feat(regulatory): 对齐附件4目录核查规则

tests/fixtures/regulatory/attachment4_outline.json

@@ -0,0 +1,8 @@
+[
+  {"code": "1", "title": "监管信息", "children": ["章节目录", "申请表", "术语/缩写词列表", "产品列表", "关联文件", "申报前与监管机构的联系情况和沟通记录", "符合性声明"]},
+  {"code": "2", "title": "综述资料", "children": ["章节目录", "概述", "产品描述", "预期用途", "申报产品上市历史", "其他需说明的内容"]},
+  {"code": "3", "title": "非临床资料", "children": ["章节目录", "产品风险管理资料", "体外诊断试剂安全和性能基本原则清单", "产品技术要求及检验报告", "分析性能研究", "稳定性研究", "阳性判断值或参考区间研究", "其他资料"]},
+  {"code": "4", "title": "临床评价资料", "children": ["章节目录", "临床评价资料"]},
+  {"code": "5", "title": "产品说明书和标签样稿", "children": ["章节目录", "产品说明书", "标签样稿", "其他资料"]},
+  {"code": "6", "title": "质量管理体系文件", "children": ["综述", "章节目录", "生产制造信息", "质量管理体系程序", "管理职责程序", "资源管理程序", "产品实现程序", "质量管理体系的测量/分析和改进程序", "其他质量体系程序信息", "质量管理体系核查文件"]}
+]

tests/test_regulatory_completeness.py

@@ -42,3 +42,30 @@ def test_completeness_check_matches_existing_files_and_reports_missing(django_us
     missing = next(finding for finding in findings if finding.rule_code == "registration_test_report")
     assert missing.severity == "blocking"
     assert missing.category == "completeness"
+
+
+def test_completeness_check_matches_attachment4_directory_names(django_user_model):
+    user = django_user_model.objects.create_user(username="owner", password="pass")
+    conversation = Conversation.objects.create(user=user, title="会话")
+    batch = FileSummaryBatch.objects.create(
+        conversation=conversation,
+        user=user,
+        batch_no="FS-A4",
+        status=FileSummaryBatch.Status.SUCCESS,
+    )
+    FileSummaryItem.objects.create(
+        batch=batch,
+        file_index=1,
+        directory_level="1. 监管信息 / 1.2 申请表",
+        file_name="注册申请表.pdf",
+        file_type="pdf",
+        relative_path="1.监管信息/1.2申请表/注册申请表.pdf",
+        storage_path="x/app.pdf",
+    )
+
+    findings = run_completeness_check(batch, load_rule_file())
+
+    assert not any(finding.rule_code == "attachment4_1_2_application_form" for finding in findings)
+    missing_qms = next(finding for finding in findings if finding.rule_code == "attachment4_6_quality_system")
+    assert missing_qms.severity == "high"
+    assert missing_qms.evidence["searched_fields"] == ["file_name", "relative_path", "directory_level"]

tests/test_regulatory_consistency.py

@@ -12,3 +12,16 @@ def test_consistency_check_reports_product_name_mismatch():
     assert len(findings) == 1
     assert findings[0].category == "consistency"
     assert "产品名称" in findings[0].title
+
+
+def test_consistency_check_reports_registration_scope_fields():
+    document_texts = {
+        "申请表.docx": "管理类别：第二类\n分类编码：6840\n注册类型：首次注册\n临床评价路径：免临床",
+        "综述资料.docx": "管理类别：第三类\n分类编码：6840\n注册类型：首次注册\n临床评价路径：临床试验",
+    }
+
+    findings = run_consistency_check(document_texts)
+    titles = [finding.title for finding in findings]
+
+    assert "管理类别在不同文件中不一致" in titles
+    assert "临床评价路径在不同文件中不一致" in titles

tests/test_regulatory_rag.py

@@ -6,6 +6,7 @@ from review_agent.regulatory_review.services.rag_citation import (
 )
 from review_agent.regulatory_review.services.rag_embedding import SiliconFlowEmbeddingProvider
 from review_agent.regulatory_review.services.rag_index import chunk_text
+from review_agent.regulatory_review.services.rag_index import collect_source_chunks
 
 
 def test_siliconflow_embedding_provider_posts_expected_payload(monkeypatch):
@@ -70,3 +71,18 @@ def test_retrieve_citations_raises_when_index_missing(settings, tmp_path):
 
     with pytest.raises(RagIndexUnavailable):
         retrieve_citations("注册检验报告", embedding_provider=lambda texts: [[0.1]])
+
+
+def test_collect_source_chunks_requires_attachment4_extraction(monkeypatch, tmp_path):
+    source_dir = tmp_path / "sources"
+    source_dir.mkdir()
+    attachment4 = source_dir / "附件 4 体外诊断试剂注册申报资料要求及说明.doc"
+    attachment4.write_bytes(b"legacy-doc")
+
+    def fail_extract(path):
+        raise RuntimeError("无法通过 LibreOffice 转换法规 .doc 材料")
+
+    monkeypatch.setattr("review_agent.regulatory_review.services.rag_index.extract_text_from_path", fail_extract)
+
+    with pytest.raises(RuntimeError, match="附件 4"):
+        collect_source_chunks(source_dir)

tests/test_regulatory_rule_loader.py

@@ -1,4 +1,5 @@
 from pathlib import Path
+import json
 
 import pytest
 from django.core.management import call_command
@@ -27,6 +28,30 @@ def test_load_rule_file_reads_demo_requirements():
     assert "essential_principles_checklist" in codes
 
 
+def test_load_rule_file_covers_attachment4_outline():
+    rule_set = load_rule_file()
+    requirements = rule_set["requirements"]
+    outline = json.loads(Path("tests/fixtures/regulatory/attachment4_outline.json").read_text(encoding="utf-8"))
+
+    for chapter in outline:
+        chapter_rule = next(
+            item for item in requirements if item["title"] == chapter["title"] and item.get("attachment4_code") == chapter["code"]
+        )
+        assert chapter_rule["attachment4_code"] == chapter["code"]
+        assert chapter_rule["severity"] == "high"
+        assert chapter_rule["citation_query"]
+        for child in chapter["children"]:
+            child_rule = next(
+                item
+                for item in requirements
+                if item["title"] == child and str(item.get("attachment4_code", "")).startswith(f"{chapter['code']}.")
+            )
+            assert child_rule["rule_id"]
+            assert child_rule["file_keywords"]
+            assert child_rule["severity"] in {"blocking", "high", "medium"}
+            assert child_rule["citation_query"]
+
+
 def test_compute_file_sha256_changes_when_file_changes(tmp_path):
     path = tmp_path / "rule.yaml"
     path.write_text("code: demo\n", encoding="utf-8")

tests/test_regulatory_structure.py

@@ -11,3 +11,15 @@ def test_structure_check_reports_missing_instruction_sections():
 
     assert any(finding.rule_code == "instructions_for_use:储存条件" for finding in findings)
     assert all("样本要求" not in finding.title for finding in findings)
+
+
+def test_structure_check_reports_missing_attachment4_outline_heading():
+    document_texts = {
+        "申报资料目录.txt": "1. 监管信息\n1.2 申请表\n2. 综述资料\n3. 非临床资料\n"
+    }
+
+    findings = run_structure_check(document_texts, load_rule_file())
+
+    missing = next(finding for finding in findings if finding.rule_code == "attachment4_4_clinical_evaluation")
+    assert missing.category == "structure"
+    assert missing.evidence["expected_title"] == "临床评价资料"

tests/test_regulatory_workflow.py

@@ -7,6 +7,7 @@ from review_agent.models import (
     FileSummaryItem,
     Message,
     RegulatoryIssue,
+    RegulatoryArtifact,
     RegulatoryReviewBatch,
     WorkflowEvent,
     WorkflowNodeRun,
@@ -201,4 +202,6 @@ def test_workflow_generates_issues_exports_and_assistant_summary(settings, tmp_p
         workflow_type="regulatory_review",
         workflow_batch_id=batch.pk,
     ).count() == 3
+    assert RegulatoryArtifact.objects.filter(batch=batch, name="text_extract_status.json").exists()
+    assert RegulatoryArtifact.objects.filter(batch=batch, name="rag_result_json.json").exists()
     assert conversation.messages.filter(role=Message.Role.ASSISTANT, content__contains="已完成 NMPA").exists()