Gating De-id Leakage in CI

Logs, baselines, and models drift. The only durable defense is a gate that runs on every change and fails closed when de-identification regresses. This skill operationalizes OpenMed's leakage-first ethos into a CI check: recall must stay above the floor and critical leakage must be exactly zero, or the build goes red.

When to use this skill

• You want a pytest test or CLI step that exits nonzero on de-id regression.
• You need to block PRs that drop PHI recall or introduce a leak.
• You want OpenMed's release gates (ReleaseGate, G1a, G8) enforced in CI.
• You maintain a synthetic held-out PHI set and want it checked automatically.

For the full gate semantics see evaluating-with-leakage-gates; this skill is about wiring it into CI so it fails the build.

Quick start, a pytest gate

# tests/eval/test_deid_leakage_gate.py
import pytest
from openmed.eval import run_suite, ReleaseGate, RELEASABLE

RECALL_FLOOR = 0.99          # direct-identifier recall floor
HELD_OUT = "eval/heldout/phi_synthetic.json"   # SYNTHETIC, committed

@pytest.fixture(scope="module")
def gate_report():
    report = run_suite(
        HELD_OUT,
        suite="golden",
        model_name="OpenMed/Privacy-PII-Detection",
        device="cpu",
        metadata={"family": "PII", "tier": "base", "policy": "hipaa_safe_harbor"},
    )
    return ReleaseGate(milestone="v1.6", policy="hipaa_safe_harbor").evaluate(report)

def test_no_critical_leakage(gate_report):
    # Hard zero: one leaked SSN/credit-card is a breach, full stop.
    assert gate_report.critical_leakage_count == 0, "critical PHI leaked"

def test_recall_floor(gate_report):
    low = {
        label: r
        for label, r in gate_report.per_label_recall.items()
        if r < RECALL_FLOOR
    }
    assert not low, f"recall below floor: {low}"