feat(precedents): איחוד cited_only↔missing_precedents — גזירת פסיקה-חסרה (#143, G2)

שתי מערכות מקבילות לאותו מושג ("פסיקה מצוטטת שטקסטה לא נקלט"): טבלת
missing_precedents (תור-רכישה ידני של היו"ר) מול case_law source_kind='cited_only'
(stubs מגרף-הציטוטים/X11). חפיפה≈0 → 31 ה-stubs לא הופיעו ב-/missing-precedents.

הכרעה (G2): missing_precedents = SoT-לתור-יחיד; cited_only = מקור-גילוי נגזר (כמו
יומונים מזינים radar). גוזרים רשומת missing_precedents 'open' לכל stub.

תיקון:
- court_citation.citation_dedup_key — מפתח-dedup **designator-aware**
  (`{designator}|{docket}`). **מתקן פגם בתוכנית-הניתוח:** dedup על מספר-בלבד היה
  ממזג בטעות אותו docket בערכאות שונות (בג"ץ 389/87 ≠ ע"א 389/87; 18 כאלה בקיים).
- סכמה V40: missing_precedents מקבל citation_norm (מפתח-dedup) + discovery_source
  (manual|cited_only|digest|writer) + index. **בלי UNIQUE** — הקורפוס מחזיק
  לגיטימית אותו docket בערכאות שונות; ייחודיות נאכפת designator-aware בנתיב-היצירה.
- create_missing_precedent: מחשב citation_norm בכתיבה (G1), מקבל discovery_source
  + linked_case_law_id. find_missing_precedent_by_citation: dedup דרך citation_norm
  (fallback ל-citation גולמי כשאין מספר).
- scripts/derive_missing_from_cited_only.py: backfill citation_norm ל-291 +
  גזירת 31 (dry-run: 31 ייווצרו, 0 deduped). linked_case_law_id=stub, status=open
  → promote-in-place בהעלאת-טקסט דרך ON CONFLICT הקיים. אידמפוטנטי.

תלוי-הקשר: #140 (הגדרת cited_only). מתואם עם #136 (digest→MP — אותו citation_norm
+ create path). תיקון-נתון יורץ אחרי הפריסה.

בדיקות: test_dedup_key_is_designator_aware (בג"ץ≠ע"א, ערר≠בל"מ, גרסאות-format
מתמזגות). כל 356 עוברות. guards נקיים.

Invariants: G2 (SoT-לתור יחיד, cited_only נגזר), G1 (citation_norm מנורמל בכתיבה),
G3 (idempotent upsert), G10 (שער-העלאה ידני נשמר), G12.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

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

@@ -134,6 +134,32 @@ def case_number_from_citation(citation: str) -> str:
     return classify(citation).case_number_norm
 
 
+def _norm_designator(prefix: str) -> str:
+    """Collapse a court/proceeding prefix to a comparison token.
+
+    Strips gershayim variants and whitespace so ``עע"מ`` / ``עע״מ`` / ``עעמ``
+    all map to the same token, while DISTINCT courts stay distinct.
+    """
+    return re.sub(r"[\s\"״׳']", "", prefix or "")
+
+
+def citation_dedup_key(citation: str) -> str:
+    """Designator-aware dedup key for a citation, or ``''`` if no number.
+
+    Returns ``f"{designator}|{case_number_norm}"`` — e.g.
+    ``בג"ץ 389/87`` → ``בגץ|389-87`` and ``ע"א 389/87`` → ``עא|389-87`` are
+    DISTINCT keys, even though both share docket ``389-87``. This is the safe
+    dedup key for ``missing_precedents`` (#143/#136): deduping on the bare number
+    alone (``case_number_from_citation``) would WRONGLY MERGE the same docket
+    across different courts (18 such collisions already exist in the corpus). A
+    prefix-less citation (bare נט-format / unknown court) yields ``|<number>``.
+    """
+    cit = classify(citation)
+    if not cit.case_number_norm:
+        return ""
+    return f"{_norm_designator(cit.court_prefix)}|{cit.case_number_norm}"
+
+
 def _split_filed(num_norm: str) -> tuple[str, str, str] | None:
     """Split a normalized NNNNN-MM-YY number into (file, month, year).
 

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

@@ -14,7 +14,7 @@ import asyncpg
 from pgvector.asyncpg import register_vector
 
 from legal_mcp import config
-from legal_mcp.services import halacha_quality
+from legal_mcp.services import court_citation, halacha_quality
 
 logger = logging.getLogger(__name__)
 
@@ -1571,6 +1571,19 @@ SCHEMA_V39_SQL = """
 ALTER TABLE case_law ADD COLUMN IF NOT EXISTS parties TEXT DEFAULT '';
 """
 
+# V40 (#143): missing_precedents gains a designator-aware dedup key
+# (citation_norm) and a provenance tag (discovery_source: manual | cited_only |
+# digest | writer). No UNIQUE constraint — the corpus legitimately holds the same
+# docket across different courts (e.g. בג"ץ 389/87 vs ע"א 389/87), so uniqueness
+# is enforced designator-aware in the create path, not by a DB constraint that
+# would over-merge distinct precedents.
+SCHEMA_V40_SQL = """
+ALTER TABLE missing_precedents ADD COLUMN IF NOT EXISTS citation_norm TEXT DEFAULT '';
+ALTER TABLE missing_precedents ADD COLUMN IF NOT EXISTS discovery_source TEXT DEFAULT 'manual';
+CREATE INDEX IF NOT EXISTS idx_missing_precedents_citation_norm
+    ON missing_precedents(citation_norm);
+"""
+
 
 # Stable, arbitrary key for the session-level advisory lock that serialises
 # schema DDL across processes. Every short-lived process (cron drains, services)
@@ -1589,7 +1602,7 @@ async def _run_schema_migrations(pool: asyncpg.Pool) -> None:
             await _apply_schema_ddl(conn)
         finally:
             await conn.execute("SELECT pg_advisory_unlock($1)", _MIGRATION_LOCK_KEY)
-    logger.info("Database schema initialized (v1-v38)")
+    logger.info("Database schema initialized (v1-v40)")
 
 
 async def _apply_schema_ddl(conn: asyncpg.Connection) -> None:
@@ -1633,6 +1646,7 @@ async def _apply_schema_ddl(conn: asyncpg.Connection) -> None:
         await conn.execute(SCHEMA_V37_SQL)
         await conn.execute(SCHEMA_V38_SQL)
         await conn.execute(SCHEMA_V39_SQL)
+        await conn.execute(SCHEMA_V40_SQL)
 
 
 async def init_schema() -> None:
@@ -7259,27 +7273,40 @@ async def create_missing_precedent(
     legal_issue: str | None = None,
     claim_quote: str | None = None,
     notes: str | None = None,
+    discovery_source: str = "manual",
+    linked_case_law_id: UUID | None = None,
 ) -> dict:
-    """Create a new missing-precedent row (status='open' by default)."""
+    """Create a new missing-precedent row (status='open' by default).
+
+    ``citation_norm`` is computed at write-time (designator-aware, #143) so
+    duplicate gaps for the same court+docket dedup correctly without merging the
+    same docket across different courts. ``discovery_source`` records how the gap
+    surfaced (manual | cited_only | digest | writer); ``linked_case_law_id`` may
+    be set up-front when the canonical identity is already known (a cited_only
+    stub awaiting its text) while ``status`` stays 'open' until the text lands.
+    """
     if not citation.strip():
         raise ValueError("citation is required")
     if cited_by_party and cited_by_party not in ALLOWED_MP_PARTIES:
         raise ValueError(
             f"cited_by_party must be one of {sorted(ALLOWED_MP_PARTIES)}"
         )
+    citation_norm = court_citation.citation_dedup_key(citation)
     pool = await get_pool()
     async with pool.acquire() as conn:
         row = await conn.fetchrow(
             """INSERT INTO missing_precedents (
                 citation, case_name, cited_in_case_id, cited_in_document_id,
                 cited_by_party, cited_by_party_name, legal_topic, legal_issue,
-                claim_quote, notes
+                claim_quote, notes, citation_norm, discovery_source,
+                linked_case_law_id
             )
-            VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10)
+            VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13)
             RETURNING *""",
             citation.strip(), case_name, cited_in_case_id, cited_in_document_id,
             cited_by_party, cited_by_party_name, legal_topic, legal_issue,
-            claim_quote, notes,
+            claim_quote, notes, citation_norm, discovery_source,
+            linked_case_law_id,
         )
     return _row_to_missing_precedent(row)
 
@@ -7443,20 +7470,31 @@ async def find_missing_precedent_by_citation(
     citation: str,
     case_id: UUID | None = None,
 ) -> dict | None:
-    """Look up an existing row by citation string (exact match) and optionally
-    cited-in case_id. Used to deduplicate auto-creation by the researcher."""
+    """Look up an existing gap by DESIGNATOR-AWARE normalized citation and
+    optionally cited-in case_id. Used to deduplicate auto-creation (researcher,
+    cited_only derivation #143, digest derivation #136).
+
+    Matches on ``citation_norm`` (``court|docket``) so formatting variants of the
+    same court+docket collapse to one gap, while the same docket across different
+    courts stays distinct. Falls back to an exact ``citation`` string match when
+    the citation carries no parseable number (citation_norm is empty)."""
+    norm = court_citation.citation_dedup_key(citation)
     pool = await get_pool()
     async with pool.acquire() as conn:
+        if norm:
+            match_col, match_val = "citation_norm", norm
+        else:
+            match_col, match_val = "citation", citation.strip()
         if case_id is not None:
             row = await conn.fetchrow(
-                "SELECT * FROM missing_precedents "
-                "WHERE citation = $1 AND cited_in_case_id = $2 LIMIT 1",
-                citation.strip(), case_id,
+                f"SELECT * FROM missing_precedents "
+                f"WHERE {match_col} = $1 AND cited_in_case_id = $2 LIMIT 1",
+                match_val, case_id,
             )
         else:
             row = await conn.fetchrow(
-                "SELECT * FROM missing_precedents WHERE citation = $1 LIMIT 1",
-                citation.strip(),
+                f"SELECT * FROM missing_precedents WHERE {match_col} = $1 LIMIT 1",
+                match_val,
             )
     return _row_to_missing_precedent(row) if row else None
 

mcp-server/tests/test_court_citation.py

@@ -4,11 +4,29 @@ from __future__ import annotations
 
 from legal_mcp.services.court_citation import (
     case_number_from_citation,
+    citation_dedup_key,
     classify,
     normalize_case_number,
 )
 
 
+def test_dedup_key_is_designator_aware():
+    """#143 — same docket across DIFFERENT courts must NOT collapse to one key
+    (deduping on the bare number would wrongly merge distinct precedents)."""
+    bagatz = citation_dedup_key('בג"ץ 389/87')
+    civil = citation_dedup_key('ע"א 389/87')
+    assert bagatz and civil and bagatz != civil, (bagatz, civil)
+    # ...while gershayim/format variants of the SAME court+docket DO collapse.
+    assert citation_dedup_key('עע"מ 9057/09') == citation_dedup_key("עע״מ 9057-09")
+    # committee proceeding-type is part of the key (ערר ≠ בל"מ).
+    assert citation_dedup_key("ערר 1192/18") != citation_dedup_key('בל"מ 1192/18')
+
+
+def test_dedup_key_empty_without_number():
+    assert citation_dedup_key("") == ""
+    assert citation_dedup_key("פסק דין בלי מספר") == ""
+
+
 def test_admin_filed_format_the_example():
     """The plan's example: עת"מ 46111-12-22 → admin, parsed into (46111,12,22)."""
     c = classify('עת"מ 46111-12-22 יכין-אפק בע"מ נ\' הוועדה המחוזית')