Upload pii_hybrid_decode.py with huggingface_hub

pii_hybrid_decode.py

@@ -142,7 +142,37 @@ def hybrid_spans(text: str, model_spans: list[dict]) -> list[dict]:
         k = (sp["start"], sp["end"], sp["type"])
         if k in seen: continue
         seen.add(k); final.append(sp)
-    return final
+    # GROUP-B cue tier: amount/date/phone/postal where the model stayed silent but a
+    # field-cue+shape is present (authoritative for those 4 types; AUTH wins; address excluded).
+    try:
+        from context_cued import group_b_cue_spans
+    except Exception:
+        return final
+    gb = [c for c in group_b_cue_spans(text)
+          if c["type"] in ("financial.amount","identity.date_of_birth","contact.phone","contact.postal_code")]
+    if not gb:
+        return final
+    aclaim = [False] * len(text)
+    for sp in final:
+        if sp["type"] in _AUTH_TYPES:
+            for i in range(sp["start"], sp["end"]): aclaim[i] = True
+    keptb = []
+    for c in gb:
+        if any(aclaim[c["start"]:c["end"]]): continue
+        s, e = c["start"], c["end"]
+        if c["type"] == "financial.amount":
+            while e > s and text[e-1] in ".,;": e -= 1
+        keptb.append({"start": s, "end": e, "type": c["type"], "text": text[s:e]})
+    rb = [(c["start"], c["end"]) for c in keptb]
+    def _ovb(sp): return any(min(e, sp["end"]) > max(s, sp["start"]) for s, e in rb)
+    GB = {"financial.amount","identity.date_of_birth","contact.phone","contact.postal_code"}
+    merged2 = [sp for sp in final if not (sp["type"] in GB and _ovb(sp))] + keptb
+    seen, out2 = set(), []
+    for sp in sorted(merged2, key=lambda s: (s["start"], s["end"])):
+        k = (sp["start"], sp["end"], sp["type"])
+        if k in seen: continue
+        seen.add(k); out2.append(sp)
+    return out2
 
 def model_spans(text: str, tok, model):
     import torch