fix: 批量/手动匹配排除同一试验已匹配患者，防止重复记录

matching_service.py:
- run_matching 改为 upsert：同一 (patient_id, trial_id) 存在记录时
  直接更新 LLM 结果，不再插入新行，彻底消除重复记录
- 仅首次匹配时向医生发送通知，重新匹配不重复推送

batch_matching_service.py:
- existing_pairs 查询从全表扫描改为按目标试验过滤，提升性能
- 只查 MatchingResult 的 patient_id/trial_id 列（select 子集）
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/app/services/batch_matching_service.py

 import asyncio
 from datetime import datetime
-from sqlalchemy import select
+from sqlalchemy import select, tuple_
 from app.database import AsyncSessionLocal
 from app.models.patient import Patient
 from app.models.trial import Trial, TrialStatus
@@ -44,12 +44,14 @@ async def run_batch_matching(job_id: str) -> None:
                 )
                 trials = trial_result.scalars().all()
 
-            # Build set of already-matched pairs to skip
-            existing_result = await db.execute(select(MatchingResult))
-            existing_pairs = {
-                (r.patient_id, r.trial_id)
-                for r in existing_result.scalars().all()
-            }
+            # Build set of already-matched pairs to skip（只查目标试验，避免全表扫描）
+            trial_ids = [t.id for t in trials]
+            existing_result = await db.execute(
+                select(MatchingResult.patient_id, MatchingResult.trial_id).where(
+                    MatchingResult.trial_id.in_(trial_ids)
+                )
+            )
+            existing_pairs = {(row.patient_id, row.trial_id) for row in existing_result.all()}
 
             # Build work list (skip already matched pairs)
             pairs = [

backend/app/services/matching_service.py

@@ -117,19 +117,41 @@ async def run_matching(
     }
     match_status = status_map.get(llm_result.get("overall_status", ""), MatchStatus.pending_review)
 
-    matching_result = MatchingResult(
-        id=str(uuid.uuid4()),
-        patient_id=patient_id,
-        trial_id=trial_id,
-        status=match_status,
-        overall_score=llm_result.get("overall_score", 0.0),
-        criterion_details=llm_result.get("criteria_results", []),
-        evidence_spans=evidence_spans,
-        llm_model_used=settings.llm_model,
+    # Upsert: 若该患者已对该试验有匹配记录，则更新而非新增
+    existing_stmt = select(MatchingResult).where(
+        MatchingResult.patient_id == patient_id,
+        MatchingResult.trial_id == trial_id,
     )
-    db.add(matching_result)
+    matching_result = (await db.execute(existing_stmt)).scalar_one_or_none()
+    is_new = matching_result is None
+
+    if matching_result:
+        matching_result.status = match_status
+        matching_result.overall_score = llm_result.get("overall_score", 0.0)
+        matching_result.criterion_details = llm_result.get("criteria_results", [])
+        matching_result.evidence_spans = evidence_spans
+        matching_result.llm_model_used = settings.llm_model
+        matching_result.matched_at = datetime.utcnow()
+    else:
+        matching_result = MatchingResult(
+            id=str(uuid.uuid4()),
+            patient_id=patient_id,
+            trial_id=trial_id,
+            status=match_status,
+            overall_score=llm_result.get("overall_score", 0.0),
+            criterion_details=llm_result.get("criteria_results", []),
+            evidence_spans=evidence_spans,
+            llm_model_used=settings.llm_model,
+        )
+        db.add(matching_result)
     await db.flush()
 
+    # 仅首次匹配时向医生发送通知，避免重复推送
+    if not is_new:
+        await db.commit()
+        await db.refresh(matching_result)
+        return matching_result
+
     # Determine recipient doctors by role
     user_result = await db.execute(
         select(User).options(selectinload(User.roles)).where(User.is_active == True)  # noqa: E712