06 · Ensemble Verification Views

Theme 03 · Ensemble · data: synthetic (unavoidably) · reimagines MET ensemble_stat + METplotpy

Four coherent ensemble diagnostics over one synthetic dataset — rank (Talagrand) histogram, spread-skill, probabilistic reliability + sharpness + Brier decomposition, and CRPS — all driven by a single under → calibrated → over dispersion knob, so the connection between ensemble character and each diagnostic is visible live.

Structure

apps/ensemble-verification/
  index.html        entry
  src/ensemble.js   ensemble generation (dispersion knob)
  src/metrics/      rankhist · spreadskill · reliability · crps · contingency · aggregate
                    (thin adapters over lib/met-stats.mjs since consolidation)
  src/plot/         canvas plotting per diagnostic
  src/selftest.js   34-check app self-test

Status & notes

• Its math was promoted into the shared lib (RHIST, CRPS, spread-skill, reliability, Brier decomposition), proven by 30,094 random differential assertions.
• Stays synthetic by necessity: the real archive is deterministic-only — no ECNT/ORANK/RHIST lines exist to ingest. The gallery badge says so explicitly. This is the honest no-data posture the lab applies everywhere.
• Single-file build: dist/ensemble-verification.html (~588 KB).