fix: preserve concurrent teacher decisions when model scoring fails

backend/tests/test_optimization.py

@@ -373,6 +373,35 @@ def test_model_declining_score_records_original_gap(sqlite_db):
     assert score.evidence_gaps[0]["reason"] == "只有装置介绍"
 
 
+def test_failed_model_call_cannot_overwrite_a_concurrent_teacher_review(sqlite_db, monkeypatch):
+    from wisegrade import scoring
+    db = sqlite_db
+    item = project(db)
+    created = services.create_batch(SimpleNamespace(course_version_id=1, rubric_version_id=1), db)
+    batch = db.get(ScoringBatch, created["id"])
+    criterion = batch.rubric_snapshot["criteria"][0]
+    batch.rubric_snapshot = {**batch.rubric_snapshot, "criteria": [criterion]}
+    batch.status, batch.progress_total, batch.progress_done = "running", 1, 0
+    score = CriterionScore(batch_id=batch.id, project_id=item.id, criterion_id=criterion["criterion_id"], status="pending")
+    db.add(score)
+    db.commit()
+    factory = sessionmaker(bind=db.get_bind(), expire_on_commit=False)
+    def teacher_then_failure(*args, **kwargs):
+        with factory() as teacher_db:
+            services.review_score(score.id, SimpleNamespace(score=0, comment="教师核定零分", reason="已核对"), teacher_db)
+        raise ModelError("模型响应超时")
+    monkeypatch.setattr(scoring, "_project_facts", lambda *args, **kwargs: {})
+    monkeypatch.setattr(scoring, "_score_criterion", teacher_then_failure)
+    failures, reviews = scoring._run_project(batch.id, item.id, factory, lambda *args: None, FLAT, lambda *a, **k: [])
+    db.expire_all()
+    saved = db.get(CriterionScore, score.id)
+    assert saved.status == "reviewed"
+    assert saved.teacher_score == 0 and saved.teacher_comment == "教师核定零分"
+    assert saved.error == ""
+    assert failures == reviews == 0
+    assert db.get(ScoringBatch, batch.id).progress_done == 1
+
+
 def test_unrebuilt_course_document_does_not_complete_task(sqlite_db):
     db = sqlite_db
     doc = KnowledgeDocument(course_version_id=1, title="无页面资料", document_type="course",

backend/wisegrade/scoring.py

@@ -387,13 +387,17 @@ def _run_project(batch_id, project_id, session_factory, client_factory, embedder
             except Exception as exc:
                 db.rollback()
                 score = db.scalar(select(CriterionScore).where(CriterionScore.batch_id == batch.id,
-                                  CriterionScore.project_id == project.id, CriterionScore.criterion_id == criterion["criterion_id"]))
+                                  CriterionScore.project_id == project.id, CriterionScore.criterion_id == criterion["criterion_id"])
+                                  .with_for_update().execution_options(populate_existing=True))
                 if not score:
                     score = CriterionScore(batch_id=batch.id, project_id=project.id, criterion_id=criterion["criterion_id"])
                     db.add(score)
-                score.status, score.error = STATUS_FAILED, str(exc)[:500]
-                db.commit()
-                failures += 1
+                # A teacher may have committed a decision while the request was in
+                # flight. The exception path must honour it just like success does.
+                if score.status != STATUS_REVIEWED and score.teacher_score is None:
+                    score.status, score.error = STATUS_FAILED, str(exc)[:500]
+                    db.commit()
+                    failures += 1
             db.execute(update(ScoringBatch).where(ScoringBatch.id == batch_id).values(progress_done=ScoringBatch.progress_done + 1))
             db.commit()
     return failures, reviews