Every possible 2026 season

Each game starts as a predicted margin: Tennessee's rating minus the opponent's, plus 2.5 points for playing at home or minus 2.5 for playing away. Both ratings are opponent-adjusted — they already account for who each team played.

Turning a margin into a win probability needs to know what a point of margin is actually worth. That relationship isn't assumed here. Nothing has been measured for this season yet, so a documented default of 13.0 points is standing in, and this note is here rather than a silent substitution.

That fitted number replaced an earlier one, and the change matters. The model used to convert margins using its average margin error, which is a different question and made it far too timid about heavy favorites — it said 85% for games favorites won 98% of the time. Measuring the relationship against outcomes instead cut the average calibration miss roughly in half on held-out games. The before and after is published.

The season is then rolled forward exactly. There is no random simulation and no sample noise: every reachable record gets its true probability, which is why the same inputs always give the same answer and why the whole thing recomputes the instant you change something.

Opponents outside the ratings pool — Furman, and any other FCS team — are given the engine's standard non-FBS rating rather than being guessed at individually.