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.
Schedule-based NFL ideas: luck regression, final weeks, travel, totals teasers (2026-10-31)
All four nullnull resultA test that found nothing. "Null" is the starting assumption that there is no real effect; a "null result" means the data gave us no reason to abandon that assumption. It does not mean the data was missing or the test failed to run.. Free nflverse closing lines, 2006–2025 regular season (5,199 games). Tests were fixed in
analyze.py before results were seen. Bets at closing lines, −110 for sides (break-even 52.38%), pushes
dropped, SEs clustered by season-week. Full output: results.json.
#3 Early-season luck regression (weeks 4–9)
Idea: teams whose record beats their underlying play are overrated. Bet the less lucky side when the luck gap is ≥ 1 SD.
| Luck measure | Bets | Less-lucky side covered | 2006–15 | 2016–25 |
|---|---|---|---|---|
| Wins vs Pythagorean | 480 | 47.5% ± 2.3 | 47.3% | 47.6% |
| One-score record | 479 | 50.1% ± 2.4 | 48.2% | 51.7% |
| Fumble recoveries | 462 | 51.7% ± 2.5 | 51.8% | 51.7% |
None beats 52.4%. The slope of ATS margin on the luck gap is within 1 SE of zero for every measure, in weeks 4–9 and in weeks 10–18. The market already prices luck.
#6 Final regular-season week
| Rule | Bets | Won | Note |
|---|---|---|---|
| Against teams at ≥ .700 (resting) | 105 | 48.6% ± 5.2 | |
| Against an eliminated team (≤ .400) facing a contender | 80 | 42.5% ± 4.6 | opposite way: dead teams covered 57.5%, a screen result |
| Against a team starting a non-regular QB | 115 | 57.4% ± 4.5 | z = 1.1 vs break-even, too small to act on |
| Over | 315 | 52.4% |
The week before the final week shows nothing either (overs 46.3%, z −2.1 vs break-even, but that's one of ~16 cells).
#7 Travel / body clock
| Rule | Bets | Won | 2006–15 | 2016–25 |
|---|---|---|---|---|
| Pacific team at Eastern home, 1 pm ET: bet home | 243 | 46.5% ± 3.2 | 50.9% | 43.1% |
| Pacific vs Eastern at night: bet Pacific | 90 | 57.8% ± 5.1 | 62.1% | 55.7% |
| Away team crossing ≥ 2 zones: bet home | 1,225 | 49.7% ± 1.5 | 51.8% | 47.9% |
The 10 am body-clock effect is gone or reversed: the market overprices it now. The night-game West
Coast edge from the sleep literature is too small a sample (90 bets in 20 years) to separate from noise.
Home teams covering less since 2016 matches falling home-field advantage (see hfa_estimation).
Totals teasers
6-point legs won 68.3% (Over) and 67.9% (Under). Break-even is 73.9% for 2-team at −120 and 72.7% for 3-team at +160. No band of closing total clears it, in either half. 7-point legs (70.9% / 70.4%) miss their 76.4% break-even by more. Totals have no key numbers like 3 and 7, so teaser points are worth much less than on sides. Totals teasers lose about 10–15%.
Verdict
About 25 rule cells tested; the closest to break-even are about what chance would produce. Nothing qualifies for pre-registration.
python3 backtests/nfl_schedule_edges/extract_fumbles.py 2006 2026 # PBP, one season per process
python3 backtests/nfl_schedule_edges/analyze.py