🏈 NFL predictions.

Model 1 · current-season trend + opponent: completed games weighted 1, 2, …, N (newer games count more), followed by the saved team-defense adjustment.

Model 2 · 80/20 opponent history: 80% current-season per-game average + 20% per-game average in up to ten verified meetings against the opponent from this or last regular season. If opponent history is missing, the current-season-only fallback is clearly labeled. One verified current-season game is enough for this experimental estimate; sample sizes are displayed. No legacy workload, minimum-edge or snap-stability gates apply to this replacement.

Expected winner and spread are sourced market context. They do not automatically mean Over/Under, and no untrained numerical game-script multiplier is applied. Model 1 retains its existing current-season history and matchup rules. The models run independently.

Both models lock their own pregame side, line and inputs. Results are graded against each saved line, with separate hits, misses and decided counts. Goblins are excluded.