from __future__ import annotations import json import sys from pathlib import Path REPO_ROOT = Path(__file__).resolve().parents[1] if str(REPO_ROOT) not in sys.path: sys.path.insert(0, str(REPO_ROOT)) from scripts.runtime.roles import resolve_runtime_profile from scripts.runtime.sources import append_packet_sources, rebuild_sources_from_packets from scripts.runtime.tasks import TaskCard from scripts.runtime.workers import PacketWorker, build_material_context, build_route_query, build_search_context, normalize_packet_against_context class FakeSearchProvider: def search(self, *, query: str, route: str, num_results: int): return [ { "title": f"{route} result for {query}", "url": f"https://example.com/{route}", "snippet": "候选证据摘要", "route": route, } ][:num_results] class FakeClient: def __init__(self, response: dict) -> None: self.response = response self.calls: list[dict[str, object]] = [] def chat_complete(self, **kwargs) -> str: self.calls.append(kwargs) return json.dumps(self.response, ensure_ascii=False) def sample_card() -> TaskCard: return TaskCard( task_id="ch01-literature", chapter_ids=["ch01"], topic_axis="literature", questions=["围绕临床证据提炼结论。"], search_routes=["scholar", "general"], output_packet="phase2/packets/ch01-literature.json", ) def test_build_search_context_assigns_stable_source_ids() -> None: context = build_search_context(sample_card(), FakeSearchProvider(), num_results_per_route=1) assert [source["id"] for source in context["candidate_sources"]] == [ "src_ch01_literature_001", "src_ch01_literature_002", ] assert context["routes_used"] == ["scholar", "general"] def test_fda_route_query_uses_english_axis_terms_not_chinese_title() -> None: card = TaskCard( task_id="ch10-fda_enforcement_precedents", chapter_ids=["ch10"], topic_axis="fda_enforcement_precedents", questions=["立即纠偏、体系补强、能力建设三层整改路线图必须绑定 owner、关闭证据和复核机制"], search_routes=["fda"], output_packet="phase2/packets/ch10-fda_enforcement_precedents.json", chapter_title="立即纠偏、体系补强、能力建设三层整改路线图必须绑定 owner、关闭证据和复核机制", ) query = build_route_query(card, "fda") assert "立即纠偏" not in query assert "CAPA" in query assert "remediation" in query assert "verification evidence" in query def test_integrated_scholar_query_does_not_leak_internal_axis_or_cjk_punctuation() -> None: card = TaskCard( task_id="ch07-chapter_integrated", chapter_ids=["ch07"], topic_axis="chapter_integrated", questions=["人员能力：培训有效性比培训记录更关键"], search_routes=["scholar"], output_packet="phase2/packets/ch07-chapter_integrated.json", chapter_title="人员能力：培训有效性比培训记录更关键", ) query = build_route_query(card, "scholar") assert "chapter_integrated" not in query assert "、" not in query assert " " not in query assert "training" in query assert "quality" in query assert not any("\u4e00" <= char <= "\u9fff" for char in query) def test_evidence_route_query_is_short_english_candidate_evidence_query() -> None: card = TaskCard( task_id="ch08-chapter_integrated", chapter_ids=["ch08"], topic_axis="chapter_integrated", questions=["运营管理需要建立跨部门节奏、问题升级、指标看板和管理层 review"], search_routes=["evidence"], output_packet="phase2/packets/ch08-chapter_integrated.json", chapter_title="运营管理需要建立跨部门节奏、问题升级、指标看板和管理层 review", ) query = build_route_query(card, "evidence") assert "evidence" in query assert "quality" in query assert "operations" in query assert "运营管理" not in query assert not any("\u4e00" <= char <= "\u9fff" for char in query) def test_packet_worker_includes_search_context_in_prompt() -> None: context = build_search_context(sample_card(), FakeSearchProvider(), num_results_per_route=1) response = { "task_id": "ch01-literature", "claims": [{"claim": "候选证据支持判断", "source_ids": ["src_ch01_literature_001"]}], "evidence_items": [{"source_id": "src_ch01_literature_001", "summary": "摘要"}], "counter_evidence": [{"claim": "仍需更多数据", "source_ids": ["src_ch01_literature_002"]}], "source_ids": ["src_ch01_literature_001", "src_ch01_literature_002"], "sources": context["candidate_sources"], "source_quality_notes": ["候选来源需要后续评级"], "open_questions": [], "raw_quotes_or_notes": [], } role = resolve_runtime_profile(profile="medium").role_for_task("evidence_packet") fake = FakeClient(response) packet = PacketWorker(role=role, client=fake, search_provider=FakeSearchProvider()).run(sample_card()) assert packet["sources"][0]["url"] == "https://example.com/scholar" assert "candidate_sources" in fake.calls[0]["user"] def test_normalize_packet_fills_source_ids_and_sources_from_context() -> None: context = { "candidate_sources": [ {"id": "src_a", "title": "A", "url": "https://example.com/a", "tier": "Tier 2", "score": 7} ] } packet = { "task_id": "ch01", "claims": [{"claim": "判断", "source_ids": ["src_a"]}], "evidence_items": [{"source_id": "src_a", "summary": "证据"}], "counter_evidence": [{"claim": "反方", "source_ids": ["src_a"]}], "source_quality_notes": [], "open_questions": [], "raw_quotes_or_notes": [], } normalized = normalize_packet_against_context(packet, context, None) assert normalized["source_ids"] == ["src_a"] assert normalized["sources"] == context["candidate_sources"] def test_material_context_is_loaded_and_allowed_as_source(tmp_path: Path) -> None: project = tmp_path / "project" material = project / "phase0/extracted/audit.md" material.parent.mkdir(parents=True) material.write_text("白帆现场发现：偏差调查未闭环。", encoding="utf-8") card = TaskCard( task_id="ch01-chapter_integrated", chapter_ids=["ch01"], topic_axis="chapter_integrated", questions=["q"], search_routes=[], output_packet="phase2/packets/ch01-chapter_integrated.json", allowed_materials=["phase0/extracted/audit.md"], ) context = build_material_context(card, project) response = { "task_id": "ch01-chapter_integrated", "claims": [{"claim": "现场材料显示偏差调查需要补强", "source_ids": [context["materials"][0]["source_id"]]}], "evidence_items": [{"source_id": context["materials"][0]["source_id"], "summary": "偏差调查未闭环。"}], "counter_evidence": [{"claim": "需与完整审计报告交叉确认", "source_ids": [context["materials"][0]["source_id"]]}], "source_ids": [context["materials"][0]["source_id"]], "sources": [ { "id": context["materials"][0]["source_id"], "title": "audit.md", "url": "phase0/extracted/audit.md", "tier": "local_material", } ], "source_quality_notes": ["本地材料作为起点证据"], "open_questions": [], "raw_quotes_or_notes": ["白帆现场发现：偏差调查未闭环。"], } fake = FakeClient(response) role = resolve_runtime_profile(profile="medium").role_for_task("evidence_packet") packet = PacketWorker(role=role, client=fake, project_root=project).run(card) assert packet["source_ids"] == [context["materials"][0]["source_id"]] assert "白帆现场发现" in fake.calls[0]["user"] def test_append_packet_sources_preserves_distinct_source_ids_for_same_url(tmp_path: Path) -> None: packet = { "sources": [ {"id": "src_a", "title": "A", "url": "https://example.com/a", "tier": "Tier 2", "score": 7}, {"id": "src_b", "title": "B", "url": "https://example.com/a", "tier": "Tier 2", "score": 7}, ] } written = append_packet_sources(tmp_path / "sources.jsonl", packet) assert written == 2 assert len((tmp_path / "sources.jsonl").read_text(encoding="utf-8").splitlines()) == 2 def test_rebuild_sources_from_packets_preserves_distinct_source_ids(tmp_path: Path) -> None: project = tmp_path / "project" packets = project / "phase2" / "packets" packets.mkdir(parents=True) packet = { "sources": [ {"id": "src_001", "title": "A", "url": "https://example.com/a"}, {"id": "src_002", "title": "A duplicate", "url": "https://example.com/a"}, {"id": "src_003", "title": "Local", "url": "phase0/extracted/local.md"}, ] } (packets / "ch01-a.json").write_text(json.dumps(packet, ensure_ascii=False), encoding="utf-8") count = rebuild_sources_from_packets(project) lines = (project / "phase2" / "sources.jsonl").read_text(encoding="utf-8").splitlines() assert count == 3 assert len(lines) == 3 assert "src_001" in lines[0] assert "src_002" in lines[1] assert "src_003" in lines[2] def test_rebuild_sources_preserves_cache_metadata(tmp_path: Path) -> None: project = tmp_path / "project" packets = project / "phase2" / "packets" packets.mkdir(parents=True) source = {"id": "src_001", "title": "A", "url": "https://example.com/a"} (packets / "ch01-a.json").write_text(json.dumps({"sources": [source]}, ensure_ascii=False), encoding="utf-8") (project / "phase2/sources.jsonl").write_text( json.dumps( { **source, "cached_text_path": "phase2/source_cache/md/src_001.md", "cache_status": "fetched", }, ensure_ascii=False, ) + "\n", encoding="utf-8", ) rebuild_sources_from_packets(project) row = json.loads((project / "phase2/sources.jsonl").read_text(encoding="utf-8")) assert row["cached_text_path"] == "phase2/source_cache/md/src_001.md" assert row["cache_status"] == "fetched"