Module calibration 

Prediction calibration — Brier scorecard with the Murphy decomposition.

Tracks recorded predictions, resolves them against reality, and produces an honest calibration scorecard:

• Brier score: mean squared error (p − o)² over resolved forecasts
• Reliability: how well predicted probabilities match observed rates
• Resolution: how much predictions separate positive from negative outcomes
• Uncertainty: base rate ō(1 − ō) — the difficulty of the problem
• Brier skill score (BSS): 1 − Brier/Uncertainty vs. climatology

The calibration gap feeds a small adjustment back into future predictions, matching the v26 simulation.calibrate bridge but with the full decomposition the v26 version lacked.

Structs

BrierScorecard

Full Brier scorecard with the Murphy decomposition.

CalibrationBin

A single calibration bin: predicted-probability range vs. observed rate.

CalibrationPrediction

A single recorded prediction, resolved or pending.

CalibrationStore

In-memory calibration store — shared per MCP server instance.