximations, not measured NFL parameters. No backtest against closing lines exists; treat all outputs as experimental uncertain forecasts.

What the win probability uses: Elo ratings only, plus a home-field adjustment of 65 Elo rating points at true home venues (no adjustment at feed-flagged neutral sites).

Contextual research (displayed, not in the calculation): Contextual research only — form, rest, injuries, head-to-head, weather, and travel do not change the win probability above.

Totals: Expected scores are centered on a shared league-wide totals baseline of 46.2 points (the mean total of the training games) — a baseline, not a matchup-specific forecast.

Score-distribution assumptions: Spread/total probabilities assume team scores are bivariate normal (score std 13.5, correlation 0.3) — documented approximations, not measured NFL parameters.

Validation status: No backtest against closing lines exists. Treat all outputs as experimental uncertain forecasts.

Data used

• all core data present