hack/tools/scripts/llm-key-hunter/results/_cache/content/chloeyouyou_acp-learning.json

{"repo": "chloeyouyou/acp-learning", "files": {"backend/run_tests.py": {"6bb8f6bbb21688466a43260d8010fc9f85ede89e": {"t": 1786068243.2854254, "n": 86304, "h": "26b9b3c956d2d6c2", "c": "\"\"\"后端核心逻辑测试（零依赖，直接 `python run_tests.py`）。\n\n锁住对错、防回归——重点覆盖踩过的隐蔽 bug：\n- 选雷\"补弱\"方向（曾经符号反了，推强不推弱）\n- 选雷\"防套路\"（不连续出同一题）\n- 质检双关闸门（雷会响 / 雷能排 / fix_check）\n- judge_fix 真判题、_error_family 解析、timeline 思维默认值聚合、sandbox。\n用真沙箱跑代码，几秒内完成。\n\"\"\"\n\nimport os\nimport re\nimport sys\nimport traceback\n\n# 主测试套件锁定 subprocess 后端：84 项用真沙箱，走 docker 会慢到几分钟。\n# 隔离性由文末「沙箱隔离」测试组单独打 docker 后端验证（docker 不可用则跳过）。\nos.environ.setdefault(\"ACP_SANDBOX\", \"subprocess\")\n# 单测不打真网：结算路径会触发先验蒸馏（真 LLM），此处关掉；\n# 先验层行为由 test_priors.py 用假 LLM 单独锁。\nos.environ.setdefault(\"ACP_PRIORS\", \"off\")\n\nsys.stdout.reconfigure(encoding=\"utf-8\")\n\nfrom sqlalchemy import create_engine\nfrom sqlalchemy.orm import sessionmaker\n\nfrom app import models  # noqa: F401  注册表到 Base.metadata\nfrom app.db import Base\nfrom app.models import (\n    CodeSnapshot, Event, ExecutionEvent, KnowledgeState, SessionMessage, Student, TutorSession,\n)\nfrom app.services import (\n    curriculum, event_engine, learning_path, mine_engine, pattern_validator, process, profile,\n    review, sandbox, teacher, timeline,\n)\n\n_engine = create_engine(\"sqlite:///:memory:\", connect_args={\"check_same_thread\": False})\nBase.metadata.create_all(_engine)\nTestSession = sessionmaker(bind=_engine, autoflush=False)\n\n_tests = []\ndef test(fn): _tests.append(fn); return fn\n\n\ndef _good_candidate() -> dict:\n    \"\"\"一道合规候选（buggy 真出 IndexError、fixed 跑出 2）。\"\"\"\n    buggy = \"def f(a):\\n    return a[3]\\nprint(f([1, 2]))\\n\"\n    fixed = \"def f(a):\\n    return a[-1]\\nprint(f([1, 2]))\\n\"\n    return {\n        \"id\": \"BP-TEST-001\", \"name\": \"测试题\", \"category\": \"boundary\", \"language\": \"python\",\n        \"knowledge_points\": [\"数组\"], \"capability_dims\": [\"Boundary_Awareness\"],\n        \"difficulty\": \"L1\", \"symptom\": \"RE\", \"symptom_sample\": \"IndexError: list index out of range\",\n        \"buggy_code\": buggy, \"fixed_code\": fixed,\n        \"mine_location\": {\"file\": \"main.py\", \"line\": 2}, \"trigger_input\": \"[1, 2]\",\n        \"root_cause\": \"下标 3 越界\", \"cognitive_root\": \"以为列表一定有第 4 个元素\",\n        \"thinking_pattern\": \"index_confusion\", \"expected_fix\": \"a[-1]\", \"fix_check\": r\"a\\[-1\\]\",\n        \"expected_output\": \"2\",\n        \"internalize_questions\": [\"为什么越界？\", \"合法下标范围是？\"],\n        \"hint_ladder\": {f\"L{i}\": f\"提示{i}\" for i in range(6)},\n    }\n\n\n# ---- 选雷：补弱（曾经符号反了，推强不推弱）----\n@test\ndef picker_选薄弱而非最强():\n    orig = profile.knowledge_mastery\n    profile.knowledge_mastery = lambda db, sid: [\n        {\"kp\": \"强项\", \"mastery\": \"熟练\", \"weak\": False},\n        {\"kp\": \"弱项\", \"mastery\": \"生疏\", \"weak\": True},\n    ]\n    try:\n        db = TestSession()\n        pats = [{\"id\": \"STRONG\", \"knowledge_points\": [\"强项\"], \"difficulty\": \"L1\"},\n                {\"id\": \"WEAK\", \"knowledge_points\": [\"弱项\"], \"difficulty\": \"L1\"}]\n        got = profile.get_weakest_pattern_id(db, \"s1\", pats)\n        assert got == \"WEAK\", f\"应推薄弱题 WEAK，实际 {got}（符号又反了？）\"\n        db.close()\n    finally:\n        profile.knowledge_mastery = orig\n\n\n# ---- 选雷：防套路（不连续出同一题）----\n@test\ndef picker_不连续出同一题():\n    orig = profile.knowledge_mastery\n    profile.knowledge_mastery = lambda db, sid: [\n        {\"kp\": \"弱项\", \"mastery\": \"生疏\", \"weak\": True},\n        {\"kp\": \"次弱\", \"mastery\": \"在学\", \"weak\": False},\n    ]\n    try:\n        db = TestSession()\n        db.add(TutorSession(id=\"sess_x\", student_id=\"s2\", pattern_id=\"WEAK\", manifest={}, history=[]))\n        db.commit()\n        pats = [{\"id\": \"WEAK\", \"knowledge_points\": [\"弱项\"], \"difficulty\": \"L1\"},\n                {\"id\": \"SECOND\", \"knowledge_points\": [\"次弱\"], \"difficulty\": \"L1\"}]\n        got = profile.get_weakest_pattern_id(db, \"s2\", pats)\n        assert got != \"WEAK\", \"上次刚做 WEAK，不该又出 WEAK\"\n        db.close()\n    finally:\n        profile.knowledge_mastery = orig\n\n\n# ---- 质检双关 ----\n@test\ndef 质检_合规候选全过():\n    problems = pattern_validator.validate_candidate(_good_candidate())\n    assert not problems, f\"合规候选不该有问题：{problems}\"\n\n@test\ndef 质检_雷不响要拦():\n    c = _good_candidate()\n    c[\"buggy_code\"] = \"x = 1\\nprint(2)\\n\"  # 两行(line:2 仍合法)、不报错，但声明 RE\n    problems = pattern_validator.validate_candidate(c)\n    assert any(\"雷没响\" in p for p in problems), f\"雷不响应被拦：{problems}\"\n\n@test\ndef 质检_fixcheck匹配带雷码要拦():\n    c = _good_candidate()\n    c[\"fix_check\"] = r\"def f\"             # 也匹配 buggy_code\n    problems = pattern_validator.validate_candidate(c)\n    assert any(\"buggy_code\" in p for p in problems), f\"应拦原样可过：{problems}\"\n\n@test\ndef 质检_跑题要拦():\n    # 订单要 arithmetic/WA，候选交了 boundary/RE → 对单不通过\n    c = _good_candidate()  # category=boundary, symptom=RE\n    task = {\"category\": \"arithmetic\", \"thinking_pattern\": \"control_semantics\", \"error_target\": \"WA\"}\n    problems = pattern_validator.validate_candidate(c, task)\n    assert any(\"跑题\" in p for p in problems), f\"对不上订单应被拦：{problems}\"\n    # 对得上订单时不因对单报错\n    task_ok = {\"category\": \"boundary\", \"thinking_pattern\": \"index_confusion\", \"error_target\": \"RE\"}\n    assert not pattern_validator.validate_candidate(c, task_ok), \"对得上订单不该被拦\"\n\n@test\ndef 质检_缺字段要拦():\n    c = _good_candidate(); del c[\"cognitive_root\"]\n    problems = pattern_validator.validate_candidate(c)\n    assert any(\"cognitive_root\" in p for p in problems), problems\n\n\n# ---- judge_fix 真判题 ----\n@test\ndef judge_带雷码不通过_修对通过():\n    bad = mine_engine.judge_fix(\"BP-BOUNDARY-001\", mine_engine.get_pattern(\"BP-BOUNDARY-001\")[\"buggy_code\"])\n    assert not bad[\"passed\"], \"带雷代码不该通过\"\n    fixed = mine_engine.get_pattern(\"BP-BOUNDARY-001\")[\"buggy_code\"].replace(\"range(len(arr) + 1)\", \"range(len(arr))\")\n    good = mine_engine.judge_fix(\"BP-BOUNDARY-001\", fixed)\n    assert good[\"passed\"], f\"正确修复应通过：{good}\"\n\n\n# ---- 事件：错误族解析 ----\n@test\ndef error_family_解析():\n    assert event_engine._error_family(\"RE\", \"Traceback...\\nIndexError: x\") == \"IndexError\"\n    assert event_engine._error_family(\"HANG\", \"\") == \"Timeout\"\n    assert event_engine._error_family(\"WA\", \"\") == \"WrongAnswer\"\n    assert event_engine._error_family(\"OK\", \"\") is None\n\n\n# ---- timeline：观察卡解析 + 思维默认值聚合 ----\n@test\ndef timeline_观察卡解析():\n    obs, guess = timeline._parse_observation(\n        [{\"role\": \"user\", \"content\": \"【观察记录】\\n我观察到：第一行少了\\n我的猜测：循环错\"}])\n    assert obs == \"第一行少了\" and guess == \"循环错\", (obs, guess)\n\n@test\ndef timeline_跨题思维默认值聚合():\n    eps = [{\"pattern_id\": \"BP-NULL-002\", \"pattern_name\": \"方法返回None未检查\"},\n           {\"pattern_id\": \"BP-NULL-003\", \"pattern_name\": \"字典取值未判空\"}]  # 同 assume_valid 簇\n    res = timeline.aggregate_thinking_patterns(eps)\n    assert res.get(\"enough\") and any(it[\"id\"] == \"assume_valid\" for it in res[\"items\"]), res\n\n\n# ---- 题库统计（曾经只吃 list、传 dict 报错）----\n@test\ndef collect_stats_兼容dict():\n    pats = mine_engine.load_patterns()  # dict[id->题]\n    s = pattern_validator.collect_stats(pats)\n    assert s[\"total\"] == len(pats) and sum(s[\"category\"].values()) == s[\"total\"], s\n\n\n# ---- 复习队列（间隔重复）----\nimport uuid as _uuid\nfrom datetime import date as _date, datetime as _dt, timedelta as _td\n\n_REVIEW_PID = \"BP-BOUNDARY-001\"  # 用真实题，name/category 能填上\n\ndef _seed_solved(db, student_id, pid, *, state, solved_days_ago: list[int]):\n    \"\"\"造一道「已解决/已内化」题 + 若干天前的 submit/OK 执行事实。\"\"\"\n    db.add(KnowledgeState(student_id=student_id, pattern_id=pid,\n                          knowledge_points=[\"数组\"], state=state))\n    base = _dt.now()\n    for d in solved_days_ago:\n        db.add(ExecutionEvent(\n            id=f\"ex_{_uuid.uuid4().hex[:12]}\", version=\"v1\", student_id=student_id,\n            session_id=f\"sess_{_uuid.uuid4().hex[:8]}\", pattern_id=pid,\n            source=\"submit\", kind=\"OK\", knowledge_points=[\"数组\"], meta={},\n            timestamp=(base - _td(days=d)).isoformat()))\n    db.commit()\n\n@test\ndef review_旧题到期进队列():\n    db = TestSession()\n    sid = \"rev_due\"\n    _seed_solved(db, sid, _REVIEW_PID, state=\"已解决\", solved_days_ago=[10])  # round1 间隔2，10>=2\n    due = review.due_queue(db, sid)\n    assert any(it[\"pattern_id\"] == _REVIEW_PID for it in due), f\"10天前解决的应到期：{due}\"\n    it = next(it for it in due if it[\"pattern_id\"] == _REVIEW_PID)\n    assert it[\"due\"] and it[\"name\"] and it[\"category\"], it\n    db.close()\n\n@test\ndef review_刚解决不到期():\n    db = TestSession()\n    sid = \"rev_fresh\"\n    _seed_solved(db, sid, _REVIEW_PID, state=\"已解决\", solved_days_ago=[0])  # 今天解决，0<2\n    due = review.due_queue(db, sid)\n    assert not any(it[\"pattern_id\"] == _REVIEW_PID for it in due), f\"刚解决不该到期：{due}\"\n    db.close()\n\n@test\ndef review_间隔随轮次升档():\n    # 已解决 round1=2天、round2=4天；已内化整体更长\n    assert review._interval_for(\"已解决\", 1) == 2\n    assert review._interval_for(\"已解决\", 2) == 4\n    assert review._interval_for(\"已内化\", 1) > review._interval_for(\"已解决\", 1)\n    # 多次解决（不同天）→ 轮次变高、间隔变长 → 同样 days_since 可能不到期\n    db = TestSession()\n    sid = \"rev_round\"\n    _seed_solved(db, sid, _REVIEW_PID, state=\"已内化\", solved_days_ago=[40, 20, 5])  # round3\n    due = review.due_queue(db, sid)\n    it = [x for x in due if x[\"pattern_id\"] == _REVIEW_PID]\n    # 最近一次 5 天前，已内化 round3 间隔=15，5<15 → 不到期\n    assert not it, f\"round3 已内化 5天前不该到期：{due}\"\n    db.close()\n\n@test\ndef review_未解决题不进队列():\n    db = TestSession()\n    sid = \"rev_open\"\n    db.add(KnowledgeState(student_id=sid, pattern_id=_REVIEW_PID,\n                          knowledge_points=[\"数组\"], state=\"已接触\"))\n    db.commit()\n    assert not review.due_queue(db, sid), \"未解决（已接触）不该进复习队列\"\n    db.close()\n\n\n# ---- 知返反幻觉：运行结果注入（截断 + 只留最新一条）----\nfrom app.services import tutor\n\n@test\ndef run_note_截断长输出():\n    note = tutor.run_result_note(\"OK\", stdout=\"x\" * 5000)\n    assert note.startswith(tutor.RUN_RESULT_PREFIX), note\n    assert \"截断\" in note and len(note) < 800, f\"长输出必须截断：{len(note)}\"\n\n@test\ndef run_note_报错只留尾部():\n    long_tb = \"Traceback...\\n\" + \"f\\n\" * 2000 + \"ZeroDivisionError: division by zero\"\n    note = tutor.run_result_note(\"RE\", stderr=long_tb)\n    assert tutor.RUN_RESULT_PREFIX in note and \"ZeroDivisionError\" in note\n    assert len(note) < 800, \"报错应只留尾部\"\n\n@test\ndef run_note_超时():\n    note = tutor.run_result_note(\"HANG\")\n    assert \"超时\" in note or \"死循环\" in note\n\n@test\ndef inject_只保留最新一条运行结果():\n    h = [{\"role\": \"user\", \"content\": \"我观察到了\"},\n         {\"role\": \"assistant\", \"content\": \"嗯\"}]\n    h = tutor.inject_run_note(h, tutor.run_result_note(\"OK\", stdout=\"第一次\"))\n    h = tutor.inject_run_note(h, tutor.run_result_note(\"RE\", stderr=\"第二次错误\"))\n    notes = [m for m in h if (m.get(\"content\") or \"\").startswith(tutor.RUN_RESULT_PREFIX)]\n    assert len(notes) == 1, f\"只该留最新一条运行结果，实际 {len(notes)}\"\n    assert \"第二次错误\" in notes[0][\"content\"], \"留下的应是最新那条\"\n    # 非运行结果的历史不能被误删\n    assert any(m[\"content\"] == \"我观察到了\" for m in h), \"普通历史不该被剔除\"\n\n\n# ---- 知返固定讲解策略：骨架 / 切路 / 断层 / 调用链 ----\ndef _strategy_session(stage=\"①发现\", stalled=0):\n    return TutorSession(id=\"strategy\", student_id=\"stu\", pattern_id=\"BP-BOUNDARY-001\",\n                        manifest={}, stage=stage, stalled_turns=stalled, history=[])\n\n\n@test\ndef teaching_strategy_正常轮先搭骨架且只推当前目标():\n    s = tutor.choose_teaching_strategy(_strategy_session(stage=\"②定位\"), \"我看到报错了\")\n    assert s.explanation_route == \"skeleton\" and not s.route_changed\n    assert \"可疑位置\" in s.core_goal, s\n\n\n@test\ndef teaching_strategy_听不懂时切换解释路径():\n    s = tutor.choose_teaching_strategy(_strategy_session(), \"还是听不懂，太抽象了\")\n    assert s.route_changed and s.explanation_route != \"skeleton\", s\n    prompt = tutor.teaching_strategy_prompt(s)\n    assert \"禁止重复上一轮定义和问法\" in prompt\n\n\n@test\ndef teaching_strategy_主动识别知识断层():\n    s = tutor.choose_teaching_strategy(_strategy_session(), \"下标是什么意思？\")\n    assert s.knowledge_gap == \"下标/索引\", s\n    assert \"先补这个最小前置知识\" in tutor.teaching_strategy_prompt(s)\n\n\n@test\ndef teaching_strategy_代码按调用链讲():\n    s = tutor.choose_teaching_strategy(_strategy_session(stalled=1), \"这段代码我看不懂\")\n    assert s.explanation_route == \"execution_trace\", s\n    assert s.code_explanation_order == tutor.CODE_EXPLANATION_ORDER\n    prompt = tutor.teaching_strategy_prompt(s)\n    assert \"入口与输入→谁调用谁→关键数据怎样变化→返回值或报错落点\" in prompt\n\n\ndef _guard_strategy(route=\"skeleton\", changed=False):\n    return tutor.TeachingStrategy(\n        core_goal=\"只推进一个小目标\",\n        explanation_route=route,\n        confusion_detected=changed,\n        route_changed=changed,\n    )\n\n\n@test\ndef reply_guard_拦直接修法与多个问题():\n    leak = tutor.assess_teaching_reply(\n        \"直接改成 range(len(arr)) 就行。你明白了吗？\",\n        _guard_strategy(), \"range(len(arr))\",\n    )\n    assert \"direct_fix_leak\" in leak.violations\n    many = tutor.assess_teaching_reply(\n        \"你先看哪一行？这个值又从哪来？\",\n        _guard_strategy(), \"range(len(arr))\",\n    )\n    assert \"question_count\" in many.violations and many.question_count == 2\n\n    one_of_two = tutor.assess_teaching_reply(\n        \"可以改成列表推导 [s for s in scores if s >= 60]。你能先说出它避开了什么吗？\",\n        _guard_strategy(),\n        \"遍历副本 for s in scores[:] 或列表推导 [s for s in scores if s >= 60]\",\n    )\n    assert \"direct_fix_leak\" in one_of_two.violations\n\n\n@test\ndef reply_guard_换路必须在回复中真实可见():\n    missing = tutor.assess_teaching_reply(\n        \"我们再想一想。你现在看到什么？\",\n        _guard_strategy(\"micro_example\", True), \"range(len(arr))\",\n    )\n    assert \"route_not_observable\" in missing.violations\n    good = tutor.assess_teaching_reply(\n        \"比如三个抽屉只编号 0 到 2，我们先只看这个小例子。你觉得编号 3 还能找到抽屉吗？\",\n        _guard_strategy(\"micro_example\", True), \"range(len(arr))\",\n    )\n    assert good.passed, good\n\n\n@test\ndef reply_guard_首次违规会纠错重试():\n    original = tutor._call_llm\n    calls = []\n    try:\n        replies = iter([\n            \"你先看哪一行？这个值从哪来？\",\n            \"比如三个抽屉只编号 0 到 2。你觉得编号 3 存在吗？\",\n        ])\n\n        def fake_call(system, history):\n            calls.append(system)\n            return tutor.TutorTurn(\n                reply=next(replies), stage_transition=None, hint_level_used=\"L0\",\n                student_progressed=False, answer_begging=False, events=[],\n            )\n\n        tutor._call_llm = fake_call\n        turn, meta = tutor._guarded_llm_turn(\n            \"system\", [], _guard_strategy(\"micro_example\", True),\n            \"range(len(arr))\", \"②定位\",\n        )\n        assert len(calls) == 2 and meta[\"attempts\"] == 2\n        assert meta[\"passed\"] is True and meta[\"fallback_used\"] is False\n        assert \"question_count\" in meta[\"violations\"] and \"三个抽屉\" in turn.reply\n    finally:\n        tutor._call_llm = original\n\n\n@test\ndef reply_guard_连续违规时安全兜底():\n    original = tutor._call_llm\n    try:\n        tutor._call_llm = lambda system, history: tutor.TutorTurn(\n            reply=\"直接改成 range(len(arr))。照着改就好。\",\n            stage_transition=None, hint_level_used=\"L0\",\n            student_progressed=False, answer_begging=False, events=[],\n        )\n        turn, meta = tutor._guarded_llm_turn(\n            \"system\", [], _guard_strategy(), \"range(len(arr))\", \"②定位\",\n        )\n        assert meta[\"passed\"] is False and meta[\"fallback_used\"] is True\n        assert \"range(len(arr))\" not in turn.reply\n        assert len(re.findall(r\"[？?]\", turn.reply)) == 1\n    finally:\n        tutor._call_llm = original\n\n\n# ---- Resource Gateway 亮点落地：动作信封 / 用途 / 风险 / Evidence 关联 ----\n@test\ndef action_governance_摘要稳定_call唯一且不泄露代码():\n    import json\n    from app.services import action_governance as ag\n    kwargs = dict(actor_id=\"ag_u1\", session_id=\"ag_s1\", action=\"sandbox.run\",\n                  purpose=\"debug_observation\", initiated_by=\"student\",\n                  payload={\"code\": \"print('secret-source')\", \"mode\": \"debug\"})\n    a = ag.authorize_action(**kwargs)\n    b = ag.authorize_action(**kwargs)\n    assert a[\"call_id\"] != b[\"call_id\"], (a, b)\n    assert a[\"request_digest\"] == b[\"request_digest\"], (a, b)\n    assert a[\"risk_level\"] == 1 and a[\"decision\"] == \"allow\" and a[\"execution\"] == \"pending\"\n    assert \"secret-source\" not in json.dumps(a), \"动作上下文只存摘要，不得复制代码原文\"\n\n\n@test\ndef action_governance_未知动作与Agent代提交失败关闭():\n    from app.services import action_governance as ag\n    try:\n        ag.authorize_action(actor_id=\"u\", session_id=\"s\", action=\"shell.exec\",\n                            purpose=\"debug\", initiated_by=\"student\", payload={})\n        assert False, \"未知动作必须拒绝\"\n    except ag.ActionDenied as exc:\n        assert \"未注册动作\" in str(exc)\n    try:\n        ag.authorize_action(actor_id=\"u\", session_id=\"s\", action=\"submission.judge\",\n                            purpose=\"judge_submission\", initiated_by=\"agent\", payload={})\n        assert False, \"知返不得代替学生提交\"\n    except ag.ActionDenied as exc:\n        assert \"不能触发\" in str(exc)\n\n\n@test\ndef action_governance_运行用途由服务端状态确定():\n    from app.services import action_governance as ag\n    assert ag.purpose_for_run(stage=\"⑤验证\", mode=\"debug\") == \"verify_hypothesis\"\n    assert ag.purpose_for_run(stage=\"②定位\", mode=\"debug\") == \"debug_observation\"\n    assert ag.purpose_for_run(stage=\"⑤验证\", mode=\"coop\") == \"collaborative_debug\"\n\n\n@test\ndef action_governance_提交事实与能力事件共享call_id():\n    from app.main import SubmitReq, submit_fix\n    from app.services import mine_engine\n    db = TestSession()\n    sid, ssid, pid = \"ag_submit_u\", \"ag_submit_s\", \"BP-BOUNDARY-001\"\n    pat = mine_engine.get_pattern(pid)\n    sess = TutorSession(\n        id=ssid, student_id=sid, pattern_id=pid,\n        manifest=mine_engine.build_manifest(sid, pat), history=[],\n        stage=\"④修复\", mine_status=\"found\", status=\"active\",\n    )\n    db.add(sess); db.commit()\n    fixed_code = pat[\"buggy_code\"].replace(\"range(len(arr) + 1)\", \"range(len(arr))\")\n    result = submit_fix(ssid, SubmitReq(code=fixed_code), db=db, me=sid, _lock=None)\n    assert result[\"passed\"] is True\n    ctx = result[\"action_context\"]\n    assert ctx[\"action\"] == \"submission.judge\" and ctx[\"risk_level\"] == 2\n    assert ctx[\"decision\"] == \"allow\" and ctx[\"execution\"] == \"completed\"\n    execution = db.query(ExecutionEvent).filter_by(session_id=ssid, source=\"submit\").one()\n    fixed_event = db.query(Event).filter_by(session_id=ssid, capability=\"Independent_Debug\").one()\n    assert execution.meta[\"action_context\"][\"call_id\"] == ctx[\"call_id\"]\n    assert fixed_event.context[\"action_context\"][\"call_id\"] == ctx[\"call_id\"]\n    db.close()\n\n\n# ---- 接着做（未完成关卡续做）：cleanup / active-sessions / abandon（设计 08）----\nfrom app.main import _cleanup_abandoned, abandon_session, get_active_sessions\n\n_AS_PID = \"BP-BOUNDARY-001\"  # 用真实题，name 能填上\n\ndef _mk_active(db, sid, ssid, *, history=None, mine_status=\"planted\", events_at=None,\n               updated_at=None):\n    \"\"\"造一个 active 会话 + 可选的若干 ExecutionEvent（timestamp 为 ISO 串）。\n    updated_at 给定则设为该值——active-sessions 现按 updated_at 排序（不再回查事件表）。\"\"\"\n    sess = TutorSession(id=ssid, student_id=sid, pattern_id=_AS_PID, manifest={},\n                        mine_status=mine_status, history=history or [], status=\"active\")\n    if updated_at:\n        sess.updated_at = updated_at\n    db.add(sess)\n    for ts in (events_at or []):\n        db.add(ExecutionEvent(\n            id=f\"ex_{_uuid.uuid4().hex[:12]}\", version=\"v1\", student_id=sid,\n            session_id=ssid, pattern_id=_AS_PID, source=\"run\", kind=\"OK\",\n            knowledge_points=[], meta={}, timestamp=ts))\n    db.commit()\n\n@test\ndef cleanup_只清纯空壳():\n    db = TestSession()\n    sid = \"cl_1\"\n    # 系统注入的 user 消息（（系统·…）开头）不算真实发言\n    sys_only = [{\"role\": \"system\", \"content\": \"运行这段代码，看看它的行为\"},\n                {\"role\": \"user\", \"content\": \"（系统·运行结果）2\"}]\n    _mk_active(db, sid, \"empty\", history=sys_only)                                   # 纯空壳 → 删\n    _mk_active(db, sid, \"has_msg\", history=[{\"role\": \"user\", \"content\": \"我觉得循环错了\"}])  # 有真实发言 → 留\n    _mk_active(db, sid, \"has_run\", history=sys_only, events_at=[_dt.now().isoformat()])     # 有运行记录 → 留\n    _cleanup_abandoned(db, sid)\n    db.commit()  # _cleanup_abandoned 只 delete 不 commit（真实调用方 create_session 负责提交）\n    left = {s.id for s in db.query(TutorSession).filter_by(student_id=sid).all()}\n    assert left == {\"has_msg\", \"has_run\"}, f\"只该清纯空壳，实际剩：{left}\"\n    db.close()\n\n@test\ndef cleanup_已定位不误删():\n    db = TestSession()\n    sid = \"cl_2\"\n    _mk_active(db, sid, \"located\", history=[], mine_status=\"found\")  # 非 planted → 不在清理范围\n    _cleanup_abandoned(db, sid)\n    assert db.query(TutorSession).filter_by(id=\"located\").first() is not None, \"已定位会话不该被清\"\n    db.close()\n\n@test\ndef active_sessions_按活跃倒序且上限5():\n    db = TestSession()\n    sid = \"as_1\"\n    base = _dt.now()\n    for i in range(6):  # 6 个；updated_at 各异，i 越大越老\n        _mk_active(db, sid, f\"s{i}\", history=[{\"role\": \"user\", \"content\": \"x\"}],\n                   updated_at=(base - _td(days=i)).isoformat())\n    sessions = get_active_sessions(sid, db=db, me=sid)[\"sessions\"]\n    assert len(sessions) == 5, f\"上限 5，实际 {len(sessions)}\"\n    assert sessions[0][\"session_id\"] == \"s0\", f\"最近活跃应置顶，实际 {sessions[0]['session_id']}\"\n    times = [s[\"last_active_at\"] for s in sessions]\n    assert times == sorted(times, reverse=True), f\"应按活跃度倒序：{times}\"\n    assert \"s5\" not in [s[\"session_id\"] for s in sessions], \"最老的应被挤出前 5\"\n    assert sessions[0][\"name\"] and sessions[0][\"stage\"], \"应带 name/stage 供大厅展示\"\n    assert sessions[0][\"display_step\"] == \"运行观察\", \"学生界面应只消费四步语言\"\n    db.close()\n\n@test\ndef active_sessions_无事件回退created_at():\n    db = TestSession()\n    sid = \"as_2\"\n    _mk_active(db, sid, \"no_ev\", history=[{\"role\": \"user\", \"content\": \"x\"}])  # 无 ExecutionEvent\n    sessions = get_active_sessions(sid, db=db, me=sid)[\"sessions\"]\n    assert len(sessions) == 1 and sessions[0][\"last_active_at\"], \"无事件应回退 created_at\"\n    db.close()\n\n@test\ndef abandon_置abandoned且不删历史与事件():\n    db = TestSession()\n    sid = \"ab_1\"\n    hist = [{\"role\": \"user\", \"content\": \"我修了第 2 行\"}]\n    _mk_active(db, sid, \"drop_me\", history=hist, events_at=[_dt.now().isoformat()])\n    abandon_session(\"drop_me\", db=db, me=sid)\n    s = db.query(TutorSession).filter_by(id=\"drop_me\").first()\n    assert s.status == \"abandoned\", f\"应置 abandoned，实际 {s.status}\"\n    assert s.history == hist, \"history 不该被删\"\n    assert db.query(ExecutionEvent).filter_by(session_id=\"drop_me\").count() == 1, \"事件流不该被删\"\n    assert not get_active_sessions(sid, db=db, me=sid)[\"sessions\"], \"放弃后不该再在 active-sessions\"\n    db.close()\n\n\n# ---- sandbox ----\n@test\ndef sandbox_基本():\n    assert sandbox.run_code(\"print(1+1)\").stdout.strip() == \"2\"\n    assert sandbox.run_code(\"raise ValueError()\").has_error\n    assert sandbox.normalize_output(\"  6  \") == \"6\"\n\n\n@test\ndef 提问训练_三要素评分单调():\n    from app.services import question_training as q\n    s0 = q._score_from_factors(False, False, False)\n    s1 = q._score_from_factors(True, False, False)\n    s2 = q._score_from_factors(True, True, False)\n    s3 = q._score_from_factors(True, True, True)\n    assert s0 < s1 < s2 < s3, (s0, s1, s2, s3)\n    assert 0 <= s0 and s3 <= 100\n\n\n@test\ndef 提问训练_空提问不调LLM得0():\n    from app.services import question_training as q\n    r = q.diagnose(\"login\", \"   \")\n    assert r[\"score\"] == 0 and r[\"degraded\"] is False\n    assert not r[\"phenomenon\"] and not r[\"context\"] and not r[\"expectation\"]\n\n\n@test\ndef 提问训练_fallback结构完整且不崩():\n    from app.services import question_training as q\n    r = q._fallback(\"随便写的提问\")\n    for k in (\"phenomenon\", \"context\", \"expectation\", \"score\", \"feedback\", \"confidence\", \"degraded\"):\n        assert k in r, k\n    assert r[\"degraded\"] is True and r[\"confidence\"] == 0.0\n\n\n@test\ndef 提问训练_资产化落ExecutionEvent且命名空间正确():\n    from app.services import question_training as q\n    db = TestSession()\n    r = {\"phenomenon\": True, \"context\": True, \"expectation\": False, \"score\": 66,\n         \"feedback\": \"现象上下文已清，缺预期\", \"confidence\": 0.9, \"degraded\": False}\n    ev = q.log_diagnosis(db, student_id=\"stu_qt1\", scenario_id=\"login\", prompt=\"登录登不上去\", result=r)\n    assert ev is not None\n    row = db.query(ExecutionEvent).filter_by(student_id=\"stu_qt1\").one()\n    assert row.source == \"qt_diagnose\"\n    assert row.pattern_id == \"QUESTION_TRAINING\"\n    assert row.kind == \"QT\"\n    assert row.knowledge_points == []                    # 空 kp → 对画像隐形\n    assert row.session_id == \"qtsess_stu_qt1\"            # 合成 session → 对轨迹隐形\n    assert row.meta[\"scenario_id\"] == \"login\" and row.meta[\"score\"] == 66\n    assert row.meta[\"expectation\"] is False\n    db.close()\n\n\n@test\ndef 提问训练_资产不进画像不进成长轨迹():\n    from app.services import question_training as q\n    from app.services import profile, timeline\n    db = TestSession()\n    # 先给该生建一个真实知识点状态 + 写若干 qt 资产\n    profile.update_knowledge_state(db, student_id=\"stu_qt2\", pattern_id=\"BP-X\",\n                                   knowledge_points=[\"循环\"], new_state=\"已接触\")\n    for _ in range(4):\n        q.log_diagnosis(db, student_id=\"stu_qt2\", scenario_id=\"login\", prompt=\"登录登不上去\",\n                        result={\"phenomenon\": True, \"context\": True, \"expectation\": False,\n                                \"score\": 66, \"feedback\": \"缺预期\", \"confidence\": 0.9, \"degraded\": False})\n    # 画像掌握度里\"循环\"的 evidence_count 不应被 qt 事件抬高（qt 是空 kp）\n    mastery = profile.knowledge_mastery(db, \"stu_qt2\")\n    loop = [m for m in mastery if m[\"kp\"] == \"循环\"]\n    assert loop and loop[0][\"evidence_count\"] == 0, loop\n    # 成长轨迹不应出现 qt 事件\n    blob = str(timeline.build_timeline(db, \"stu_qt2\"))\n    assert \"qt_diagnose\" not in blob and \"QUESTION_TRAINING\" not in blob\n    db.close()\n\n\n@test\ndef 提问训练_短板统计样本不足返回None_够了返回最常漏():\n    from app.services import question_training as q\n    db = TestSession()\n    # 2 条 < 门槛(3) → None\n    for _ in range(2):\n        q.log_diagnosis(db, student_id=\"stu_qt3\", scenario_id=\"login\", prompt=\"x\",\n                        result={\"phenomenon\": True, \"context\": True, \"expectation\": False,\n                                \"score\": 66, \"feedback\": \"f\", \"confidence\": 0.9, \"degraded\": False})\n    assert q.weakness_summary(db, \"stu_qt3\") is None\n    # 再加 2 条（共 4 条，都缺预期）→ 返回 expectation\n    for _ in range(2):\n        q.log_diagnosis(db, student_id=\"stu_qt3\", scenario_id=\"login\", prompt=\"x\",\n                        result={\"phenomenon\": True, \"context\": True, \"expectation\": False,\n                                \"score\": 66, \"feedback\": \"f\", \"confidence\": 0.9, \"degraded\": False})\n    w = q.weakness_summary(db, \"stu_qt3\")\n    assert w and w[\"factor\"] == \"expectation\" and w[\"miss_count\"] == 4\n    db.close()\n\n\n@test\ndef 提问训练_degraded与低置信不参与短板统计():\n    from app.services import question_training as q\n    db = TestSession()\n    # 3 条都是 degraded 或低置信 → 不参与 → 样本不足 → None\n    q.log_diagnosis(db, student_id=\"stu_qt4\", scenario_id=\"login\", prompt=\"x\",\n                    result={\"phenomenon\": False, \"context\": False, \"expectation\": False,\n                            \"score\": 12, \"feedback\": \"f\", \"confidence\": 0.0, \"degraded\": True})\n    for _ in range(2):\n        q.log_diagnosis(db, student_id=\"stu_qt4\", scenario_id=\"login\", prompt=\"x\",\n                        result={\"phenomenon\": False, \"context\": False, \"expectation\": False,\n                                \"score\": 12, \"feedback\": \"f\", \"confidence\": 0.5, \"degraded\": False})\n    assert q.weakness_summary(db, \"stu_qt4\") is None\n\n\n@test\ndef 提问训练_custom场景注册且无隐藏背景():\n    from app.services import question_training as q\n    assert \"custom\" in q.SCENARIOS and q.SCENARIOS[\"custom\"][\"brief\"] == \"\"\n\n\n@test\ndef 提问训练_custom资产化与预置共存且短板合并():\n    from app.services import question_training as q\n    db = TestSession()\n    sid = \"stu_custom\"\n    # 2 条 custom + 2 条预置，都缺预期 → 短板统计应合并到 4 条、判 expectation\n    for _ in range(2):\n        q.log_diagnosis(db, student_id=sid, scenario_id=\"custom\", prompt=\"我自己的问题x\",\n                        result={\"phenomenon\": True, \"context\": True, \"expectation\": False,\n                                \"score\": 66, \"feedback\": \"f\", \"confidence\": 0.9, \"degraded\": False})\n    for _ in range(2):\n        q.log_diagnosis(db, student_id=sid, scenario_id=\"login\", prompt=\"预置场景x\",\n                        result={\"phenomenon\": True, \"context\": True, \"expectation\": False,\n                                \"score\": 66, \"feedback\": \"f\", \"confidence\": 0.9, \"degraded\": False})\n    # 资产里 custom 与预置共存\n    rows = db.query(ExecutionEvent).filter_by(student_id=sid, source=\"qt_diagnose\").all()\n    sids = {(r.meta or {}).get(\"scenario_id\") for r in rows}\n    assert sids == {\"custom\", \"login\"}, sids\n    # 短板统计合并 4 条\n    w = q.weakness_summary(db, sid)\n    assert w and w[\"factor\"] == \"expectation\" and w[\"sample_size\"] == 4\n    db.close()\n\n\ndef _qres(p, c, e, conf=0.9, degraded=False):\n    return {\"phenomenon\": p, \"context\": c, \"expectation\": e,\n            \"score\": 0, \"feedback\": \"f\", \"confidence\": conf, \"degraded\": degraded}\n\n\n@test\ndef 提问训练计分_三要素全_Prompt_Design正分点亮AI协作维度():\n    from app.services import question_training as q\n    from app.services import profile\n    db = TestSession()\n    sid = \"stu_sc1\"\n    # 计分前：AI协作维度无分\n    before = profile.get_profile(db, sid)\n    ai_before = [d for d in before[\"dimensions\"] if d[\"name\"] == \"AI协作能力\"][0]\n    assert ai_before[\"score\"] is None\n    # 三要素全 → +2\n    ev = q.score_diagnosis(db, student_id=sid, scenario_id=\"login\",\n                           prompt=\"登录报NPE在第32行，期望登录成功\", result=_qres(True, True, True))\n    assert ev is not None and ev.capability == \"Prompt_Design\" and ev.delta == 2\n    after = profile.get_profile(db, sid)\n    ai_after = [d for d in after[\"dimensions\"] if d[\"name\"] == \"AI协作能力\"][0]\n    assert ai_after[\"score\"] is not None, after\n    assert after[\"vector\"][\"Prompt_Design\"][\"events_count\"] == 1\n    db.close()\n\n\n@test\ndef 提问训练计分_degraded和单要素不记分():\n    from app.services import question_training as q\n    db = TestSession()\n    assert q.score_diagnosis(db, student_id=\"s\", scenario_id=\"login\", prompt=\"x\",\n                             result=_qres(True, True, True, degraded=True)) is None\n    assert q.score_diagnosis(db, student_id=\"s\", scenario_id=\"login\", prompt=\"只有现象\",\n                             result=_qres(True, False, False)) is None  # 1要素→delta0→不记\n    assert db.query(Event).filter_by(student_id=\"s\", capability=\"Prompt_Design\").count() == 0\n    db.close()\n\n\n@test\ndef 提问训练计分_纯帮我看看记负分():\n    from app.services import question_training as q\n    db = TestSession()\n    ev = q.score_diagnosis(db, student_id=\"s_neg\", scenario_id=\"login\", prompt=\"代码报错了帮我看看\",\n                           result=_qres(False, False, False))\n    assert ev is not None and ev.delta == -1\n    db.close()\n\n\n@test\ndef 提问训练计分_日上限6封顶():\n    from app.services import question_training as q\n    db = TestSession()\n    sid = \"stu_cap\"\n    # 不同提问各 +2，3 条到 +6\n    for i in range(3):\n        ev = q.score_diagnosis(db, student_id=sid, scenario_id=\"login\",\n                               prompt=f\"完整提问版本{i}\", result=_qres(True, True, True))\n        assert ev is not None, i\n    # 第 4 条超上限 → 不记\n    assert q.score_diagnosis(db, student_id=sid, scenario_id=\"login\",\n                             prompt=\"完整提问版本4\", result=_qres(True, True, True)) is None\n    db.close()\n\n\n@test\ndef 提问训练计分_同句今日不重复记():\n    from app.services import question_training as q\n    db = TestSession()\n    sid = \"stu_dup\"\n    p = \"登录报NPE在第32行，期望登录成功\"\n    assert q.score_diagnosis(db, student_id=sid, scenario_id=\"login\", prompt=p, result=_qres(True, True, True)) is not None\n    assert q.score_diagnosis(db, student_id=sid, scenario_id=\"login\", prompt=p, result=_qres(True, True, True)) is None\n    db.close()\n\n\n@test\ndef 提问训练计分_泛泛求代办2要素降档不记():\n    from app.services import question_training as q\n    db = TestSession()\n    # 2要素(现象+上下文) + \"帮我看看\" → C档降为0、不记\n    ev = q.score_diagnosis(db, student_id=\"s_vague\", scenario_id=\"login\",\n                           prompt=\"登录报错了，帮我看看\", result=_qres(True, True, False))\n    assert ev is None\n    assert db.query(Event).filter_by(student_id=\"s_vague\", capability=\"Prompt_Design\").count() == 0\n    db.close()\n\n\n@test\ndef 提问训练计分_泛泛求代办0要素仍负分():\n    from app.services import question_training as q\n    db = TestSession()\n    ev = q.score_diagnosis(db, student_id=\"s_v0\", scenario_id=\"login\",\n                           prompt=\"帮我看看\", result=_qres(False, False, False))\n    assert ev is not None and ev.delta == -1\n    db.close()\n\n\n@test\ndef 提问训练计分_明确协作请求不算泛泛():\n    from app.services import question_training as q\n    # 含\"判断/还是\"等明确协作请求 → 不被当泛泛求代办（即使有\"帮我\"）\n    assert q._is_vague_handoff(\"可以帮我判断是接口返回结构问题，还是前端取值逻辑问题吗？\") is False\n    assert q._is_vague_handoff(\"帮我看看\") is True\n    assert q._is_vague_handoff(\"登录报错了，帮我看看\") is True\n    db = TestSession()\n    # 2要素 + 明确协作请求 → 正常 +1（不降档）\n    ev = q.score_diagnosis(db, student_id=\"s_collab\", scenario_id=\"login\",\n                           prompt=\"登录报错，是接口问题还是前端取值问题？\", result=_qres(True, True, False))\n    assert ev is not None and ev.delta == 1\n    db.close()\n\n\n@test\ndef 提问训练计分_三要素完整不受泛泛影响():\n    from app.services import question_training as q\n    db = TestSession()\n    # 三要素齐全即便带\"帮我看看\"也照记 +2（已是完整提问）\n    ev = q.score_diagnosis(db, student_id=\"s_full3\", scenario_id=\"login\",\n                           prompt=\"登录报NPE在第32行，期望登录成功，帮我看看\", result=_qres(True, True, True))\n    assert ev is not None and ev.delta == 2\n    db.close()\n\n\n@test\ndef 提问训练计分_低置信入库但不进画像():\n    from app.services import question_training as q\n    from app.models import CapabilityScore\n    db = TestSession()\n    sid = \"stu_lowconf\"\n    ev = q.score_diagnosis(db, student_id=sid, scenario_id=\"login\",\n                           prompt=\"低置信完整提问\", result=_qres(True, True, True, conf=0.5))\n    assert ev is not None  # 事件入库\n    # 但 confidence<0.7 不进画像分\n    assert db.query(CapabilityScore).filter_by(student_id=sid, capability=\"Prompt_Design\").first() is None\n    db.close()\n\n\n@test\ndef 状态机_修复阶段未提交时LLM不得擅自跃迁():\n    \"\"\"硬门回归：离开④修复只能靠 judge_fix 提交通过。LLM 在对话里填 stage_transition=⑤\n    也不放行——否则代码从未判过就能走到⑤⑥甚至被标「已内化」（真实漏洞，2026-06-22 修）。\"\"\"\n    from app.services import mine_engine, tutor\n    from app.models import TutorSession\n    orig_get, orig_load, orig_llm = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        pat = _good_candidate()\n        mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n        mine_engine.load_patterns = lambda: {pat[\"id\"]: pat}\n        tutor._call_llm = lambda system, history: tutor.TutorTurn(\n            reply=\"继续\", stage_transition=\"⑤验证\", hint_level_used=\"L1\",\n            student_progressed=True, answer_begging=False, events=[])\n        db = TestSession()\n        sess = TutorSession(id=\"sess_gate1\", student_id=\"g1\", pattern_id=pat[\"id\"],\n                            manifest=mine_engine.build_manifest(\"g1\", pat), history=[],\n                            stage=\"④修复\", mine_status=\"found\")\n        db.add(sess); db.commit()\n        tutor.run_turn(db, sess, \"改个边界就行\")\n        assert sess.stage == \"④修复\", f\"未提交不该离开④，实际到了 {sess.stage}\"\n        assert sess.mine_status == \"found\", sess.mine_status\n        saved = (db.query(SessionMessage).filter_by(session_id=sess.id, role=\"tutor\")\n                 .order_by(SessionMessage.seq.desc()).first())\n        assert saved.meta[\"teaching_strategy\"][\"explanation_route\"] == \"skeleton\"\n        assert saved.meta[\"stage_before\"] == \"④修复\" and saved.meta[\"stage_after\"] == \"④修复\"\n        assert saved.meta[\"reply_guard\"][\"fallback_used\"] is True\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = orig_get, orig_load, orig_llm\n\n\n@test\ndef 状态机_提交通过后可正常进验证():\n    \"\"\"对照：mine_status=fixed（提交已判过）时，④→⑤ 的跃迁正常放行，不被新门误伤。\"\"\"\n    from app.services import mine_engine, tutor\n    from app.models import TutorSession\n    orig_get, orig_load, orig_llm = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        pat = _good_candidate()\n        mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n        mine_engine.load_patterns = lambda: {pat[\"id\"]: pat}\n        tutor._call_llm = lambda system, history: tutor.TutorTurn(\n            reply=\"继续\", stage_transition=\"⑤验证\", hint_level_used=\"L1\",\n            student_progressed=True, answer_begging=False, events=[])\n        db = TestSession()\n        sess = TutorSession(id=\"sess_gate2\", student_id=\"g2\", pattern_id=pat[\"id\"],\n                            manifest=mine_engine.build_manifest(\"g2\", pat), history=[],\n                            stage=\"④修复\", mine_status=\"fixed\")  # 已提交判过\n        db.add(sess); db.commit()\n        tutor.run_turn(db, sess, \"我想测空列表\")\n        assert sess.stage == \"⑤验证\", f\"已 fixed 应可进⑤，实际 {sess.stage}\"\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = orig_get, orig_load, orig_llm\n\n\n# ════════ 导师纯规则函数（无 LLM、无 DB，最便宜的回归网）════════\n\n@test\ndef 受挫检测_学生本人受挫与程序卡住要分开():\n    from app.services import tutor\n    # 学生本人受挫 → True\n    for s in [\"我不会\", \"完全没思路\", \"太难了\", \"想放弃\", \"给个提示\", \"啊啊啊\", \"我卡住了\"]:\n        assert tutor.detect_frustration(s), f\"应判受挫: {s}\"\n    # 描述程序卡住（不是本人受挫）→ False\n    for s in [\"程序卡住了跑不完\", \"代码一直在循环卡住\", \"运行卡住\"]:\n        assert not tutor.detect_frustration(s), f\"程序卡死≠学生受挫: {s}\"\n    # 负向断言别误伤：会不会/知不知道 不算受挫\n    for s in [\"会不会越界\", \"我知不知道得测一下\", \"这样对不对\"]:\n        assert not tutor.detect_frustration(s), f\"不该误判受挫: {s}\"\n\n\n@test\ndef 错误签名识别_命中返回剧本_未命中None():\n    from app.services import tutor\n    sig, pb = tutor.detect_error_signature(\"我猜会报 IndexError\")\n    assert sig == \"IndexError\" and \"meaning\" in pb\n    assert tutor.detect_error_signature(\"不知道哪里错\") is None\n\n\n@test\ndef 定位判定_引用雷行或点行号才算():\n    from app.services import mine_engine, tutor\n    orig = mine_engine.get_pattern\n    try:\n        pat = _good_candidate()  # 雷在第 2 行：return a[3]\n        mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n        assert tutor.message_points_at_mine(\"我觉得 return a[3] 有问题\", pat[\"id\"])\n        assert tutor.message_points_at_mine(\"第 2 行不对吧\", pat[\"id\"])\n        assert not tutor.message_points_at_mine(\"不知道在哪\", pat[\"id\"])\n    finally:\n        mine_engine.get_pattern = orig\n\n\n@test\ndef 边界解释判定_要同时有边界输入和预期():\n    from app.services import tutor\n    assert tutor.message_explains_boundary_test(\"空列表传进去，预期返回 None\")\n    assert not tutor.message_explains_boundary_test(\"空列表试试\")      # 缺预期\n    assert not tutor.message_explains_boundary_test(\"应该返回正确结果\")  # 缺具体边界输入\n\n\n@test\ndef 内化实质性_纯表态不算_含机制才算():\n    from app.services import tutor\n    for s in [\"我懂了\", \"好的\", \"嗯嗯\", \"以后注意\"]:\n        assert not tutor.is_substantive(s), f\"纯表态不该算: {s}\"\n    assert tutor.is_substantive(\"因为下标从0数，长度3时最大下标是2，访问3就越界了\")\n\n\n# ════════ run_turn 状态机跃迁（stub LLM，覆盖规则跃迁与硬门）════════\n\ndef _stub_turn(tutor, **kw):\n    base = dict(reply=\"嗯\", stage_transition=None, hint_level_used=\"L0\",\n                student_progressed=True, answer_begging=False, events=[])\n    base.update(kw)\n    return lambda system, history: tutor.TutorTurn(**base)\n\ndef _arena_session(tutor, mine_engine, *, stage, mine_status=\"planted\"):\n    pat = _good_candidate()\n    mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n    mine_engine.load_patterns = lambda: {pat[\"id\"]: pat}\n    from app.models import TutorSession\n    db = TestSession()\n    sess = TutorSession(id=f\"s_{stage}_{__import__('uuid').uuid4().hex[:6]}\", student_id=\"rt1\", pattern_id=pat[\"id\"],\n                        manifest=mine_engine.build_manifest(\"rt1\", pat), history=[],\n                        stage=stage, mine_status=mine_status)\n    db.add(sess); db.commit()\n    return db, sess, pat\n\n\n@test\ndef 跃迁_发现阶段贴出报错签名自动进定位():\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        tutor._call_llm = _stub_turn(tutor)  # LLM 不报跃迁\n        db, sess, pat = _arena_session(tutor, mine_engine, stage=\"①发现\")\n        tutor.run_turn(db, sess, \"它会报 IndexError 吧\")\n        assert sess.stage == \"②定位\", sess.stage  # 规则层据报错签名自动跃迁\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 跃迁_定位阶段指认雷行自动进归因且雷标found():\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        tutor._call_llm = _stub_turn(tutor)\n        db, sess, pat = _arena_session(tutor, mine_engine, stage=\"②定位\")\n        tutor.run_turn(db, sess, \"问题在 return a[3] 这行\")\n        assert sess.stage == \"③归因\", sess.stage\n        assert sess.mine_status == \"found\", sess.mine_status\n        assert sess.attribution_step == \"variable_trace\"\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 跃迁_内化三轴达标则升级已内化并结束会话():\n    from app.services import mine_engine, profile, tutor\n    from app.models import KnowledgeState\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        tutor._call_llm = _stub_turn(\n            tutor, internalization=tutor.InternalizationScore(cause=True, locate=True, prevent=False))\n        db, sess, pat = _arena_session(tutor, mine_engine, stage=\"⑥内化\", mine_status=\"fixed\")\n        tutor.run_turn(db, sess, \"因为下标越界，定位靠看报错行和追踪变量，下次先查 range 边界\")\n        assert sess.mine_status == \"internalized\", sess.mine_status\n        assert sess.status == \"completed\", sess.status\n        ks = db.get(KnowledgeState, (\"rt1\", pat[\"id\"]))\n        assert ks and ks.state == \"已内化\", ks\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n# ════════ 算分 / 画像（profile）════════\n\n@test\ndef 算分_冷启动加倍且封顶100():\n    from app.services import event_engine\n    from app.models import CapabilityScore\n    db = TestSession()\n    # INITIAL=30；冷启动×2（events<10）；规则事件 conf=1.0 → 每条 +3×3×2 = +18\n    for _ in range(4):\n        event_engine.emit(db, student_id=\"sc_up\", session_id=\"x\", capability=\"Independent_Debug\",\n                          delta=3, producer=\"rule\", evidence={\"summary\": \"fix\"})\n    r = db.get(CapabilityScore, (\"sc_up\", \"Independent_Debug\"))\n    assert r.score == 100.0, r.score        # 30→48→66→84→100，4 条即封顶\n    assert r.events_count == 4\n    db.close()\n\n\n@test\ndef 算分_负向事件下探不破0():\n    from app.services import event_engine\n    from app.models import CapabilityScore\n    db = TestSession()\n    for _ in range(2):  # 30 -18 -18 → 钳到 0\n        event_engine.emit(db, student_id=\"sc_dn\", session_id=\"x\", capability=\"Independent_Debug\",\n                          delta=-3, producer=\"rule\", evidence={\"summary\": \"neg\"})\n    r = db.get(CapabilityScore, (\"sc_dn\", \"Independent_Debug\"))\n    assert r.score == 0.0, r.score\n    db.close()\n\n\n@test\ndef 算分_低置信LLM事件入库但不进分():\n    from app.services import event_engine\n    from app.models import CapabilityScore, Event\n    db = TestSession()\n    event_engine.emit(db, student_id=\"sc_lc\", session_id=\"x\", capability=\"Root_Cause_Reasoning\",\n                      delta=3, producer=\"llm_judge\", confidence=0.5, evidence={\"summary\": \"弱证据\"})\n    assert db.query(Event).filter_by(student_id=\"sc_lc\").count() == 1          # 事件入库\n    assert db.get(CapabilityScore, (\"sc_lc\", \"Root_Cause_Reasoning\")) is None  # 但不进画像分\n    db.close()\n\n\n@test\ndef 掌握度_状态映射与薄弱判定():\n    from app.services import profile\n    db = TestSession()\n    sid = \"km1\"\n    profile.update_knowledge_state(db, student_id=sid, pattern_id=\"P_A\",\n                                   knowledge_points=[\"kpA\"], new_state=\"已内化\")\n    profile.update_knowledge_state(db, student_id=sid, pattern_id=\"P_B\",\n                                   knowledge_points=[\"kpB\"], new_state=\"已解决\")\n    profile.update_knowledge_state(db, student_id=sid, pattern_id=\"P_C\",\n                                   knowledge_points=[\"kpC\"], new_state=\"已接触\")\n    m = {x[\"kp\"]: x for x in profile.knowledge_mastery(db, sid)}\n    assert m[\"kpA\"][\"mastery\"] == \"掌握\" and m[\"kpA\"][\"weak\"] is False\n    assert m[\"kpB\"][\"mastery\"] == \"在学\" and m[\"kpB\"][\"weak\"] is True   # 已解决→建议内化\n    assert m[\"kpC\"][\"mastery\"] == \"生疏\" and m[\"kpC\"][\"weak\"] is True   # 已接触→还没真正解决\n    db.close()\n\n\n# ════════ AI 共脑调试 coop（B0，不连 LLM 的部分）════════\n\n@test\ndef coop_建会话_mode为coop且伪pattern_id():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start(db, \"co1\", \"off_by_one\")\n    assert out[\"sample_id\"] == \"off_by_one\" and out[\"code\"]\n    s = db.get(TutorSession, out[\"session_id\"])\n    assert (s.manifest or {}).get(\"mode\") == \"coop\"\n    assert s.pattern_id == \"coop_off_by_one\"\n    assert s.history and s.history[0][\"role\"] == \"user\"   # 首条=样本求助\n    db.close()\n\n\n@test\ndef coop_非法样本回退第一个():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start(db, \"co2\", \"不存在的样本\")\n    assert out[\"sample_id\"] in coop.SAMPLES\n    db.close()\n\n\n@test\ndef coop_run落ExecutionEvent_mode_coop且不查pattern():\n    from app.services import coop\n    from app.models import ExecutionEvent\n    db = TestSession()\n    out = coop.start(db, \"co3\", \"off_by_one\")\n    # 跑一段真代码（不依赖任何 pattern，伪 id 不会崩）\n    r = coop.run(db, out[\"session_id\"], \"print(1+1)\")\n    assert r[\"stdout\"].strip() == \"2\"\n    ev = db.query(ExecutionEvent).filter_by(session_id=out[\"session_id\"]).one()\n    assert ev.knowledge_points == []                 # 对画像天然隐形\n    assert (ev.meta or {}).get(\"mode\") == \"coop\"\n    ctx = (ev.meta or {}).get(\"action_context\") or {}\n    assert ctx[\"call_id\"] == r[\"action_context\"][\"call_id\"]\n    assert ctx[\"purpose\"] == \"collaborative_debug\" and ctx[\"risk_level\"] == 1\n    assert ctx[\"decision\"] == \"allow\" and ctx[\"execution\"] == \"completed\"\n    db.close()\n\n\n@test\ndef 续学决策_固定四级优先级且理由可解释():\n    active = [{\"session_id\": \"s1\", \"pattern_id\": \"p1\", \"name\": \"未完成题\",\n               \"stage\": \"④修复\"}]\n    due = [{\"pattern_id\": \"p2\", \"name\": \"复习题\", \"days_since\": 5, \"interval_days\": 2}]\n    recs = [{\"id\": \"p3\", \"name\": \"推荐题\", \"reason\": \"补强边界意识\",\n             \"knowledge_points\": [\"数组边界\"]}]\n\n    resume = learning_path.decide_next_action(active, due, recs)\n    assert resume[\"action\"][\"kind\"] == \"resume\"\n    assert resume[\"action\"][\"reason_code\"] == \"unfinished_first\"\n    assert resume[\"action\"][\"evidence\"][\"ref\"] == \"s1\"\n    assert \"修改验证\" in resume[\"action\"][\"detail\"] and \"修复\" not in resume[\"action\"][\"detail\"]\n    assert resume[\"pending\"] == {\"active\": 1, \"reviews\": 1, \"recommendations\": 1}\n\n    review_action = learning_path.decide_next_action([], due, recs)\n    assert review_action[\"action\"][\"kind\"] == \"review\"\n    assert review_action[\"action\"][\"target\"][\"mode\"] == \"review\"\n    assert \"5 天前\" in review_action[\"action\"][\"detail\"]\n\n    recommend = learning_path.decide_next_action([], [], recs)\n    assert recommend[\"action\"][\"kind\"] == \"recommend\"\n    assert recommend[\"action\"][\"detail\"] == \"补强边界意识\"\n\n    start = learning_path.decide_next_action([], [], [])\n    assert start[\"action\"][\"kind\"] == \"start\"\n    assert start[\"decision_order\"] == [\"resume\", \"review\", \"recommend\", \"start\"]\n\n\n@test\ndef coop_run把真实结果注入对话():\n    from app.services import coop, tutor\n    db = TestSession()\n    out = coop.start(db, \"co4\", \"off_by_one\")\n    coop.run(db, out[\"session_id\"], \"print(42)\")\n    s = db.get(TutorSession, out[\"session_id\"])\n    notes = [m for m in s.history if m[\"content\"].startswith(tutor.RUN_RESULT_PREFIX)]\n    assert len(notes) == 1 and \"42\" in notes[0][\"content\"]\n    db.close()\n\n\n@test\ndef coop_resolve标completed():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start(db, \"co5\", \"off_by_one\")\n    assert coop.resolve(db, out[\"session_id\"])[\"status\"] == \"completed\"\n    s = db.get(TutorSession, out[\"session_id\"])\n    assert s.status == \"completed\"\n    db.close()\n\n\n@test\ndef 开题干预_短TTL缓存命中与清除():\n    from app.services import timeline\n    db = TestSession()\n    timeline.clear_recurring_cache()\n    # 首次算并入缓存\n    r1 = timeline.recurring_for_student(db, \"cache_u\")\n    assert \"cache_u\" in timeline._RECURRING_CACHE\n    # TTL 内再取是同一对象（走缓存、没重算）\n    r2 = timeline.recurring_for_student(db, \"cache_u\")\n    assert r1 is r2, \"TTL 内应返回缓存的同一结果\"\n    # 清缓存后重算，得到新对象\n    timeline.clear_recurring_cache()\n    assert \"cache_u\" not in timeline._RECURRING_CACHE\n    r3 = timeline.recurring_for_student(db, \"cache_u\")\n    assert r3 is not r1, \"清缓存后应重新计算\"\n    timeline.clear_recurring_cache()\n    db.close()\n\n\n@test\ndef coop_防污染_不进成长轨迹():\n    from app.services import coop, timeline\n    db = TestSession()\n    out = coop.start(db, \"co6\", \"off_by_one\")\n    coop.run(db, out[\"session_id\"], \"print(1)\")\n    tl = timeline.build_timeline(db, \"co6\")\n    assert tl[\"episodes\"] == []        # coop 会话被 build_timeline 排除\n    db.close()\n\n\n@test\ndef coop_防污染_不抬高知识点掌握度():\n    from app.services import coop, profile\n    db = TestSession()\n    out = coop.start(db, \"co7\", \"off_by_one\")\n    coop.run(db, out[\"session_id\"], \"print(1)\")\n    # coop 的 ExecutionEvent knowledge_points=[] → 不产生任何 kp 掌握度\n    assert profile.knowledge_mastery(db, \"co7\") == []\n    db.close()\n\n\n@test\ndef coop_错误session_id返回None():\n    from app.services import coop\n    db = TestSession()\n    assert coop.get(db, \"不存在\") is None\n    assert coop.run(db, \"不存在\", \"print(1)\") is None\n    assert coop.message(db, \"不存在\", \"hi\") is None\n    assert coop.resolve(db, \"不存在\") is None\n    db.close()\n\n\n@test\ndef coop_自带代码_建custom会话且存code():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start_custom(db, \"cc1\", \"print('hi')\\n\", \"它不输出东西\")\n    assert out[\"sample_id\"] is None and out[\"title\"] == \"我的代码\"\n    s = db.get(TutorSession, out[\"session_id\"])\n    assert (s.manifest or {}).get(\"custom\") is True\n    assert s.pattern_id == \"coop_custom\"\n    assert s.manifest[\"code\"] == \"print('hi')\"       # code 存进 manifest，刷新可恢复\n    assert s.history[0][\"content\"] == \"它不输出东西\"   # 描述作首条求助\n    db.close()\n\n\n@test\ndef coop_自带代码_空代码返回None_超长截断():\n    from app.services import coop\n    db = TestSession()\n    assert coop.start_custom(db, \"cc2\", \"   \") is None       # 空代码不建会话\n    big = \"x = 1\\n\" * 5000                                   # 远超上限\n    out = coop.start_custom(db, \"cc2\", big)\n    s = db.get(TutorSession, out[\"session_id\"])\n    assert len(s.manifest[\"code\"]) == coop.MAX_CODE_LEN      # 截断到上限\n    db.close()\n\n\n@test\ndef coop_自带代码_无描述给默认求助():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start_custom(db, \"cc3\", \"print(1)\")\n    assert \"看看\" in out[\"ask\"]    # 没填描述 → 默认一句求助（导师仍会主动问预期）\n    db.close()\n\n\n@test\ndef coop_自带代码_get恢复manifest里的code():\n    from app.services import coop\n    db = TestSession()\n    out = coop.start_custom(db, \"cc4\", \"a = 42\\nprint(a)\")\n    got = coop.get(db, out[\"session_id\"])\n    assert got[\"code\"] == \"a = 42\\nprint(a)\" and got[\"title\"] == \"我的代码\"\n    db.close()\n\n\n@test\ndef coop_自带代码_防污染不进轨迹():\n    from app.services import coop, timeline\n    db = TestSession()\n    out = coop.start_custom(db, \"cc5\", \"print(1)\")\n    coop.run(db, out[\"session_id\"], \"print(1)\")\n    assert timeline.build_timeline(db, \"cc5\")[\"episodes\"] == []\n    db.close()\n\n\n@test\ndef coop_get不能拿到闯关会话():\n    \"\"\"coop 的 get 只认 mode=coop 会话，普通闯关 session 一律 None（隔离）。\"\"\"\n    from app.services import coop\n    db = TestSession()\n    s = TutorSession(id=\"normal_sess\", student_id=\"co8\", pattern_id=\"BP-X\",\n                     manifest={\"mode\": \"debug\", \"mines\": [{}]}, history=[])\n    db.add(s); db.commit()\n    assert coop.get(db, \"normal_sess\") is None\n    db.close()\n\n\n@test\ndef coopB2_埋点_resolve记协作信号且不进画像():\n    from app.services import coop\n    from app.models import ExecutionEvent, CapabilityScore, Event\n    db = TestSession()\n    out = coop.start(db, \"b2a\", \"off_by_one\")\n    coop.run(db, out[\"session_id\"], \"print(1)\")      # 真实动手运行一次\n    coop.resolve(db, out[\"session_id\"])\n    sig = db.query(ExecutionEvent).filter_by(session_id=out[\"session_id\"], source=\"coop_resolve\").all()\n    assert len(sig) == 1 and sig[0].kind == \"COLLAB\"\n    assert sig[0].meta[\"collab\"][\"runs\"] == 1\n    assert sig[0].knowledge_points == []\n    # 埋点不计分：无能力分、无能力事件\n    assert db.query(CapabilityScore).filter_by(student_id=\"b2a\").count() == 0\n    assert db.query(Event).filter_by(student_id=\"b2a\").count() == 0\n    db.close()\n\n\n@test\ndef coopB2_埋点_没动手运行不记():\n    from app.services import coop\n    from app.models import ExecutionEvent\n    db = TestSession()\n    out = coop.start(db, \"b2b\", \"off_by_one\")\n    coop.resolve(db, out[\"session_id\"])    # 没点过运行 → 不算有效协作\n    assert db.query(ExecutionEvent).filter_by(session_id=out[\"session_id\"], source=\"coop_resolve\").count() == 0\n    db.close()\n\n\n@test\ndef coopB2_埋点_不进成长轨迹():\n    from app.services import coop, timeline\n    db = TestSession()\n    out = coop.start(db, \"b2c\", \"off_by_one\")\n    coop.run(db, out[\"session_id\"], \"print(1)\")\n    coop.resolve(db, out[\"session_id\"])\n    assert timeline.build_timeline(db, \"b2c\")[\"episodes\"] == []\n    db.close()\n\n\n@test\ndef 画像_维度带覆盖度trained_total():\n    from app.services import event_engine, profile\n    db = TestSession()\n    # 只练 1 个 Debug 子能力（共 5 个）→ trained=1 total=5，治 #18 虚高的语境\n    for _ in range(3):\n        event_engine.emit(db, student_id=\"cov1\", session_id=\"x\", capability=\"Independent_Debug\",\n                          delta=3, producer=\"rule\", evidence={\"summary\": \"fix\"})\n    deb = [d for d in profile.get_profile(db, \"cov1\")[\"dimensions\"] if d[\"name\"] == \"Debug能力\"][0]\n    assert deb[\"trained\"] == 1 and deb[\"total\"] == 5, deb\n    ai = [d for d in profile.get_profile(db, \"cov1\")[\"dimensions\"] if d[\"name\"] == \"AI协作能力\"][0]\n    assert ai[\"trained\"] == 0 and ai[\"total\"] == 3, ai\n\n\n@test\ndef 画像_18虚高修法A_单练不顶满维度():\n    from app.services import event_engine, profile\n    db = TestSession()\n    # 只练独立调试（权重 0.25）刷到封顶，其余 4 个 Debug 子能力零证据\n    for _ in range(30):\n        event_engine.emit(db, student_id=\"v18\", session_id=\"x\", capability=\"Independent_Debug\",\n                          delta=3, producer=\"rule\", evidence={\"summary\": \"fix\"})\n    deb = [d for d in profile.get_profile(db, \"v18\")[\"dimensions\"] if d[\"name\"] == \"Debug能力\"][0]\n    # 修法 A：Independent_Debug=100，全维度权重和=1.0 → 100*0.25/1.0=25，不再虚高到 100\n    assert deb[\"score\"] is not None and deb[\"score\"] <= 30, f\"单练一项不该顶满维度，实际 {deb['score']}\"\n    # 未练的维度仍是 None（无任何证据不给分）\n    ai = [d for d in profile.get_profile(db, \"v18\")[\"dimensions\"] if d[\"name\"] == \"AI协作能力\"][0]\n    assert ai[\"score\"] is None, ai\n    db.close()\n\n\n@test\ndef 灵犀_新用户无历史_有活动后有天数():\n    from app.services import coop, presence\n    db = TestSession()\n    assert presence.presence_signals(db, \"ling1\") == {\n        \"has_history\": False, \"last_active_at\": None, \"days_since\": None}\n    # 建个 coop 会话（产生 TutorSession）→ 算\"来过\"\n    coop.start(db, \"ling1\", \"off_by_one\")\n    p = presence.presence_signals(db, \"ling1\")\n    assert p[\"has_history\"] is True and p[\"days_since\"] == 0   # 刚活动 → 0 天\n    db.close()\n\n\n# ════════ judge_fix 剩余分支（WA 答案错 / HANG 死循环）════════\n\n@test\ndef judge_跑通但输出不符判WA():\n    from app.services import mine_engine\n    pat = _good_candidate()   # expected_output = \"2\"\n    orig = mine_engine.get_pattern\n    try:\n        mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n        # 能跑通、不报错，但输出 1 ≠ 期望 2 → WA\n        r = mine_engine.judge_fix(pat[\"id\"], \"def f(a):\\n    return a[0]\\nprint(f([1, 2]))\\n\")\n        assert r[\"kind\"] == \"WA\" and r[\"passed\"] is False, r\n    finally:\n        mine_engine.get_pattern = orig\n\n\n@test\ndef judge_死循环判HANG():\n    from app.services import mine_engine\n    pat = _good_candidate()\n    orig = mine_engine.get_pattern\n    try:\n        mine_engine.get_pattern = lambda pid: {pat[\"id\"]: pat}[pid]\n        r = mine_engine.judge_fix(pat[\"id\"], \"while True:\\n    pass\\n\")   # 超时（真跑约 4s）\n        assert r[\"kind\"] == \"HANG\" and r[\"passed\"] is False, r\n    finally:\n        mine_engine.get_pattern = orig\n\n\n# ════════ run_turn 状态机剩余分支（③归因步进 / 止损降档）════════\n\n@test\ndef 跃迁_归因步进_变量追踪到规则对照():\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        # 学生答对当前步（variable_trace）且有进展 → 步进到 rule_compare\n        tutor._call_llm = _stub_turn(tutor, attribution_step=\"variable_trace\", student_progressed=True)\n        db, sess, pat = _arena_session(tutor, mine_engine, stage=\"③归因\")\n        tutor.run_turn(db, sess, \"len 是 3，i 会取到 0,1,2,3\")\n        assert sess.attribution_step == \"rule_compare\", sess.attribution_step\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 止损_受挫时提示级别升一格():\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        tutor._call_llm = _stub_turn(tutor, hint_level_used=\"L0\", student_progressed=False)\n        db, sess, pat = _arena_session(tutor, mine_engine, stage=\"②定位\")\n        assert sess.hint_level == \"L0\"\n        tutor.run_turn(db, sess, \"我不会，完全没思路\")   # 受挫 → 止损共情层升一格\n        assert sess.hint_level == \"L1\", sess.hint_level\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef coop_样本库七道且每道可建会话():\n    from app.services import coop\n    assert len(coop.SAMPLES) == 7, len(coop.SAMPLES)\n    assert len(coop.list_samples()) == 7\n    db = TestSession()\n    for sid in coop.SAMPLES:\n        out = coop.start(db, \"samp\", sid)\n        assert out[\"sample_id\"] == sid and out[\"code\"] and out[\"title\"] and out[\"ask\"], sid\n    db.close()\n\n\n@test\ndef 报错翻译_各错误返回中文_无报错返回None():\n    from app.services import tutor\n    assert \"死循环\" in (tutor.explain_error(\"HANG\", \"\") or \"\")\n    assert tutor.explain_error(\"RE\", \"IndexError: list index out of range\")\n    assert \"下一步\" in tutor.explain_error(\"RE\", \"KeyError: 'x'\")   # 翻译含\"怎么查\"方向\n    assert tutor.explain_error(\"OK\", \"\") is None                 # 没报错不硬塞\n    assert tutor.explain_error(\"RE\", \"完全不认识的乱码\") is None   # 识别不了返回 None\n\n\n# ---- 身份 / 越权（M1 D2-3）：token 签发可用、坏 token 拒绝、越权访问他人数据 403 ----\n\n@test\ndef token_签发校验round_trip():\n    from app.security import sign_token, verify_token\n    t = sign_token(\"stu_A\")\n    assert verify_token(t) == \"stu_A\"                     # 正常签发可解回\n    assert verify_token(t[:-1] + (\"0\" if t[-1] != \"0\" else \"1\")) is None  # 篡改签名 → 拒绝\n    assert verify_token(\"garbage\") is None and verify_token(None) is None\n\n\n@test\ndef 越权_看他人画像403():\n    from fastapi import HTTPException\n    from app.main import get_profile\n    db = TestSession()\n    try:\n        # 我是 stu_A，带自己的 token 去要 stu_B 的画像 → 必须 403\n        get_profile(\"stu_B\", db=db, me=\"stu_A\")\n        assert False, \"越权访问他人画像未被拦截\"\n    except HTTPException as e:\n        assert e.status_code == 403\n    # 要自己的画像 → 放行（返回结构正常）\n    assert get_profile(\"stu_A\", db=db, me=\"stu_A\")[\"student_id\"] == \"stu_A\"\n    db.close()\n\n\n@test\ndef 越权_操作他人会话403():\n    from fastapi import HTTPException\n    from app.main import _get_session\n    db = TestSession()\n    _mk_active(db, \"owner_X\", \"sess_X\", history=[{\"role\": \"user\", \"content\": \"x\"}])\n    try:\n        _get_session(db, \"sess_X\", \"intruder_Y\")   # 别人的会话\n        assert False, \"越权操作他人会话未被拦截\"\n    except HTTPException as e:\n        assert e.status_code == 403\n    assert _get_session(db, \"sess_X\", \"owner_X\").id == \"sess_X\"  # 本人放行\n    db.close()\n\n\n# ---- 代码快照（M2）：diff_stats 计算 + run/submit 双写 ----\n\n@test\ndef 快照_diff计算_首版与增量():\n    # 首版（无上一版）：非空行全算改动\n    d0 = process.compute_diff_stats(None, \"a = 1\\n\\nb = 2\\n\")\n    assert d0[\"lines_changed\"] == 2, d0\n    # 增量：改 1 行\n    d1 = process.compute_diff_stats(\"a = 1\\nb = 2\\n\", \"a = 1\\nb = 3\\n\")\n    assert d1[\"lines_changed\"] == 1, d1\n\n\n@test\ndef 快照_touched雷行判定():\n    pat = mine_engine.get_pattern(\"BP-BOUNDARY-001\")\n    buggy = pat[\"buggy_code\"]\n    line = pat[\"mine_location\"][\"line\"]\n    # 原样带雷码 → 没动雷行\n    d_same = process.compute_diff_stats(None, buggy, buggy, line)\n    assert d_same[\"touched_mine_line\"] is False, d_same\n    # 把雷行 range(len(arr)+1) 改掉 → 动过雷行\n    fixed = buggy.replace(\"range(len(arr) + 1)\", \"range(len(arr))\")\n    d_fix = process.compute_diff_stats(None, fixed, buggy, line)\n    assert d_fix[\"touched_mine_line\"] is True, d_fix\n    # 无 pattern 信息（coop）→ None\n    d_none = process.compute_diff_stats(None, \"print(1)\")\n    assert d_none[\"touched_mine_line\"] is None, d_none\n\n\n@test\ndef 快照_diff摘要_盲改与定向改():\n    # 没动雷行 → 盲改提示\n    s_blind = process.diff_summary({\"lines_changed\": 1, \"touched_mine_line\": False})\n    assert \"没落在\" in s_blind, s_blind\n    # 动了雷行仍不对 → 定向改提示\n    s_aim = process.diff_summary({\"lines_changed\": 1, \"touched_mine_line\": True})\n    assert \"关键\" in s_aim and \"方向对\" in s_aim, s_aim\n    # 大面积改动附注\n    assert \"多处试探\" in process.diff_summary({\"lines_changed\": 8, \"touched_mine_line\": False})\n    # coop / 无雷信息 → 空串（不打扰）\n    assert process.diff_summary({\"lines_changed\": 3, \"touched_mine_line\": None}) == \"\"\n\n\n@test\ndef 快照_record_seq递增且落库():\n    db = TestSession()\n    s = TutorSession(id=\"snap_s1\", student_id=\"u\", pattern_id=\"BP-BOUNDARY-001\",\n                     manifest={}, history=[], status=\"active\")\n    db.add(s)\n    db.commit()\n    process.record_snapshot(db, s, \"print(1)\", \"ex_1\")\n    process.record_snapshot(db, s, \"print(2)\", \"ex_2\")\n    db.commit()\n    snaps = (db.query(CodeSnapshot).filter_by(session_id=\"snap_s1\")\n             .order_by(CodeSnapshot.seq).all())\n    assert [x.seq for x in snaps] == [1, 2], \"seq 应从 1 递增\"\n    assert snaps[0].code == \"print(1)\" and snaps[1].code == \"print(2)\"\n    assert snaps[1].execution_event_id == \"ex_2\"\n    db.close()\n\n\n@test\ndef 教师总览_聚合进度与班级过滤():\n    db = TestSession()\n    # 两个学生，一个属 A 班、一个属 B 班；进度不同\n    db.add(Student(id=\"t_a\", name=\"小A\", class_id=\"classA\"))\n    db.add(Student(id=\"t_b\", name=\"小B\", class_id=\"classB\"))\n    db.add(KnowledgeState(student_id=\"t_a\", pattern_id=\"BP-BOUNDARY-001\",\n                          knowledge_points=[], state=\"已内化\"))\n    db.add(KnowledgeState(student_id=\"t_a\", pattern_id=\"BP-BOUNDARY-003\",\n                          knowledge_points=[], state=\"已解决\"))\n    db.add(KnowledgeState(student_id=\"t_b\", pattern_id=\"BP-LOOP-001\",\n                          knowledge_points=[], state=\"已接触\"))\n    db.commit()\n    allv = teacher.build_class_overview(db)\n    assert allv[\"total\"] == 2\n    a = next(r for r in allv[\"students\"] if r[\"student_id\"] == \"t_a\")\n    assert a[\"solved\"] == 2 and a[\"internalized\"] == 1 and a[\"name\"] == \"小A\"\n    b = next(r for r in allv[\"students\"] if r[\"student_id\"] == \"t_b\")\n    assert b[\"solved\"] == 0 and b[\"touched\"] == 1\n    # 按班过滤\n    only_a = teacher.build_class_overview(db, class_id=\"classA\")\n    assert only_a[\"total\"] == 1 and only_a[\"students\"][0][\"student_id\"] == \"t_a\"\n    db.close()\n\n\n@test\ndef 教师端_共性卡点与学生钻取均为派生():\n    db = TestSession()\n    for sid in (\"td_a\", \"td_b\"):\n        db.add(Student(id=sid, name=sid, class_id=\"C1\"))\n        db.add(TutorSession(\n            id=f\"sess_{sid}\", student_id=sid, pattern_id=\"BP-BOUNDARY-001\",\n            manifest={}, history=[], stage=\"②定位\", status=\"active\",\n        ))\n    db.commit()\n    process.record_message(\n        db, \"sess_td_a\", \"tutor\", \"换个例子\",\n        meta={\n            \"teaching_strategy\": {\"explanation_route\": \"micro_example\", \"route_changed\": True,\n                                  \"knowledge_gap\": \"下标/索引\"},\n            \"confusion_detected\": True, \"stage_before\": \"②定位\", \"stage_after\": \"②定位\",\n            \"student_progressed\": False,\n        },\n    )\n    db.commit()\n    overview = teacher.build_class_overview(db, \"C1\")\n    common = overview[\"common_stumbling_blocks\"][0]\n    assert common[\"student_count\"] == 2 and common[\"stage\"] == \"②定位\", common\n    detail = teacher.build_student_detail(db, \"td_a\")\n    assert detail[\"student\"][\"student_id\"] == \"td_a\"\n    assert detail[\"sessions\"][0][\"session_id\"] == \"sess_td_a\"\n    assert detail[\"teaching_metrics\"][\"route_changed_turns\"] == 1\n    assert teacher.build_student_detail(db, \"missing\") is None\n    db.close()\n\n\n@test\ndef 课程配置_单元完整且题都真实存在():\n    units = curriculum.load_units()\n    assert len(units) == 4, f\"应 4 个单元，实际 {len(units)}\"\n    lib = mine_engine.load_patterns()\n    total = 0\n    for u in units:\n        assert u[\"patterns\"], f\"{u['id']} 无题\"\n        for pid in u[\"patterns\"]:\n            assert pid in lib, f\"{u['id']} 配了不存在的题 {pid}\"\n            total += 1\n    assert total == 21, f\"单元覆盖题数应 21，实际 {total}\"\n\n\n@test\ndef 课程派生_进度反映知识状态():\n    db = TestSession()\n    sid = \"cur_u\"\n    # 一道题标已内化、一道已解决，其余未接触\n    db.add(KnowledgeState(student_id=sid, pattern_id=\"BP-BOUNDARY-001\",\n                          knowledge_points=[], state=\"已内化\"))\n    db.add(KnowledgeState(student_id=sid, pattern_id=\"BP-BOUNDARY-003\",\n                          knowledge_points=[], state=\"已解决\"))\n    db.commit()\n    cur = curriculum.build_curriculum(db, sid)\n    u1 = next(u for u in cur[\"units\"] if u[\"id\"] == \"U1\")\n    assert u1[\"internalized\"] == 1 and u1[\"solved\"] == 2, u1  # 已内化也计入 solved\n    by_pid = {lv[\"pattern_id\"]: lv for lv in u1[\"levels\"]}\n    assert by_pid[\"BP-BOUNDARY-001\"][\"state\"] == \"已内化\"\n    assert by_pid[\"BP-BOUNDARY-003\"][\"state\"] == \"已解决\"\n    assert by_pid[\"BP-BOUNDARY-002\"][\"state\"] == \"未接触\"\n    # 单元顺序按 order\n    assert [u[\"id\"] for u in cur[\"units\"]] == [\"U1\", \"U2\", \"U3\", \"U4\"]\n    db.close()\n\n\n@test\ndef 回放_合并快照与消息按时间成步():\n    import time as _time\n    db = TestSession()\n    s = TutorSession(id=\"rp_s1\", student_id=\"u\", pattern_id=\"BP-BOUNDARY-001\",\n                     manifest={}, history=[], status=\"active\")\n    db.add(s); db.commit()\n    replay_ctx = {\"call_id\": \"call_replay\", \"purpose\": \"debug_observation\",\n                  \"decision\": \"allow\", \"execution\": \"completed\"}\n    ev = event_engine.log_execution(db, student_id=\"u\", session_id=\"rp_s1\",\n                                    pattern_id=\"BP-BOUNDARY-001\", source=\"run\", kind=\"RE\",\n                                    action_context=replay_ctx)\n    process.record_snapshot(db, s, \"print(1)\", ev.id)\n    process.record_message(db, \"rp_s1\", \"system\", \"（系统·运行结果）报错\", \"run_result\")\n    db.commit()\n    _time.sleep(0.01)\n    process.record_message(db, \"rp_s1\", \"student\", \"我觉得是循环\", \"chat\")\n    db.commit()\n    rp = process.build_replay(db, \"rp_s1\")\n    assert rp[\"code_versions\"] == 1 and rp[\"pattern_name\"]\n    kinds = [x[\"kind\"] for x in rp[\"steps\"]]\n    assert \"code\" in kinds and kinds.count(\"msg\") == 2, kinds\n    code_step = next(x for x in rp[\"steps\"] if x[\"kind\"] == \"code\")\n    assert code_step[\"source\"] == \"run\" and code_step[\"result\"] == \"RE\"\n    assert code_step[\"action_context\"][\"call_id\"] == \"call_replay\"\n    assert rp[\"mainline_steps\"] == 1 and code_step[\"mainline\"] is True\n    ats = [x[\"at\"] for x in rp[\"steps\"]]\n    assert ats == sorted(ats), \"回放步应按时间升序\"\n    assert process.build_replay(db, \"不存在\") is None\n    db.close()\n\n\n@test\ndef 回放2_从失败到定向修改再通过标出转折点():\n    steps = [\n        {\"version\": 1, \"result\": \"RE\", \"result_label\": \"报错\", \"error_family\": \"IndexError\",\n         \"diff_stats\": {\"touched_mine_line\": False}},\n        {\"version\": 2, \"result\": \"RE\", \"result_label\": \"报错\", \"error_family\": \"TypeError\",\n         \"diff_stats\": {\"touched_mine_line\": True}},\n        {\"version\": 3, \"result\": \"OK\", \"result_label\": \"通过\", \"error_family\": None,\n         \"diff_stats\": {\"touched_mine_line\": True}},\n    ]\n    points = process.derive_turning_points(steps)\n    types = [point[\"type\"] for point in points]\n    assert types == [\"first_failure\", \"targeted_change\", \"error_changed\", \"breakthrough\"], types\n    assert steps[2][\"annotations\"][0][\"type\"] == \"breakthrough\"\n\n\n@test\ndef 回放主线_只归拢真实动作转折观察总结与换路():\n    steps = [\n        {\"kind\": \"msg\", \"role\": \"tutor\", \"content\": \"先看整体\", \"meta\": {}},\n        {\"kind\": \"code\", \"version\": 1, \"annotations\": []},\n        {\"kind\": \"msg\", \"role\": \"student\", \"content\": \"【观察记录】\\n我观察到：报错\", \"meta\": {}},\n        {\"kind\": \"msg\", \"role\": \"tutor\", \"content\": \"换个例子\", \"meta\": {\n            \"teaching_strategy\": {\"route_changed\": True, \"explanation_route\": \"micro_example\"},\n        }},\n        {\"kind\": \"code\", \"version\": 2, \"annotations\": [\n            {\"type\": \"breakthrough\", \"title\": \"关键突破\", \"detail\": \"通过\"},\n        ]},\n        {\"kind\": \"msg\", \"role\": \"student\", \"content\": \"【思考总结】\\n我想记住：先验证\", \"meta\": {}},\n        {\"kind\": \"msg\", \"role\": \"system\", \"content\": \"完成\", \"meta\": {}},\n    ]\n    count = process.derive_replay_mainline(steps)\n    assert count == 5\n    assert not steps[0][\"mainline\"] and not steps[-1][\"mainline\"]\n    assert steps[1][\"mainline_reasons\"] == [\"first_execution\"]\n    assert steps[2][\"mainline_reasons\"] == [\"student_observation\"]\n    assert steps[3][\"mainline_reasons\"] == [\"teaching_route_changed\"]\n    assert steps[4][\"mainline_reasons\"] == [\"turning_point\", \"latest_execution\"]\n    assert steps[5][\"mainline_reasons\"] == [\"student_summary\"]\n\n\n@test\ndef 消息时间线_record_seq递增且落库():\n    db = TestSession()\n    s = TutorSession(id=\"msg_s1\", student_id=\"u\", pattern_id=\"BP-BOUNDARY-001\",\n                     manifest={}, history=[], status=\"active\")\n    db.add(s); db.commit()\n    process.record_message(db, \"msg_s1\", \"student\", \"我觉得循环错了\", \"chat\")\n    process.record_message(db, \"msg_s1\", \"tutor\", \"为什么这么认为？\", \"chat\",\n                           meta={\"teaching_strategy\": {\"explanation_route\": \"skeleton\"}})\n    process.record_message(db, \"msg_s1\", \"system\", \"（系统·运行结果）报错\", \"run_result\")\n    db.commit()\n    msgs = (db.query(SessionMessage).filter_by(session_id=\"msg_s1\")\n            .order_by(SessionMessage.seq).all())\n    assert [m.seq for m in msgs] == [1, 2, 3], \"seq 应从 1 递增\"\n    assert [m.role for m in msgs] == [\"student\", \"tutor\", \"system\"]\n    assert msgs[2].kind == \"run_result\"\n    assert msgs[1].meta[\"teaching_strategy\"][\"explanation_route\"] == \"skeleton\"\n    db.close()\n\n\n@test\ndef 教学指标_困惑后两轮内恢复可重放计算():\n    db = TestSession()\n    s = TutorSession(id=\"metric_s1\", student_id=\"u\", pattern_id=\"BP-BOUNDARY-001\",\n                     manifest={}, history=[], status=\"active\")\n    db.add(s); db.commit()\n    process.record_message(db, s.id, \"tutor\", \"换个小例子\", meta={\n        \"teaching_strategy\": {\"explanation_route\": \"micro_example\", \"route_changed\": True,\n                              \"knowledge_gap\": \"下标/索引\"},\n        \"confusion_detected\": True, \"stage_before\": \"②定位\", \"stage_after\": \"②定位\",\n        \"student_progressed\": False,\n    })\n    process.record_message(db, s.id, \"tutor\", \"沿执行链再走一次\", meta={\n        \"teaching_strategy\": {\"explanation_route\": \"execution_trace\", \"route_changed\": False,\n                              \"knowledge_gap\": None},\n        \"confusion_detected\": False, \"stage_before\": \"②定位\", \"stage_after\": \"③归因\",\n        \"student_progressed\": True,\n    })\n    db.commit()\n    metrics = process.teaching_metrics(db, s.id)\n    assert metrics[\"tutor_turns\"] == 2 and metrics[\"confusion_turns\"] == 1\n    assert metrics[\"recovered_within_two_turns\"] == 1 and metrics[\"recovery_rate\"] == 1.0\n    assert metrics[\"knowledge_gap_turns\"] == 1 and metrics[\"routes\"][\"micro_example\"] == 1\n    db.close()\n\n\n# ---- 生产安全门槛 / 教师认证头 ----\n@test\ndef 生产安全门槛_生产缺隔离或秘密时失败关闭():\n    from app import runtime_security\n    keys = (\"ACP_ENV\", \"ACP_SANDBOX\", \"ACP_ADMIN_TOKEN\", \"ACP_AUTH_SECRET\", \"DEEPSEEK_API_KEY\")\n    old_env = {key: os.environ.get(key) for key in keys}\n    old_check = runtime_security.sandbox.docker_available\n    try:\n        os.environ[\"ACP_ENV\"] = \"production\"\n        os.environ[\"ACP_SANDBOX\"] = \"subprocess\"\n        for key in keys[2:]:\n            os.environ.pop(key, None)\n        runtime_security.sandbox.docker_available = lambda: False\n        try:\n            runtime_security.validate_production_environment()\n            assert False, \"生产环境未隔离时必须拒绝启动\"\n        except RuntimeError as exc:\n            assert \"production_security_gate\" in str(exc) and \"Docker\" in str(exc)\n\n        os.environ[\"ACP_SANDBOX\"] = \"docker\"\n        for key in keys[2:]:\n            os.environ[key] = \"configured-for-test\"\n        runtime_security.sandbox.docker_available = lambda: True\n        assert runtime_security.validate_production_environment()[\"ready\"] is True\n    finally:\n        runtime_security.sandbox.docker_available = old_check\n        for key, value in old_env.items():\n            if value is None:\n                os.environ.pop(key, None)\n            else:\n                os.environ[key] = value\n\n\n@test\ndef 教师认证_优先Bearer且错误口令拒绝():\n    from fastapi import HTTPException\n    from app.main import _check_admin\n    old = os.environ.get(\"ACP_ADMIN_TOKEN\")\n    try:\n        os.environ[\"ACP_ADMIN_TOKEN\"] = \"teacher-secret\"\n        _check_admin(authorization=\"Bearer teacher-secret\")\n        try:\n            _check_admin(authorization=\"Bearer wrong\")\n            assert False, \"错误教师口令必须拒绝\"\n        except HTTPException as exc:\n            assert exc.status_code == 403\n    finally:\n        if old is None:\n            os.environ.pop(\"ACP_ADMIN_TOKEN\", None)\n        else:\n            os.environ[\"ACP_ADMIN_TOKEN\"] = old\n\n\n# ---- 会话回合锁（M1 D4-5）：串行化同会话并发，防丢写 + 堵 TOCTOU ----\n\n@test\ndef 会话锁_同会话串行无丢写():\n    import threading\n    import time\n    from app.concurrency import session_turn\n    # 模拟 history 的「读整块→改→写整块」竞态：无锁时并发会互相覆盖、丢元素\n    shared = {\"list\": []}\n\n    def worker(val):\n        with session_turn(\"sessX\"):\n            cur = list(shared[\"list\"])   # 读\n            time.sleep(0.02)             # 放大竞态窗口\n            shared[\"list\"] = cur + [val]  # 整块写回\n\n    threads = [threading.Thread(target=worker, args=(i,)) for i in range(5)]\n    for t in threads:\n        t.start()\n    for t in threads:\n        t.join()\n    assert len(shared[\"list\"]) == 5, f\"同会话应串行、无丢写，实际 {shared['list']}\"\n\n\n@test\ndef 会话锁_不同会话用不同锁():\n    from app.concurrency import _lock_for\n    assert _lock_for(\"a\") is _lock_for(\"a\")   # 同 id 复用同一把锁\n    assert _lock_for(\"a\") is not _lock_for(\"b\")  # 不同 id 不互相阻塞\n\n\n# ---- 沙箱隔离测试组（M1 D1）：显式打 docker 后端，验证学生代码逃不出容器。\n# docker 不可用时整组跳过——本地无 docker 的开发机照样能跑主套件。----\n\nclass _SkipTest(Exception):\n    \"\"\"docker 不可用：跳过该测试（不计失败）。\"\"\"\n\n\ndef _docker_run(code, timeout=4.0):\n    \"\"\"强制用 docker 后端跑一段代码；docker 不可用则抛 _SkipTest。\"\"\"\n    old = os.environ.get(\"ACP_SANDBOX\")\n    os.environ[\"ACP_SANDBOX\"] = \"auto\"\n    try:\n        if sandbox.active_backend() != \"docker\":\n            raise _SkipTest(\"docker 不可用\")\n        os.environ[\"ACP_SANDBOX\"] = \"docker\"\n        return sandbox.run_code(code, timeout)\n    finally:\n        if old is None:\n            os.environ.pop(\"ACP_SANDBOX\", None)\n        else:\n            os.environ[\"ACP_SANDBOX\"] = old\n\n\n@test\ndef 沙箱docker_正常代码照跑():\n    r = _docker_run(\"print(sum([1,2,3]))\")\n    assert r.stdout.strip() == \"6\" and not r.has_error\n\n\n@test\ndef 沙箱docker_读不到宿主密钥():\n    r = _docker_run(\"import os; print(os.environ.get('DEEPSEEK_API_KEY'))\")\n    assert r.stdout.strip() == \"None\"   # 容器环境里没有密钥\n\n\n@test\ndef 沙箱docker_读不到宿主acp_db():\n    r = _docker_run(\"import os; print(os.path.exists('/data/acp.db'), os.path.exists('acp.db'))\")\n    assert r.stdout.strip() == \"False False\"   # 宿主数据库不可见\n\n\n@test\ndef 沙箱docker_无网络():\n    r = _docker_run(\"import socket; socket.create_connection(('1.1.1.1',53),2); print('CONNECTED')\")\n    assert \"CONNECTED\" not in r.stdout and r.has_error   # 连不出去\n\n\n@test\ndef 沙箱docker_根文件系统只读():\n    r = _docker_run(\"open('/evil.txt','w').write('x'); print('WROTE')\")\n    assert \"WROTE\" not in r.stdout and r.has_error   # 根只读，写不进去\n\n\n@test\ndef 沙箱docker_死循环超时():\n    r = _docker_run(\"while True: pass\", timeout=3.0)\n    assert r.timed_out\n\n\n# ════════ 经历记忆（第二层记忆：结算沉淀 + 相关度召回 + 泄题门控）════════\n\ndef _exp_pattern(pid, name, root_cause=\"下标 3 越界\"):\n    p = dict(_good_candidate())\n    p.update(id=pid, name=name, root_cause=root_cause)\n    return p\n\n\ndef _exp_session(tutor, mine_engine, patterns, *, student_id, pattern_id,\n                 stage, mine_status=\"planted\", mode=None):\n    \"\"\"自定学生/题目的会话（_arena_session 硬编码 rt1，经历测试需隔离学生）。\"\"\"\n    mine_engine.get_pattern = lambda pid: patterns[pid]\n    mine_engine.load_patterns = lambda: patterns\n    from app.models import TutorSession\n    db = TestSession()\n    manifest = mine_engine.build_manifest(student_id, patterns[pattern_id])\n    if mode:\n        manifest[\"mode\"] = mode\n    sess = TutorSession(id=f\"s_exp_{__import__('uuid').uuid4().hex[:6]}\",\n                        student_id=student_id, pattern_id=pattern_id,\n                        manifest=manifest, history=[], stage=stage, mine_status=mine_status)\n    db.add(sess); db.commit()\n    return db, sess\n\n\n@test\ndef 经历检索_内容词召回且纯虚词不算():\n    from app.services import experience\n    s = experience.relevance_score(\"数组下标越界 IndexError\", \"他在《测试题》踩了 IndexError，根因是数组下标越界\")\n    assert s >= experience.RELEVANCE_FLOOR, f\"内容词高度重叠应过门槛: {s}\"\n    s2 = experience.relevance_score(\"我觉得可能是不知道怎么了\", \"他自己就是这个样子的没什么\")\n    assert s2 < experience.RELEVANCE_FLOOR, f\"纯虚词撞上不算相关: {s2}\"\n\n\n@test\ndef 经历检索_跑题记忆宁可空手():\n    from app.services import experience\n    s = experience.relevance_score(\"除以零 ZeroDivisionError 分母 平均数\",\n                                   \"在《测试题》遇到 IndexError，根因是下标 3 越界\")\n    assert s < experience.RELEVANCE_FLOOR, f\"不相关经历不应过门槛: {s}\"\n\n\n@test\ndef 经历沉淀_内化结算落库且摘要不含修复代码():\n    from app.models import StudentExperience\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        pats = {p[\"id\"]: p for p in [_exp_pattern(\"BP-EXP-001\", \"经历题甲\")]}\n        tutor._call_llm = _stub_turn(\n            tutor, internalization=tutor.InternalizationScore(cause=True, locate=True, prevent=False))\n        db, sess = _exp_session(tutor, mine_engine, pats, student_id=\"exp1\",\n                                pattern_id=\"BP-EXP-001\", stage=\"⑥内化\", mine_status=\"fixed\")\n        tutor.run_turn(db, sess, \"因为下标越界，定位靠追踪变量取值，下次先查边界\")\n        assert sess.status == \"completed\"\n        exp = db.query(StudentExperience).filter_by(student_id=\"exp1\").one()\n        assert \"经历题甲\" in exp.summary and \"下标 3 越界\" in exp.summary, exp.summary\n        assert \"成因机制\" in exp.summary and \"定位方法\" in exp.summary, exp.summary\n        # 落库端泄题硬门控：摘要与关键词绝不含 expected_fix 片段\n        fix = pats[\"BP-EXP-001\"][\"expected_fix\"]\n        assert fix not in exp.summary and all(fix not in k for k in exp.keywords), \"摘要泄了修复代码\"\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 经历召回_新题注入过往经历且同题排除():\n    from app.services import experience, mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        pats = {p[\"id\"]: p for p in [\n            _exp_pattern(\"BP-EXP-011\", \"经历题乙\"), _exp_pattern(\"BP-EXP-012\", \"经历题丙\")]}\n        # 先内化完成\"经历题乙\"沉淀一条经历\n        tutor._call_llm = _stub_turn(\n            tutor, internalization=tutor.InternalizationScore(cause=True, locate=True, prevent=True))\n        db, sess = _exp_session(tutor, mine_engine, pats, student_id=\"exp2\",\n                                pattern_id=\"BP-EXP-011\", stage=\"⑥内化\", mine_status=\"fixed\")\n        tutor.run_turn(db, sess, \"因为下标越界，定位靠追踪变量，下次先查边界\")\n\n        # 换一道同族新题：system 应注入过往经历（带硬规则），召回按相关度过门槛\n        captured = {}\n        stub = _stub_turn(tutor)\n        tutor._call_llm = lambda system, history: (captured.__setitem__(\"sys\", system), stub(system, history))[1]\n        db2, sess2 = _exp_session(tutor, mine_engine, pats, student_id=\"exp2\",\n                                  pattern_id=\"BP-EXP-012\", stage=\"①发现\")\n        tutor.run_turn(db2, sess2, \"程序好像不太对\")\n        assert \"过往共同经历\" in captured[\"sys\"] and \"经历题乙\" in captured[\"sys\"], \"应召回过往经历\"\n        assert \"绝不点破\" in captured[\"sys\"], \"注入段应携带泄题硬规则\"\n\n        # 同题重开：exclude_pattern_id 应排除，system 无经历段（变式不剧透）\n        captured.clear()\n        db3, sess3 = _exp_session(tutor, mine_engine, pats, student_id=\"exp2\",\n                                  pattern_id=\"BP-EXP-011\", stage=\"①发现\")\n        tutor.run_turn(db3, sess3, \"程序好像不太对\")\n        assert \"过往共同经历\" not in captured[\"sys\"], \"同题经历不得召回（防变式剧透）\"\n        db.close(); db2.close(); db3.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 经历隔离_coop会话不沉淀():\n    from app.services import experience, mine_engine, tutor\n    o1, o2 = mine_engine.get_pattern, mine_engine.load_patterns\n    try:\n        pats = {p[\"id\"]: p for p in [_exp_pattern(\"BP-EXP-021\", \"经历题丁\")]}\n        db, sess = _exp_session(tutor, mine_engine, pats, student_id=\"exp3\",\n                                pattern_id=\"BP-EXP-021\", stage=\"⑥内化\",\n                                mine_status=\"fixed\", mode=\"coop\")\n        assert experience.record_on_completion(db, sess) is None, \"coop 会话不得沉淀经历\"\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns = o1, o2\n\n\n@test\ndef 兜底升级_LLM挂掉时按阶段给确定性引导():\n    from app.services import mine_engine, tutor\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        def boom(system, history):\n            raise RuntimeError(\"api down\")\n        tutor._call_llm = boom\n        db, sess = _exp_session(tutor, mine_engine,\n                                {p[\"id\"]: p for p in [_exp_pattern(\"BP-EXP-031\", \"经历题戊\")]},\n                                student_id=\"exp4\", pattern_id=\"BP-EXP-031\", stage=\"③归因\")\n        sess.attribution_step = \"variable_trace\"\n        r = tutor.run_turn(db, sess, \"然后呢\")\n        assert tutor._SAFE_STAGE_QUESTIONS[\"③归因\"] in r[\"reply\"], r[\"reply\"]  # 不再只说\"走神\"\n        assert sess.stage == \"③归因\" and r[\"events_emitted\"] == [], \"兜底轮状态不变、不落事件\"\n        db.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\ndef _snap(db, session_id, seq, touched):\n    \"\"\"造一条只带 touched_mine_line 判定的快照（过程旁白测试用）。\"\"\"\n    db.add(CodeSnapshot(id=f\"cs_strug_{session_id}_{seq}\", session_id=session_id, seq=seq,\n                        code=\"x\", diff_stats={\"lines_changed\": 1, \"touched_mine_line\": touched},\n                        timestamp=f\"2026-07-16T00:0{seq}:00+00:00\"))\n\n\n@test\ndef 过程旁白_同题重开与试探开局信号():\n    from app.services import process\n    from app.models import TutorSession\n    db = TestSession()\n    # 同题重开：strug1 在 P1 上有 2 局没走完\n    for i, stage in enumerate((\"②定位\", \"②定位\")):\n        db.add(TutorSession(id=f\"s_strug_p1_{i}\", student_id=\"strug1\", pattern_id=\"P-STRUG-1\",\n                            manifest={\"mines\": []}, history=[], stage=stage, mine_status=\"planted\"))\n    # 试探开局惯性：另两题各有失败后两版才碰关键行的快照链\n    for i in range(2):\n        sid = f\"s_strug_o{i}\"\n        db.add(TutorSession(id=sid, student_id=\"strug1\", pattern_id=f\"P-STRUG-O{i}\",\n                            manifest={\"mines\": []}, history=[], stage=\"④修复\", mine_status=\"fixed\"))\n        for seq, touched in enumerate((False, False, True), start=1):\n            _snap(db, sid, seq, touched)\n    db.commit()\n    text = process.recent_struggle_signal(db, \"strug1\", \"P-STRUG-1\", exclude_session_id=\"cur\")\n    assert \"第 3 次\" in text and \"②定位\" in text, f\"应报同题重开与停留位置: {text}\"\n    assert \"先读证据\" in text, f\"应报试探开局惯性: {text}\"\n    assert process.recent_struggle_signal(db, \"strug_fresh\", \"P-STRUG-1\") == \"\", \"无历史必须零信号\"\n    # 只有 1 局没走完 → 不算重复卡点；定向开局的样本不凑惯性\n    db.add(TutorSession(id=\"s_strug_p2_0\", student_id=\"strug2\", pattern_id=\"P-STRUG-2\",\n                        manifest={\"mines\": []}, history=[], stage=\"②定位\", mine_status=\"planted\"))\n    sid = \"s_strug2_o0\"\n    db.add(TutorSession(id=sid, student_id=\"strug2\", pattern_id=\"P-STRUG-O9\",\n                        manifest={\"mines\": []}, history=[], stage=\"④修复\", mine_status=\"fixed\"))\n    for seq, touched in enumerate((True, False), start=1):\n        _snap(db, sid, seq, touched)\n    db.commit()\n    assert process.recent_struggle_signal(db, \"strug2\", \"P-STRUG-2\") == \"\", \"单局卡壳/定向开局不该报信号\"\n    db.close()\n\n\n@test\ndef 过程旁白_注入导师system且带硬规则():\n    from app.services import mine_engine, tutor\n    from app.models import TutorSession\n    o1, o2, o3 = mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm\n    try:\n        pats = {p[\"id\"]: p for p in [_exp_pattern(\"BP-STRUG-011\", \"旁白题甲\")]}\n        captured = {}\n        stub = _stub_turn(tutor)\n        tutor._call_llm = lambda system, history: (captured.__setitem__(\"sys\", system), stub(system, history))[1]\n        # 造 2 局没走完的历史，再开第 3 局\n        db, sess = _exp_session(tutor, mine_engine, pats, student_id=\"strug3\",\n                                pattern_id=\"BP-STRUG-011\", stage=\"①发现\")\n        for i in range(2):\n            db.add(TutorSession(id=f\"s_strug3_{i}\", student_id=\"strug3\", pattern_id=\"BP-STRUG-011\",\n                                manifest=sess.manifest, history=[], stage=\"②定位\", mine_status=\"planted\"))\n        db.commit()\n        tutor.run_turn(db, sess, \"程序好像不太对\")\n        assert \"过程旁白\" in captured[\"sys\"] and \"第 3 次\" in captured[\"sys\"], \"应注入跨局旁白\"\n        assert \"绝不向学生照念\" in captured[\"sys\"], \"旁白必须带不照念硬规则\"\n        assert \"不得据此跳过或放松任何阶段门控\" in captured[\"sys\"], \"旁白必须带门控硬规则\"\n        # 无历史学生：零注入，行为与现状一致\n        captured.clear()\n        db2, sess2 = _exp_session(tutor, mine_engine, pats, student_id=\"strug4\",\n                                  pattern_id=\"BP-STRUG-011\", stage=\"①发现\")\n        tutor.run_turn(db2, sess2, \"程序好像不太对\")\n        assert \"过程旁白\" not in captured[\"sys\"], \"无信号必须零注入\"\n        db.close(); db2.close()\n    finally:\n        mine_engine.get_pattern, mine_engine.load_patterns, tutor._call_llm = o1, o2, o3\n\n\n@test\ndef 足迹周报_窗口聚合卡点与进步():\n    from datetime import datetime, timedelta, timezone as _tz\n    from app.models import SessionMessage as SM, TutorSession as TS\n    recent_ts = datetime.now(_tz.utc).isoformat()\n    old_ts = (datetime.now(_tz.utc) - timedelta(days=30)).isoformat()\n    # 两个 Episode：一个本周已内化（快照证据=试探开局），一个 30 天前的（必须被窗口排除）\n    episodes = [\n        {\"pattern_id\": \"BP-NULL-002\", \"pattern_name\": \"方法返回None未检查\", \"session_id\": \"fp_s1\",\n         \"outcome\": \"已内化\", \"last_at\": recent_ts,\n         \"rounds\": [{\"day\": recent_ts[:10], \"attempts\": [{\"time\": recent_ts}, {\"time\": recent_ts}]}]},\n        {\"pattern_id\": \"BP-NULL-003\", \"pattern_name\": \"字典取值未判空\", \"session_id\": \"fp_s2\",\n         \"outcome\": \"已解决\", \"last_at\": old_ts,\n         \"rounds\": [{\"day\": old_ts[:10], \"attempts\": [{\"time\": old_ts}]}]},\n    ]\n    db = TestSession()\n    db.add(TS(id=\"fp_tut1\", student_id=\"fp1\", pattern_id=\"BP-NULL-002\",\n              manifest={\"mines\": []}, history=[], stage=\"②定位\", mine_status=\"planted\"))\n    db.add(SM(id=\"sm_fp1\", session_id=\"fp_tut1\", seq=1, role=\"tutor\", kind=\"chat\", content=\"x\",\n              meta={\"support_mode\": True, \"stage_before\": \"②定位\"}, created_at=recent_ts))\n    for seq, touched in enumerate((False, False, True), start=1):\n        _snap(db, \"fp_s1\", seq, touched)\n    db.commit()\n    fp = timeline.debug_footprint(db, \"fp1\", episodes)\n    assert fp[\"enough\"] and fp[\"attempts\"] == 2 and fp[\"patterns_touched\"] == 1, \\\n        f\"30 天前的尝试必须被窗口排除: {fp}\"\n    assert fp[\"internalized\"] == [\"方法返回None未检查\"], fp\n    assert fp[\"stuck_step\"] == \"定位原因\" and fp[\"stuck_count\"] == 1, \"卡点应翻成四步语言\"\n    assert fp[\"targeted_progress\"] == [\"方法返回None未检查\"], \"试探开局且解决的题应报进步\"\n    assert fp[\"loosened\"] and \"正常\" in fp[\"loosened\"][0][\"name\"], \"高频簇内化应报松动\"\n    assert timeline.debug_footprint(db, \"fp_empty\", [])[\"enough\"] is False, \"无活动 enough=False\"\n    db.close()\n\n\n# 认知先验层测试组：在 test_priors.py 独立维护（假 LLM、零外部依赖、可单跑），\n# 这里并入主套件一起跑。顶部已设 ACP_PRIORS=off，结算路径不打真网。\nimport test_priors as _priors  # noqa: E402\n_tests.extend(_priors._tests)\n\n\ndef main():\n    passed = failed = skipped = 0\n    for fn in _tests:\n        try:\n            fn()\n            print(f\"[PASS] {fn.__name__}\")\n            passed += 1\n        except _SkipTest as e:\n            print(f\"[SKIP] {fn.__name__}: {e}\")\n            skipped += 1\n        except Exception as e:\n            print(f\"[FAIL] {fn.__name__}: {e}\")\n            traceback.print_exc()\n            failed += 1\n    print(f\"\\n{passed} 过 / {failed} 败 / {skipped} 跳过 / 共 {len(_tests)}\")\n    sys.exit(1 if failed else 0)\n\n\nif __name__ == \"__main__\":\n    main()\n"}}}}