Upload pii_hybrid_decode.py with huggingface_hub

pii_hybrid_decode.py

@@ -120,7 +120,29 @@ def hybrid_spans(text: str, model_spans: list[dict]) -> list[dict]:
         if k in seen: continue
         if len(text[sp["start"]:sp["end"]].strip()) < 3: continue  # drop fragments
         seen.add(k); uniq.append(sp)
-    return uniq
+    # CONTEXT tier: cue-gated alphanumeric IDs (Passport No:/Policy #/MRN: ...) — recovers
+    # the structure-less IDs the model can't learn. Authoritative for their types; AUTH wins.
+    try:
+        from context_cued import context_cued_spans, CONTEXT_TYPES
+    except Exception:
+        return uniq
+    cued = context_cued_spans(text)
+    if not cued:
+        return uniq
+    auth_claim = [False] * len(text)
+    for sp in uniq:
+        if sp["type"] in _AUTH_TYPES:
+            for i in range(sp["start"], sp["end"]): auth_claim[i] = True
+    kept = [c for c in cued if not any(auth_claim[c["start"]:c["end"]])]
+    rng = [(c["start"], c["end"]) for c in kept]
+    def _ov(sp): return any(min(e, sp["end"]) > max(s, sp["start"]) for s, e in rng)
+    merged = [sp for sp in uniq if not (sp["type"] in CONTEXT_TYPES and _ov(sp))] + kept
+    seen, final = set(), []
+    for sp in sorted(merged, key=lambda s: (s["start"], s["end"])):
+        k = (sp["start"], sp["end"], sp["type"])
+        if k in seen: continue
+        seen.add(k); final.append(sp)
+    return final
 
 def model_spans(text: str, tok, model):
     import torch