Cohort Brier Dashboard

Aggregated Brier calibration sliced by asset class, tier, horizon, and regime. Validates the honesty + calibration claim end-to-end. Cohorts with N<30 show calibrating pills — scores are NEVER fabricated. Brier 0.25 = no-skill benchmark (always predict p=0.5).

Global Mean Brier

Claims

Resolved: —

Pending: —

Brier Score Reference Scale

0
perfect 0.25
no-skill (p=0.5) 1.0
anti-skill

Lower is better. Brier = (p−outcome)² averaged. 0.25 = random baseline (always predict 50%). <0.15 = skillful. >0.25 = anti-skill (worse than random).

By Class

By Tier

By Horizon

By Regime

Synthetic-backfill cohort separated: only prospective claims shown here. Add ?synthetic=1 to view backfill cohort (not yet populated).

Honesty contract: Every Brier number on this page is computed live from forecast-ledger-supabase.summarize(). No values are fabricated. Cohorts with N<30 are permanently suppressed until the threshold is crossed.

N disclosed on every row — format Brier 0.21 · N=142, never naked numbers.
Brier 0.25 reference shown inline — never dressed as "good".
ECE (Expected Calibration Error, Naeini 2015) requires per-claim pairs; shown when available.
Synthetic-backfill separated — add ?synthetic=1 to include (Memo C Lane B-2 migration required).
See Track Record for per-claim history · Calibration Scoreboard for per-type cohort view.

methodology_hash: cohort-brier-render-2026-05-01-v1