Uncertainty-disclosure completeness audit¶

openmed.clinical.uncertainty_disclosure provides a local, deterministic structural audit for guarded claim metadata. It checks whether each claim has uncertainty categories, reason codes, evidence or provenance references, a recognized review state, and bounded display limits. It does not interpret the claim, assess clinical correctness, certify compliance, or make a clinical decision.

from openmed.clinical.uncertainty_disclosure import audit_uncertainty_disclosures

claims = [
    {
        "claim_id": "synthetic-claim-001",
        "uncertainty_disclosure": {
            "uncertainty_categories": ["epistemic"],
            "reason_codes": ["reason.synthetic"],
            "evidence_references": ["evidence.synthetic.001"],
            "review_state": "pending",
            "display_hints": {"max_chars": 240, "max_items": 4},
        },
    }
]

report = audit_uncertainty_disclosures(claims)
assert report.is_complete

The default display bounds are max_chars from 1 through 4096 and max_items from 1 through 100. max_lines is available from 1 through 100, and callers can choose which bounded hints are required with required_display_hints. Every supplied known hint is checked, including optional hints. Multiple aliases for the same bound are rejected as ambiguous. Direct report construction requires canonical hexadecimal digests and integer counts consistent with its findings. Callers may also provide required_categories and a minimum evidence-reference count. The fields can be top-level or nested under uncertainty_disclosure, uncertainty, or metadata containers.

Reports are safe for audit logs and review tooling: claim identifiers become opaque SHA-256 keys, findings use fixed issue codes, and aggregate issue counts are emitted instead of category names, reason codes, references, display values, or other claim metadata. The implementation performs no network calls and does not write files.