New to these reports? Start here
- Dotted-underlined words have a plain-English definition — hover or tap them. Every term is also on the glossary page.
- "Null" means we found nothing, not that something broke. Most reports here are negative results, on purpose — knowing an idea doesn't work is the point.
- Two questions get asked separately. First, is the effect real? Second, is it already priced into the betting odds? An effect can be completely real and still useless to bet on.
- A "calibration" row is a self-check. It runs the same method on something already known to be true. If that fails, the whole report is unreliable — so it's reported alongside the findings.
- If a confidence interval includes zero, the real effect might be nothing at all, so no claim gets made.
How predictable are Super Bowl winners and final fours? (2026-10-22)
sim.py: at weeks 1, 6, 12 and the end of the regular season, fit team
ratings from the closing spreads seen so far, simulate the rest of the season
and the playoffs 20,000 times, and grade the title and final-four odds against
what happened, 2016-2025 (10 seasons).
The ratings are the spread market's view of each team, so this measures how predictable titles are given the market's game-by-game view, not a model of our own. Settings (recency half-life 2 weeks, light shrinkage) were picked on how well week-K ratings predict week K+1 spreads, using no outcomes. The rating drift from week K to season end is estimated leaving the graded season out. The last regular-season week's lines are skipped (rested starters).
Results
| Checkpoint | Favorite's avg title odds | Favorite won | Champion's avg odds | Champion's rank (10 seasons) | Final-four teams in the top 4 |
|---|---|---|---|---|---|
| After week 1 | 10% | 0 of 10 | 4% | 25, 6, 6, 3, 21, 3, 12, 13, 6, 26 | 11 of 40 |
| After week 6 | 20% | 0 of 10 | 8% | 2, 7, 2, 2, 9, 4, 4, 2, 6, 13 | 18 of 40 |
| After week 12 | 23% | 2 of 10 | 15% | 1, 2, 2, 9, 6, 4, 1, 4, 5, 10 | 21 of 40 |
| Playoffs start | 25% | 3 of 10 | 16% | 1, 2, 3, 5, 3, 7, 1, 5, 5, 1 | 25 of 40 |
Uniform guessing gives each team 3.1% to win and 12.5% to make the final four.
- The ceiling is low. Even at playoff time the best team is only about a 1-in-4 title shot; single elimination is mostly noise between good teams.
- Week 6 already sorts contenders. The champion was in the top 2 in 4 of 10 seasons and in the top 6 in 8 of 10. Only 2018 NE (9th) and 2025 SEA (13th) came from further back.
- Calibrated. Weeks 6-18 pooled, final four: predicted 9% / 22% / 38% / 61% in four bins, happened 8% / 21% / 40% / 66%. Title: teams given 15-50% won 20% (predicted 20%); teams under 3% won 1 of 449.
- Week 1 is handicapped: one game of lines per team. A preseason checkpoint needs futures or win totals, which is what the historical futures purchase is for.
2026 after week 6 (projection_2026_wk6.csv) vs the futures market
Sim: BUF 12.3%, KC 10.5%, SF 10.2%, JAX 7.6%, DET 7.0%, SEA 6.7%. Pinnacle futures 2026-10-22 (proportional no-vig): SF 13.6%, KC 9.9%, BUF 9.8%, BAL 7.2%, LA 6.2%, JAX 6.0%. The spread market rates BUF best (+4.4); the futures market prefers SF. The sim is crude (no injuries, simple tiebreaks), so a gap like this is not a bet.
Next
- Futures captured daily from 2026-10-22 (
nfl_futures_snapshots, 1 credit). - After the Nov 1 reset: buy historical Super Bowl futures (Odds API historical works, probed 2021-10-20, 10 credits a snapshot) at the same four checkpoints for 2020-2025, about 24 snapshots / 240 credits, and grade futures against this sim and against outcomes.
Futures vs our simulation, 2020–25 (2026-11-02)
24 historical Super Bowl futures snapshots (8 books, 240 credits) at week 1 / 6 / 12 / playoff
start, devigged per book and median across books. Script grade_futures.py → results_futures.json.
| Point | Champion's probability: market / sim | Log losslog lossA score for probability forecasts that punishes confident wrong answers harshly. Lower is better.: market / sim | Brier (32 teams): market / sim |
|---|---|---|---|
| Week 1 | 6.3% / 3.7% | 2.94 / 3.50 | .934 / .972 |
| Week 6 | 7.4% / 7.7% | 2.80 / 2.76 | .928 / .939 |
| Week 12 | 10.2% / 12.1% | 2.34 / 2.28 | .887 / .870 |
| Playoffs | 12.1% / 13.7% | 2.19 / 2.14 | .887 / .855 |
| All 24 | 2.57 / 2.67 | .9092 / .9091 |
Reading: - The market is clearly better before the season, when our ratings start from scratch. - From week 6 the simulation is level with it, sometimes a hair better. That's 6 seasons, so it's noise-sized. - Every book carries ~20% margin on the futures board (Pinnacle 19.8%), so even a model level with the market loses about 20% betting futures. - Champions' market rank at playoff start: 5, 1, 1, 5, 5, 1 (2020–25). - Use futures as a benchmark for the title-odds page, not as a bet.