Merge pull request 'fix(precedents): נרמול case_number עמיד-להתנגשות — מדלג ומתעד, לא קורס (#145)' (#266) from worktree-backfill-citations-run into main

mcp-server/src/legal_mcp/services/db.py

@@ -4112,6 +4112,20 @@ async def get_case_law_by_citation(case_number: str) -> dict | None:
     return _row_to_case_law(row) if row else None
 
 
+async def case_number_collides(case_number: str, exclude_id: UUID) -> bool:
+    """True if assigning ``case_number`` to a NON-internal row would violate the
+    partial unique index ``uq_case_law_external_number`` (``case_number`` WHERE
+    source_kind <> 'internal_committee') — i.e. another non-internal row already owns
+    that docket. Lets a caller SKIP the identity normalization (a duplicate to dedupe
+    later) instead of crashing the whole operation on the unique violation."""
+    pool = await get_pool()
+    return bool(await pool.fetchval(
+        "SELECT 1 FROM case_law WHERE case_number = $1 AND id <> $2 "
+        "AND source_kind <> 'internal_committee' LIMIT 1",
+        case_number, exclude_id,
+    ))
+
+
 async def create_external_case_law(
     case_number: str,
     case_name: str,

mcp-server/src/legal_mcp/services/precedent_metadata_extractor.py

@@ -364,7 +364,21 @@ async def apply_to_record(
         and cn_clean != cur_cn
         and (overwrite_case_number or citation_shaped)
     ):
-        fields_to_update["case_number"] = cn_clean
+        # Skip (don't crash) when the clean docket already belongs to ANOTHER
+        # non-internal row — a duplicate to dedupe later, not this run's concern.
+        # Writing it would hit uq_case_law_external_number and abort the whole merge
+        # (including the citation). No-silent-swallow: log the skip.
+        if (
+            record.get("source_kind") != "internal_committee"
+            and await db.case_number_collides(cn_clean, case_law_id)
+        ):
+            logger.warning(
+                "metadata_extractor: case_number normalization %r→%r skipped — docket "
+                "already owned by another non-internal row (likely duplicate)",
+                cur_cn, cn_clean,
+            )
+        else:
+            fields_to_update["case_number"] = cn_clean
 
     # parties — store the extracted "עורר נ' משיב" line (the re-derivable basis for
     # the deterministic citation). Only fill when empty; chair edits are preserved.