Stabilize near-gate regional sampling

scripts/build_failure_driven_yolo_sampling.py

@@ -51,6 +51,8 @@ def build_sampling(
     context_negative_repeat: int = 6,
     max_region_share: float = 0.65,
     sampling_round: int = 0,
+    precision_guard_band: float = 0.03,
+    recall_guard_band: float = 0.03,
 ) -> tuple[list[str], dict[str, Any]]:
     if assessment.get("status") != "continue_training_loop":
         raise ValueError("Failure-driven sampling requires a failed assessment")
@@ -66,6 +68,8 @@ def build_sampling(
         raise ValueError("max_region_share must be in (0, 1]")
     if sampling_round < 0:
         raise ValueError("sampling_round must be non-negative")
+    if precision_guard_band < 0 or recall_guard_band < 0:
+        raise ValueError("Guard bands must be non-negative")
 
     samples = {item["sample_slug"]: item for item in manifest["samples"]}
     gates = assessment["gates"]
@@ -77,12 +81,12 @@ def build_sampling(
         region
         for region, metrics in regions.items()
         if metrics["f1"] < gates["min_region_f1"]
-        or metrics["recall"] < gates["min_region_recall"]
+        or metrics["recall"] < gates["min_region_recall"] + recall_guard_band
     }
     weak_precision_regions = {
         region
         for region, metrics in regions.items()
-        if metrics["precision"] < gates["min_region_precision"]
+        if metrics["precision"] < gates["min_region_precision"] + precision_guard_band
     }
     background = assessment.get("background")
     background_failed = bool(
@@ -206,6 +210,8 @@ def build_sampling(
         "context_negative_repeat": context_negative_repeat,
         "max_region_share": max_region_share,
         "sampling_round": sampling_round,
+        "precision_guard_band": precision_guard_band,
+        "recall_guard_band": recall_guard_band,
         "source_train_tile_count": sum(
             1
             for tile in summary["tiles"]
@@ -235,6 +241,8 @@ def main() -> int:
     parser.add_argument("--context-negative-repeat", type=int, default=6)
     parser.add_argument("--max-region-share", type=float, default=0.65)
     parser.add_argument("--sampling-round", type=int)
+    parser.add_argument("--precision-guard-band", type=float, default=0.03)
+    parser.add_argument("--recall-guard-band", type=float, default=0.03)
     args = parser.parse_args()
 
     summary = json.loads(args.summary.read_text(encoding="utf-8"))
@@ -255,6 +263,8 @@ def main() -> int:
         context_negative_repeat=args.context_negative_repeat,
         max_region_share=args.max_region_share,
         sampling_round=sampling_round,
+        precision_guard_band=args.precision_guard_band,
+        recall_guard_band=args.recall_guard_band,
     )
     args.output_dir.mkdir(parents=True, exist_ok=True)
     train_list = args.output_dir / "train-failure-driven.txt"