[cschanhniem]

Implements H6 from NEXT_100_PR_MAP: simulation uncertainty calibration report.

What

• src/openamp_foundry/simulation/uncertainty_calibration_report.py: SUC- prefix schema
• SimulationUncertaintyCalibrationReport dataclass: 14 fields
• VALID_SIMULATION_MODULES (7 values): membrane_proxy, structure_proxy, ensemble_checker, external_adapter, dummy_module, baseline_registry, other
• VALID_CALIBRATION_METHODS (6 values): isotonic_regression, platt_scaling, temperature_scaling, histogram_binning, empirical_coverage_only, none_applied
• VALID_UNCERTAINTY_ASSESSMENT_STATUSES (5 values): well_calibrated, overconfident, underconfident, insufficient_data, not_assessed
• OVERCONFIDENCE_THRESHOLD = 0.15, MIN_SAMPLES_FOR_CALIBRATION = 10
• 10 validation rules: prefix, module, method, status, coverage range, calibration_error consistency, insufficient_data gate, dry_lab_only, overconfidence_flag consistency, non-blank fields
• validate_uncertainty_calibration_report(), format_uncertainty_calibration_report()
• tests/simulation/test_uncertainty_calibration_report.py: 63 tests, 5 classes
• BASELINE: 8221 -> 8284

Why

A simulation module that claims 80% confidence intervals should actually contain the true value 80% of the time. Without SUC- records, over-confident modules silently inflate trust in rankings. The overconfidence_flag and cheap_baseline_beats_module fields give calibration reviewers a machine-checkable gate before a module gets ranking authority.

Disconfirming pass

• dry_lab_only enforced on all reports (cannot be set False)
• insufficient_data status required when n_samples < 10 (prevents over-claiming calibration from small samples)
• calibration_error must equal abs(expected - empirical) within 0.01 tolerance (prevents manual override of error magnitude)