test: use PostgreSQL and live model integrations

backend/tests/test_workflow.py

@@ -4,37 +4,13 @@ from pathlib import Path
 import pymupdf as fitz
 import pytest
 from fastapi.testclient import TestClient
-from sqlalchemy import select
 
-from wisegrade.config import get_settings
 from wisegrade.db import get_db
-from wisegrade.llm import ConsistencyResult, FactResult, ScoreResult
 from wisegrade.main import app
-from wisegrade.models import CourseVersion, CriterionScore, RubricVersion, ScoringBatch, TeacherRevision
-from wisegrade.rag import retrieve
 from wisegrade.scoring import run_batch
 from wisegrade.storage import scan
 
 
-def fake_embed(texts):
-    return [[1.0] + [0.0] * 1023 for _ in texts]
-
-
-class FakeModel:
-    def json_call(self, _prompt, payload, schema, images=None):
-        if schema is FactResult:
-            return FactResult(topic="自动化实践示例")
-        if schema is ConsistencyResult:
-            return ConsistencyResult()
-        criterion = payload["criterion"]
-        evidence = payload["evidence"][0]
-        return ScoreResult(criterion_id=criterion["criterion_id"], score=criterion["max_score"] / 2,
-                           max_score=criterion["max_score"], reason="示例证据支持部分要求",
-                           evidence=[{"chunk_id": evidence["chunk_id"], "page": evidence["page"],
-                                      "quote": evidence["text"][:min(12, len(evidence["text"]))]}],
-                           missing_evidence=[], suggestion="补充量化论证", confidence=.7)
-
-
 def make_pdf(path, title):
     pdf = fitz.open()
     page = pdf.new_page()
@@ -70,14 +46,20 @@ def test_docs_examples_end_to_end(environment, session_factory):
         response = client.post("/api/scoring-batches", json={"course_version_id": 1, "rubric_version_id": 1})
         assert response.status_code == 200, response.text
         batch_id = response.json()["id"]
-        run_batch(batch_id, session_factory=session_factory, client_factory=lambda db, batch: FakeModel(), embedder=fake_embed,
-                  retriever=lambda db, **kwargs: retrieve(db, embedder=fake_embed, **kwargs))
+        run_batch(batch_id, session_factory=session_factory)
+        progress = client.get(f"/api/scoring-batches/{batch_id}/progress").json()
+        assert progress["status"] == "completed", progress
+        assert progress["progress_done"] == progress["progress_total"] == 22
         result = client.get(f"/api/projects/{project_id}/scores?batch_id={batch_id}").json()
-        assert result["totals"]["overall"] == 50
-        assert all(row["evidence"] and row["evidence"][0]["page"] >= 1 for row in result["rows"])
-        assert client.get(result["rows"][0]["evidence"][0]["image_url"]).content[:8] == b"\x89PNG\r\n\x1a\n"
+        assert result["totals"]["overall"] is not None
+        assert all(row["status"] == "scored" and
+                   (row["model_score"] == 0 or row["evidence"] and row["evidence"][0]["page"] >= 1)
+                   for row in result["rows"]), result["rows"]
+        evidence_row = next(row for row in result["rows"] if row["evidence"] and row["evidence"][0]["image_url"])
+        assert client.get(evidence_row["evidence"][0]["image_url"]).content[:8] == b"\x89PNG\r\n\x1a\n"
         score_id = result["rows"][0]["score_id"]
-        reviewed = client.put(f"/api/criterion-scores/{score_id}/review", json={"score": 20, "comment": "教师核定", "reason": "核查原文"})
+        teacher_score = 0 if result["rows"][0]["model_score"] != 0 else 1
+        reviewed = client.put(f"/api/criterion-scores/{score_id}/review", json={"score": teacher_score, "comment": "教师核定", "reason": "核查原文"})
         assert reviewed.status_code == 200
         assert reviewed.json()["model_score"] != reviewed.json()["teacher_score"]
         assert len(client.get(f"/api/criterion-scores/{score_id}/revisions").json()) == 1