Decision D14 · 14 of 21
Calibration gate thresholds
Per model class
DECIDED
- Status
- DECIDED (7 Oct 2026, Vinayak)
- Options
- One global coverage/NIS/ALS cut-off; or per model class; or per plant
- Trade-offs
- Uniformity vs fit
- Recommendation
- Per model class. Estimators and forecasters: 85–95% empirical coverage for nominal 90% intervals on unseen data, and NIS inside its 95% chi-squared band, checked after ALS and before ACI per horizon. Targets, not results.
- Why
- One global number would be wrong for most classes.
- Evidence
- 7 Oct briefing intake; sources in evidence ledger section F.
- Reversible
- Per class.
- Blocks
- Other classes set thresholds when each first ships.
Page history: last 2 changes
- docs(research): retire stale research to archive/research-2026-10 with a register
ab84821 - docs(technical): split decision board into DECISIONS.md
b4db9d4