33.1%
Go Rate on Clear-Go Calls (Margin ≥ 1 pp)
95% CI: [30.2%, 36.2%]
37.07 pp
WP Sacrificed per Team per Game
95% CI: [34.20, 39.86] pp
32.3%
Overall Coach Decision Agreement
95% CI: [30.6%, 33.9%]
3,997
Total 2025 Decisions Audited
Held-Out Unseen Season
Sensitivity Analysis: -5.0 pp Conversion Probability Haircut
To test whether coach conservatism could be justified by model optimism, we imposed an aggressive uniform 5.0 percentage point haircut on every conversion probability ($\hat{P}_{ ext{conv}} - 0.05$).
| Scenario |
Clear-Go Opportunities |
Go Rate on Clear Go |
95% Bootstrap CI |
WP Lost / Team-Game |
95% Bootstrap CI |
| Baseline Model |
1,198 |
33.1% |
[30.2%, 36.2%] |
37.07 pp |
[34.20, 39.86] |
| Sensitivity (-5 pp Haircut) |
1,084 |
34.9% |
[31.9%, 38.0%] |
34.09 pp |
[31.50, 36.61] |
Empirical Conclusion: Even under an extreme 5 pp penalty, 90.5% of Clear-Go situations remain mathematically Clear-Go, and coaches still only went 34.9% of the time. The conservatism gap is not an artifact of model overconfidence.