TENNIS / Surface evidence
Surface-Prior Brier Delta
Keeps the published surface-specific versus surface-blind Brier comparison and its fold direction count together for each tour.
Published measurements
Choose a measurement
Inspect 6 held-out fold results
Surface-specific versus surface-blind priors
Each table keeps one tour separate. Lower Brier means lower squared probability error; a positive surface-minus-blind difference means higher error for the surface-specific variant.
ATP
| Source fold | Test states | Surface-blind Brier | Surface-specific Brier | Surface minus blind | Source row |
|---|---|---|---|---|---|
| 0 | 8,780 | 0.166648 | 0.166744 | +0.000096 | Source pathingame_surface_context.tours.atp.n_folds[0] |
| 1 | 8,780 | 0.163305 | 0.163347 | +0.000042 | Source pathingame_surface_context.tours.atp.n_folds[1] |
| 2 | 8,780 | 0.166239 | 0.166463 | +0.000224 | Source pathingame_surface_context.tours.atp.n_folds[2] |
WTA
| Source fold | Test states | Surface-blind Brier | Surface-specific Brier | Surface minus blind | Source row |
|---|---|---|---|---|---|
| 0 | 2,646 | 0.148274 | 0.149926 | +0.001652 | Source pathingame_surface_context.tours.wta.n_folds[0] |
| 1 | 2,645 | 0.151989 | 0.154370 | +0.002381 | Source pathingame_surface_context.tours.wta.n_folds[1] |
| 2 | 2,646 | 0.151421 | 0.154863 | +0.003442 | Source pathingame_surface_context.tours.wta.n_folds[2] |
Differences subtract the two rounded scores published for the same fold. Scores and differences remain in Brier units. Test states can include repeated observations of a match; independent game counts and calendar test dates are not published here.
Fold identifiers and order follow the source. These full-receipt tables do not change with the main analysis search; the summary CSV contains the tour rows. The recorded surface-prior comparison was rejected. The fold display does not establish a general surface effect or a new validation result.
Inspect the published tennis fold source2 matching rows; 2 contain surface minus blind brier.
CSV includes 2 matching rows, all fields, and source context; missing measurements stay blank. Published source dates and row provenance are included when available; structured provenance uses JSON cells.
- Measured rows
- 2 / 20 unavailable
- Median
- 0.001307Middle ranked value
- Middle 50%
- 0.000714 to 0.00189925th to 75th percentile
- Observed range
- 0.000121 to 0.002492Minimum to maximum
Summaries describe the filtered rows with equal row weight. Row counts are not independent game or event counts.
Measurement availability8 / 8 fields have values
Reference population: All published groups. 2 rows published.
Text search does not change this reference population.
Availability describes published values only. It does not establish data quality or independent game counts, and measurement windows can differ.
Showing 2 of 2 matching numeric rows. Open a row to inspect all its fields. Rankings apply only to the published subset.