Policy-impact digest¶

openmed.risk.policy_impact provides an offline dry run for reviewing a candidate policy version before promotion. It evaluates the effective action, gate, and waiver state for each safe resource type and aggregates the result by type.

from openmed.risk import evaluate_policy_impact

baseline = {
    "name": "baseline-v1",
    "actions": {"clinical_note": "keep"},
    "gates": {"clinical_note": "leakage"},
    "waivers": {"clinical_note": False},
}
candidate = {
    "name": "candidate-v2",
    "actions": {"clinical_note": "redact"},
    "gates": {"clinical_note": ["budget", "leakage"]},
    "waivers": {"clinical_note": True},
}

impact = evaluate_policy_impact(
    baseline,
    candidate,
    [{"resource_type": "clinical_note", "count": 4}],
)
print(impact.to_json())

The digest contains policy version names, safe resource-type counts, and counts for each action, gate, and waiver transition. It excludes resource identifiers, payload values, waiver reasons, and unknown policy fields. Its sha256: value is computed from the canonical counts-only payload, so changing the order of synthetic resources does not change the digest.

The simulator is read-only: it copies normalized scalar settings and never mutates a live policy or budget. It performs no network calls. The result is review evidence, not a compliance certification or a clinical decision.

The root openmed.risk namespace exports the policy description as PolicyImpactVersion and the comparison function as compare_policy_impact_versions. The defining module retains its local PolicyVersion and compare_policy_versions names. The root compare_policy_versions function continues to refer to policy migration.