fix(learning): extract precedent metadata inline on final-decision enroll

When a chair's signed final decision is uploaded it is enrolled into the
precedent library, but _enroll_final_in_library only set deterministic
fields (citation, proceeding_type, date) and copied subject_tags from the
case's subject_categories — which is usually empty. The Gemini metadata
pass (subject_tags / summary / headnote / key_quote) was never triggered,
so the row sat at metadata_extraction_status='pending' with no subject
tags until a drain happened to pick it up. In practice a freshly-enrolled
final showed an empty "תגיות נושא" in the precedent edit UI.

Fix: after enrollment + citation, call the existing reextract_metadata
path inline (G2 — one path, full status lifecycle). It runs on Gemini
Flash over REST (GOOGLE_GEMINI_API_KEY, already in Coolify), so it is
container-safe — unlike the halacha path (claude CLI, host-only).
apply_to_record fills only empty fields, so the deterministic seeds and
any chair-curated subject_categories are preserved. Result surfaced in the
upload response (out["metadata"]); failures logged, upload still succeeds.

Also corrects two now-stale comments: the "Gemini returns no_metadata for
internal decisions" note in the enroll loop, and the "MCP-tool-only path"
docstring on reextract_metadata (true for halacha, not for Gemini REST).

Invariants: G1 (fill at source on enroll, not a read-time workaround),
G2 (reuse reextract_metadata — no parallel extraction path). LLM call is
Gemini REST, not claude_session, so the container LLM-call constraint holds.

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

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

@@ -383,7 +383,10 @@ async def reextract_metadata(
     appeal_subtype, and case_name when it equals the citation). User
     values are preserved.
 
-    **MCP-tool-only path** — same constraint as :func:`reextract_halachot`.
+    **Container-safe** — unlike :func:`reextract_halachot` (claude CLI, host-only),
+    metadata extraction runs on Gemini Flash over REST (GOOGLE_GEMINI_API_KEY), so
+    this path is callable from the FastAPI container too. The final-decision
+    enrollment loop (``_enroll_final_in_library``) calls it inline on upload.
     """
     from legal_mcp.services import precedent_metadata_extractor
 

web/app.py

@@ -3643,9 +3643,11 @@ async def _enroll_final_in_library(
         out["error"] = "no final text extracted"
         return out
 
-    # Deterministic metadata from the case record — the Gemini metadata extractor is
-    # tuned for EXTERNAL rulings and returns no_metadata for internal decisions, so we
-    # populate proceeding_type / date / tags / summary / citation ourselves (no LLM).
+    # Deterministic seeds from the case record first — proceeding_type / date /
+    # citation are identity, not inference, so we set them ourselves (no LLM) and the
+    # later Gemini pass (reextract_metadata, below) is fills-empty-only and won't
+    # touch them. subject_tags/summary seeded here from the case (chair-curated
+    # subject_categories win when present); when empty, Gemini fills them inline.
     district = "ירושלים"
     proceeding_type = (case.get("proceeding_type") or "ערר").strip()
     decision_date = case.get("decision_date") or case.get("hearing_date")
@@ -3685,6 +3687,26 @@ async def _enroll_final_in_library(
     except Exception as e:
         logger.warning("citation build failed for %s: %s", case_number, e)
 
+    # Fill the LLM-derived metadata (subject_tags, summary, headnote, key_quote)
+    # inline via Gemini (REST — container-safe; GOOGLE_GEMINI_API_KEY in Coolify),
+    # reusing the same reextract_metadata path the UI/drain use (G2 — one path, full
+    # status lifecycle). apply_to_record fills ONLY empty fields, so the deterministic
+    # seeds above (proceeding_type / date / citation, plus any chair-curated
+    # subject_categories) are preserved. Without this the row sat at
+    # metadata_extraction_status='pending' with empty tags until a drain happened to
+    # pick it up, so a freshly-enrolled final showed no subject tags in the edit UI.
+    # Best-effort — surfaced in the return value, never silently swallowed.
+    try:
+        from legal_mcp.services import precedent_library as plib_service
+        meta = await plib_service.reextract_metadata(UUID(case_law_id))
+        out["metadata"] = {
+            "status": meta.get("status"),
+            "fields": meta.get("fields") or [],
+        }
+    except Exception as e:
+        logger.warning("inline metadata extraction failed for %s: %s", case_number, e)
+        out["metadata_error"] = str(e)
+
     # The precedents this decision cites → link to the library; flag the ones not found.
     try:
         await cit_tools.extract_internal_citations(case_law_id=case_law_id, limit=0)