Skip to content
Browse › Kernel Transfer
Analytics module · as of 2026-07-22

Kernel transfer: one honest cross-sport calibration axis

This table's only cross-sport-comparable number is the Murphy reliability component, for 2 of 4 sports where a market-side probability decomposition exists (mlb moneyline, soccer_intl moneyline). CRPS (mlb totals/margin), Brier-without-d...
View full size ↗
confirmednull (a finding)not testabledescriptivepending
Kernel Transfer
Chart: Kernel Transfer -- This table's only cross-sport-comparable number is the Murphy reliability component, for 2 of 4 sports where a market-side probability decomposition exists (mlb moneyline, soccer_intl moneyline). CRPS (mlb totals/margin), Brier-without-d...
scripts/platformkit/analytics_showcase/out/kernel_transfer.json2026-07-22

What it means

The kernel-transfer table composes Murphy reliability across NBA, MLB, soccer, and tennis, but only MLB and soccer have a market probability to decompose against. In both, the model's calibration component trails the market's. The gaps agree in sign but differ in magnitude by roughly 5.9x, which is suggestive rather than a proven pattern.

Caveats & confounds

n=2 sports is not enough to call this a general cross-sport pattern; it is a same-direction coincidence worth re-checking if a third sport gets a market-side decomposition, not a claim.

Method. composition only -- reliability/resolution numbers copied verbatim from murphy_decomposition.py / soccer_calibration_pack.py output; verdicts copied/summarized verbatim from each source artifact, nothing recomputed

Ask Scout about this
Compare the model and the market on MLB in-game moneyline.Which sports can be compared on the same axis at all?Is the model better at win probability or at totals?Ask anything →