How good are these forecasts?
Every forecast scored on seasons the model never saw while its settings were chosen.
Held-back seasons
| Forecast | Forecasts | Log loss | Favourite won |
|---|---|---|---|
| The model | 706 | 0.620 | 66% |
| Base rate | 706 | 0.696 | 53% |
| Recent form | 706 | 0.641 | 62% |
| Betting market | 706 | 0.603 | 68% |
Against the betting markets
Each market's price before the start, scored on the same finished games as the model. A negative figure means the market was more accurate.
| Market | Since | Priced | Model log loss | Market log loss | Model vs market |
|---|---|---|---|---|---|
| Bookmakers (Pinnacle, DraftKings, bet365) | 25 Jan 2026 | 139 | 0.624 | 0.620 | −0.6% |
| Polymarket | 15 Jan 2024 | 707 | 0.620 | 0.603 | −2.8% |
| Kalshi | 4 Sep 2025 | 417 | 0.628 | 0.614 | −2.3% |
The playoff odds seed each conference the NFL's way: four division winners by record, then three wild cards; the top seed has a bye and faces the lowest seed left.
Ties in the simulated table break on point differential. The NFL's full tiebreaker list (head to head, division record, common games) is not applied yet.
EPA per play is shown beside the rating in the Teams explorer. It does not move the forecast; it is there so you can see which teams' results run ahead of their play.
Only the quarterback is adjusted for; other injuries move the market straight away and the model only after the results come in.
Travel and rest are in the file but not in the model. Games under a dome or a closed roof get no weather.
Questions and answers
How accurate are PodiumOracle's NFL predictions?
On 706 games from held-back seasons (2024 to 2026), which the model never saw while it was tuned, the side it favoured won 66% of the time. Its log loss was 0.620, against 0.696 for the base rate and 0.641 for recent form, and 0.603 for the closing betting odds. Against the betting markets on the same finished games, its log loss was 0.624 against the bookmakers' 0.620 over 139 games, 0.620 against Polymarket's 0.603 over 707 games and 0.628 against Kalshi's 0.614 over 417 games (lower is better).