Calibration of the market sensor we cite (Brier, log loss, reliability bins)

The grading dataset for the market sensor we cite, and the published home of these numbers — forward (contamination-proof) and out-of-sample retrospective scores with confidence intervals. Freshness is PER-COMPONENT: the forward slice regenerates daily as markets mature; the retrospective slice is a fixed benchmark that re-runs on engine changes. Each slice carries its own generatedAt — read those, not the top-level asOf. Edge-cached 1 h. Rows: crowd (the prediction-market price) and base_rate (the naive null it is scored against), in both slices. No forecaster of ours is published — we sell a maintained, source-traced record and cite market prices as one graded sensor; we serve current state, not forecasts (since 2026-08-19, restated 2026-10-07). The payload states this in whatThisCovers.