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.
Pick'em tiebreakers: what to actually guess
2026-09-09. 2,895 games and 192 full weeks, 2015-2025, against closing lines.
Three questions that look similar and are not.
1. Guessing a game's total score
As an ESTIMATION problem the answer is boring: the closing total is already the best single guess.
| guess | MAEmean absolute errorAverage size of the miss, ignoring direction. If a projection is off by 3 one week and -5 the next, the MAE is 4. Lower is better. |
|---|---|
| closing total | 10.43 |
| closing total, best constant shift (−0.5) | 10.42 |
| the league mean, 45.6 | 10.96 |
The line is unbiased (mean error +0.40) and no constant shift improves it by more than a rounding error. Note how little it beats a flat 45.6 — half a point of MAE. Totals are mostly unpredictable; SD around the line is 13.2.
But a tiebreaker is not scored on your error. It is scored on being closest. If everyone guesses the line, the line is the most contested number on the board, and being exactly right there still loses to whoever is a point closer. Simulating against opponents who cluster at the line (SD 3), using the real error distribution rather than a Gaussian:
| your deviation from the line | 0 | 2 | 4 | 6 | 8 | 10 | 12 | 15 | 20 |
|---|---|---|---|---|---|---|---|---|---|
| win rate vs 4 opponents | 10% | 18% | 30% | 34% | 32% | 31% | 27% | 24% | 18% |
| vs 9 opponents | 2% | 6% | 19% | 29% | 30% | 29% | 26% | 22% | 17% |
| vs 19 opponents | 1% | 2% | 9% | 24% | 29% | 28% | 25% | 22% | 16% |
⇒ deviate 6–8 points from the closing total, in either direction. In a 20-person pool that turns a 1% win rate into 29%. The optimum is interior: far enough to escape the crowd, near enough that outcomes are still dense.
⚠ This rests on an assumption I could not measure: that opponents cluster near the line. I have no data on what real pool entrants guess. If they scatter more (the same simulation with herd SD 6 puts the optimum at 12 and the payoff at 20% vs 5%), or if they pile on round numbers like 40/45/50 rather than the line, the right deviation changes. The direction of the advice is robust — do not guess the consensus — the exact number is not.
2 & 3. Highest and lowest scoring team of the week
These are order statistics, not estimates. You are not asked who scores most on average; you are asked who tops a 32-team maximum.
That made me expect variance to matter as much as mean — a volatile offence should win a maximum more often than a steady one with the same average. It does not. Adding a team's prior-season scoring SD makes the pick worse:
| criterion | picks the week's HIGHEST scorer |
|---|---|
| highest implied team total | 13.0% |
| implied + 0.5 × SD | 11.5% |
| implied + 1.0 × SD | 11.5% |
| highest game total (pace) | 8.9% |
| random | 3.1% |
| criterion | picks the week's LOWEST scorer |
|---|---|
| lowest implied team total | 18.2% |
| implied − 0.5 × SD | 14.6% |
| implied − 1.0 × SD | 12.5% |
| lowest game total (pace) | 6.8% |
| random | 3.1% |
The likely reason is the one this project keeps meeting: a team's scoring SD is itself noisy year to year, so the variance term adds noise rather than signal.
Two things worth knowing beyond the criterion:
- The low scorer is easier to call than the high scorer (18.2% vs 13.0%). Bad offences are more reliably bad than good offences are reliably explosive, and the week's low scorer usually got blown out — which the spread already flags.
- Expect to be wrong. The single highest-implied team wins the week only 13% of the time; the actual high scorer sits 7th in implied total in a median week. Top-3 implied covers just 24.5% of weeks. Both criteria beat random by 4–6×, and that is all they do.
For calibrationcalibrationA deliberate sanity check on the method itself: run it on something already known to be true. If it fails to detect the known thing, the method is broken and its other results mean nothing.: the week's high scorer averages 44.4 points (range 33–70), the low scorer 4.1 (range 0–11). The highest-implied team averages 30.2 and the lowest-implied 14.3 — so if the tiebreaker asks for the SCORE rather than the team, guess well above the implied total for the high and well below it for the low.
Summary
| tiebreaker | strategy |
|---|---|
| game total | closing total ± 6–8, never the line itself |
| highest scoring team | highest implied team total; ignore volatility |
| lowest scoring team | lowest implied team total; ignore volatility |
| high team's score | ~44, not the ~30 the implied total suggests |
| low team's score | ~4, not the ~14 the implied total suggests |