[cschanhniem]

Summary

• New module: src/openamp_foundry/evidence/charge_distribution_report.py
• compute_charge_report(sequences, labels) → ChargeDistributionReport
• Reveals the charge shortcut: if benchmark positives have much higher charge than negatives, a simple charge threshold rule might explain most discrimination
• ChargeDistributionReport fields: n_positive, n_negative, mean/median charge for both groups, fraction_positive_high_charge, fraction_negative_high_charge, charge_ratio, charge_shortcut_likely (bool), shortcut_warning (str), positive_charges, negative_charges
• Shortcut detection: fraction_positive_high_charge >= 0.60 (60%) OR charge_ratio >= 1.5 triggers warning
• format_charge_report() for human-readable text output
• 64 tests in tests/evidence/test_charge_distribution_report.py covering: net charge proxy, basic compute, shortcut detection, edge cases, median/mean stats, format report, integration
• BASELINE 6835→6898
• Closes Phase C C3 — every benchmark can now be analyzed for the charge shortcut via a machine-checkable signal