Calibration / Brier audit
Inspect the Brier reconstruction remainder.
This view separates the published reliability, resolution, and uncertainty components for each population, then shows the difference between each binned reconstruction and the published Brier score.
Reading trail
Read first: Read the calibration reliability view before assigning meaning to a Brier component.
Next question: Which published reliability components can be compared across sports?
Read the analysis: What carries across sports (sources regenerated)
| Sport | Sport n rows | Population | Population n | Brier | Reconstructed Brier (derived) | Signed remainder (derived) |
|---|---|---|---|---|---|---|
| MLB | n rows=27,351 | Model | n=27,351 | 0.166759 | 0.168327 | 0.168327 - 0.166759 = 0.001568 |
| MLB | n rows=27,351 | Market | n=27,351 | 0.151008 | 0.151258 | 0.151258 - 0.151008 = 0.000250 |
| International soccer | n rows=4,265 | Model | n=4,265 | 0.332522 | 0.366753 | 0.366753 - 0.332522 = 0.034231 |
| International soccer | n rows=4,265 | Market | n=4,265 | 0.195157 | 0.230310 | 0.230310 - 0.195157 = 0.035153 |
The reconstruction is computed from ten probability bins. A binned reconstruction can differ from the published Brier; this table reports that signed remainder without assigning a cause to it.
Source fields: murphy_decomposition.json -> sports[sport].n_rows; model_prob and market_prob: n, brier, reliability, resolution, uncertainty, reconstructed_brier. Reconstructed Brier and signed remainder are shown as derived values; remainder = reconstructed_brier - brier.