Keep rejected training regressions from propagating

2026-07-29 23:33:53 +02:00
parent b8e40ba7a8
commit a5b321fdb6
2 changed files with 60 additions and 2 deletions
@@ -266,3 +266,34 @@ def test_threshold_selection_uses_worst_region_then_aggregate() -> None:
        ]
    }
    assert MODULE.select_calibration_threshold(report)["threshold"] == 0.2
+
+
+def test_rejected_candidate_score_prioritizes_weakest_release_gate() -> None:
+    gates = {
+        "min_aggregate_f1": 0.55,
+        "min_region_f1": 0.45,
+        "min_region_precision": 0.5,
+        "min_region_recall": 0.4,
+    }
+    incumbent = {
+        "gates": gates,
+        "calibration": {
+            "aggregate": {"f1": 0.58},
+            "regions": {
+                "flanders": {"f1": 0.34, "precision": 0.38, "recall": 0.31},
+                "wallonia": {"f1": 0.60, "precision": 0.50, "recall": 0.75},
+            },
+        },
+    }
+    regressed = {
+        "gates": gates,
+        "calibration": {
+            "aggregate": {"f1": 0.60},
+            "regions": {
+                "flanders": {"f1": 0.31, "precision": 0.45, "recall": 0.24},
+                "wallonia": {"f1": 0.62, "precision": 0.52, "recall": 0.77},
+            },
+        },
+    }
+
+    assert MODULE.rejected_candidate_score(incumbent) > MODULE.rejected_candidate_score(regressed)