tests/test_safety.py

@@ -6,6 +6,7 @@ from pathlib import Path
 import pytest
 import yaml
 
+from app.core.privacy import scrub
 from app.core.safety import (
     CrisisScript,
     SafetyResult,
@@ -40,6 +41,12 @@ def test_acute_recall_is_100_percent(case: dict) -> None:
     assert check(ScrubbedText(case["text"])).level == RiskLevel.ACUTE
 
 
+@pytest.mark.parametrize("case", ACUTE, ids=_ids(ACUTE))
+def test_acute_recall_survives_scrub(case: dict) -> None:
+    # Real pipeline: scrub() runs first, so PII placeholders must not break crisis detection.
+    assert check(scrub(case["text"])).level == RiskLevel.ACUTE
+
+
 @pytest.mark.parametrize("case", NOT_ACUTE, ids=_ids(NOT_ACUTE))
 def test_non_acute_cases_match_label(case: dict) -> None:
     # Exact level documents the calibration; the hard requirement is "never ACUTE" for these.