Get Calibration

Post-selection calibration over graded bets in a date range.

Bins each placed bet's stated probability against how often it won, and the claimed edge against the realized edge, per market. suggested_lambda is the blend-to-market weight implied by the claimed-vs-realized gap (the Phase 4 shrinkage input). Shares the pure aggregation with the offline report script.

The result is cached for 15 minutes under the exact (start_date, end_date, sport) the caller asked for, in the shared cache in api.calibration_cache, so a repeated 30-day admin refresh returns without running the query again. Concurrent callers that miss the same key await one shared asyncio.Task.

This route holds its own one-token AnyIO capacity limiter and the public transparency route holds a separate one, so each API process runs at most two calibration queries at once and an admin request never waits behind a public refresh. Callers awaiting a token do not occupy the default AnyIO thread capacity.