deidkit 0.1.3__tar.gz → 0.1.4__tar.gz

{deidkit-0.1.3/src/deidkit.egg-info → deidkit-0.1.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deidkit
-Version: 0.1.3
+Version: 0.1.4
 Summary: Schema-driven pseudonymization for clinical/tabular datasets: synthetic names, interval-preserving date shifting, multi-stage free-text PII detection, and before/after audit exports.
 Author: Meridian Data
 License: MIT

{deidkit-0.1.3 → deidkit-0.1.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "deidkit"
-version = "0.1.3"
+version = "0.1.4"
 description = "Schema-driven pseudonymization for clinical/tabular datasets: synthetic names, interval-preserving date shifting, multi-stage free-text PII detection, and before/after audit exports."
 readme = "README.md"
 requires-python = ">=3.9"

{deidkit-0.1.3 → deidkit-0.1.4}/src/deidkit/detect/pipeline.py

@@ -238,4 +238,9 @@ def _score_person(d: Detection) -> float:
         score += 0.10
     if len(methods) >= 2:
         score += 0.20
-    return min(score, 0.99)
+    # Round to 2 dp: the score is built from 2-decimal weights, and the mode
+    # thresholds are 2-decimal (0.80 / 0.65 / 0.35). Without this, float dust
+    # (0.30 + 0.35 = 0.6499999…) drops a lone NER hit just under the 0.65
+    # "balanced" threshold, so balanced would never accept single-model NER hits
+    # as documented.
+    return round(min(score, 0.99), 2)

{deidkit-0.1.3 → deidkit-0.1.4}/src/deidkit/version.py

@@ -1,3 +1,3 @@
 """Single source of truth for the package version."""
 
-__version__ = "0.1.3"
+__version__ = "0.1.4"

{deidkit-0.1.3 → deidkit-0.1.4/src/deidkit.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: deidkit
-Version: 0.1.3
+Version: 0.1.4
 Summary: Schema-driven pseudonymization for clinical/tabular datasets: synthetic names, interval-preserving date shifting, multi-stage free-text PII detection, and before/after audit exports.
 Author: Meridian Data
 License: MIT

{deidkit-0.1.3 → deidkit-0.1.4}/src/deidkit.egg-info/SOURCES.txt

@@ -67,4 +67,5 @@ tests/test_ner_model_helpers.py
 tests/test_qa_fixes.py
 tests/test_qa_freetext_leaks.py
 tests/test_qa_hardening.py
-tests/test_schema_lang.py
+tests/test_schema_lang.py
+tests/test_scoring.py

deidkit-0.1.4/tests/test_scoring.py

@@ -0,0 +1,24 @@
+"""Person-score threshold behaviour.
+
+A lone statistical-NER hit must score exactly at the 'balanced' threshold (0.65),
+not 0.6499999… — otherwise balanced mode silently fails to accept single-model
+NER hits (the documented behaviour) because of float dust.
+"""
+
+from deidkit.detect.pipeline import _score_person, _THRESHOLDS
+from deidkit.detect.types import Detection, PERSON
+
+
+def test_lone_ner_hit_scores_exactly_at_balanced_threshold():
+    d = Detection(0, 15, "Yeimy Bocanegra", PERSON, {"ner"})
+    score = _score_person(d)
+    assert score == 0.65                          # not 0.6499999999999999
+    assert score >= _THRESHOLDS["balanced"]       # balanced (0.65) accepts it
+    assert score < _THRESHOLDS["conservative"]    # conservative (0.80) -> review
+
+
+def test_score_is_two_decimal_quantized():
+    # a mix that would accumulate float error still lands on a clean 2-dp value
+    d = Detection(0, 10, "X", PERSON, {"ner"}, meta={"run_len": 2})
+    s = _score_person(d)
+    assert round(s, 2) == s