Objective
Implement $cocopod calibrate-rubric by reading recent rubric result history and updating each rubric entry's calibration metadata. A rule whose evaluated samples all score at or below 0.4 is marked never_fires; a rule with no samples is marked insufficient_evidence.
Exit Criteria
- Every rubric entry has sample count, maximum observed probability, status, and calibration timestamp.
- Calibration does not rewrite instruction text.
Anti-Rationalization
| Shortcut / Temptation | Why It Fails |
|---|---|
| Mark an unsampled rule as never firing | No evidence is different from low-scoring evidence. |
| Use wall-clock age as the signal | Calibration is based on observed operations. |