[cschanhniem]

Implements Q3 from NEXT_100_PR_MAP.

What

• src/openamp_foundry/evidence/post_experiment_calibration_update.py: PCU- schema; PostExperimentCalibrationUpdate dataclass (16 fields); validate_post_experiment_calibration_update(); build_post_experiment_calibration_update() (enforces dry_lab_only=False, human_reviewed=True); format_post_experiment_calibration_update()
• VALID_UPDATE_DIRECTIONS (5): increase_threshold, decrease_threshold, no_change, expand_candidate_pool, restrict_candidate_pool
• VALID_CALIBRATION_TRIGGER_TYPES (7): hit_rate_above_baseline, hit_rate_below_baseline, novel_family_detected, safety_signal_observed, systematic_prediction_error, insufficient_data, scheduled_review
• tests/evidence/test_post_experiment_calibration_update.py: 63 tests in 5 classes
• BASELINE: 8711 → 8774

Disconfirming pass

• dry_lab_only=True blocked (tested) — real experiment-triggered updates only
• human_reviewed=False blocked (tested) — no auto-approved calibration changes
• n_whr_records=0 blocked (tested) — evidence required before any calibration change
• n_whr_records/whr_ids length mismatch blocked (tested)
• update_magnitude out of [0,1] blocked (tested)
• empty rationale and limitations blocked (tested)