Forecast record
Every forecast we published, counted once and checked against the official result. Each kind of forecast gets its own yardstick: how far a projection missed, how well a probability was calibrated. There is no single accuracy number here on purpose. A yardage projection and a win chance answer different questions.
Void means the player did not play, or the game ended in a push or tie. Not measurable means the official result has no line for it. Neither ever counts as a miss. 0 forecasts were withdrawn before kickoff and stay listed, marked withdrawn.
NFL
| Forecast | Kind | Measured | How it did | Latest game | Details |
|---|---|---|---|---|---|
| Anytime touchdown | Probability | 767 | Brier 0.170 log loss 0.520 · forecast 23.6% vs happened 23.3% · ECE 0.034 | 2026-10-05 | Details → CSV |
| Receiving yards Pick record 131–148–1 | Projection | 657 | Avg miss 23.4 median miss 16.4 · bias -9.0 · 77.8% inside our 80% range | 2026-10-05 | Details → CSV |
| Receptions Pick record 136–140 | Projection | 657 | Avg miss 1.7 median miss 1.0 · bias -0.6 · 88.6% inside our 80% range | 2026-10-05 | Details → CSV |
| Rushing yards Pick record 80–65 | Projection | 346 | Avg miss 18.7 median miss 11.5 · bias -10.1 · 84.4% inside our 80% range | 2026-10-05 | Details → CSV |
| Team score | Projection | 212 | Avg miss 7.9 median miss 6.5 · bias +0.3 · 78.3% inside our 80% range | 2026-10-05 | Details → CSV |
| Passing yards | Projection | 120 | Avg miss 70.7 median miss 56.7 · bias -28.0 · 81.7% inside our 80% range | 2026-10-05 | Details → CSV |
| Winning margin | Projection | 106 | Avg miss 10.9 median miss 8.5 · bias +0.7 · 75.5% inside our 80% range | 2026-10-05 | Details → CSV |
| Game total points | Projection | 106 | Avg miss 10.0 median miss 8.0 · bias +0.6 · 85.9% inside our 80% range | 2026-10-05 | Details → CSV |
| Game winner Pick record 61–43 | Probability | 104 | Brier 0.236 log loss 0.664 · forecast 51.9% vs happened 52.9% · ECE 0.049 | 2026-10-05 | Details → CSV |
MLB
| Forecast | Kind | Measured | How it did | Latest game | Details |
|---|---|---|---|---|---|
| Moneyline pick Pick record 411–390 | Probability | 801 | Brier 0.253 log loss 0.700 · forecast 57.2% vs happened 51.3% · ECE 0.059 | 2026-10-05 | Details → CSV |
| Run line pick Pick record 504–297 | Probability | 801 | Brier 0.234 log loss 0.661 · forecast 59.9% vs happened 62.9% · ECE 0.032 | 2026-10-05 | Details → CSV |
| Total runs pick Pick record 374–386–36 | Probability | 760 | Brier 0.266 log loss 0.728 · forecast 58.7% vs happened 49.2% · ECE 0.095 | 2026-10-05 | Details → CSV |
| Home run (Homer Nukes) | Probability | 196 | Brier 0.150 log loss 0.479 · forecast 23.9% vs happened 17.3% · ECE 0.069 | — | Details → CSV |
Premier League
| Forecast | Kind | Measured | How it did | Latest game | Details |
|---|---|---|---|---|---|
| Anytime goalscorer | Probability | 1,212 | Brier 0.068 log loss 0.249 · forecast 8.6% vs happened 8.1% · ECE 0.015 | 2026-09-20 | Details → CSV |
| 1+ shot on target | Probability | 1,212 | Brier 0.153 log loss 0.470 · forecast 24.1% vs happened 23.7% · ECE 0.024 | 2026-09-20 | Details → CSV |
| Match result (1X2) | Probabilities | 46 | Log loss 1.052 Brier 0.644 · likeliest outcome happened 43.5% · a blind guess scores 1.099 | 2026-09-20 | Details → CSV |
| Over 2.5 goals | Probability | 46 | Brier 0.263 log loss 0.747 · forecast 52.3% vs happened 50.0% · ECE 0.192 | 2026-09-20 | Details → CSV |
Ligue 1
UFC
Published but not in this record yet
- NFL · player props (before 2026-09-20) — unresolved names only: Week 1 and Week 2 Thursday props come from the week reconciliation through an exact roster crosswalk (latest capture before kickoff, unique full name on the team). A row whose name matches no one, or several people, on that roster is left out rather than guessed.
- NFL · nfl_score_shape: Published derived distribution (key numbers, OT/tie) with no settlement owner.
- MLB · projected score / simulation-median total: Published on game pages; no owner grades it. UNMEASURED.
- MLB · player-prop leans: Every MLB prop market is DEMOTED to market context (RESEARCH), so it is not public forecast history.
- EPL · BTTS / clean sheet / double chance / scorelines: Published as derived markets; no owner grades them. UNMEASURED.
- UFC · method / round: Published heads with no forward grader. UNMEASURED.
- NBA · game model (v0 / v0.1): SHADOW / PRIVATE_RESEARCH — never public forecast history.
How this record works
One row per forecast: the last version we published before the game started, read from the record of that forecast and never re-predicted. A forecast that appeared on several pages (a game page, a Top-5 board, Ask) is still one row. Projections are measured by how far they missed. Probabilities are measured by Brier score and log loss, where lower is better, plus calibration: when we said 30%, did it happen about 30% of the time? A pick record appears only where we published a pick. Sportsbook lines are shown for context and are never part of these numbers.
Products are graded separately, card by card: Suggested parlays and the paper bankroll. A four-leg card is one product result here and four forecasts in this record.