Tennis Showcase
tennis has a served live model but no ingame_grade_joined corpus locally -- this showcase surfaces the two REAL preregistered gate receipts that ARE present (pregame cross-corpus + ingame surface-context detail layer), not a recomputed i...
How to read this figure
Two preregistered gate receipts, the cross-corpus pregame prior (lower held-out Brier is better) and the in-game surface-context layer (H1 surface-specific versus H0 surface-blind).
Continue with the data
Open the interactive analysis: Surface-Prior Brier DeltaWhat it means
The base (state, time) model, fit on one tour and tested leak-free on the other, is beaten by adding the prior in both directions, calibration only, no market claim. But making the hold prior surface-specific does not help: the pooled Brier change is only +0.000121 on ATP and +0.002492 on WTA, so the planted null dies.
Caveats and confounds
CALIBRATION only (held-out Brier / Diebold-Mariano). Tennis has a served live model but no in-game grade corpus locally, so this surfaces the two receipts that exist, not a recomputed in-game summary.