refactor(intelligent-eval): 可见性接缝收敛（Phase 1）

将「已删即 404」语义收进 IntelligentEvalRepository 单一接缝，消除三处独立裁决；
任务监控开始隐藏已删评估的任务（本 Phase 唯一刻意行为变化）。

- repository.py 新增 visible() 谓词与 require_live_eval() 服务接缝；
  六处裸谓词统一走它，get()/get_including_deleted() 语义不变。
- decision_logs.py 删除本地 _require_eval，三处调用迁至 repository 接缝；
  count_decisions 由 len(.all()) 改为 func.count。
- task_queue.py list_tasks 与 stats 过滤已删评估的任务（行为变化）。
- web/routers/intelligent_evals.py: _require_eval_exists → _require_live_eval，
  把 LookupError 翻译为 404；expired 会话 Markdown 标注下沉至
  read_model.report_markdown_by_eval；配置快照 11 字段序列化收至
  config_snapshot.snapshot_to_dict 单一出口。
- AGENTS.md 登记可见性纪律（已知陷阱 #6）。
- 补 characterization 测试锁定四处契约；更新 task_queue 测试以使用
  真实 eval_id（可见性过滤后字面 eval_id 不再可见）。

AGENTS.md

@@ -197,3 +197,5 @@ tests/
 4. **SQLite 连接池**：必须用 `StaticPool` + `check_same_thread=False`。使用默认连接池会导致多线程下连接耗尽（async + FastAPI 线程池混合场景）。
 
 5. **`.env` 在 t480 上需手动维护**：部署脚本不会自动同步 `.env` 文件，新增环境变量需 SSH 到主机手动补全。容易导致 OpenClaw 代理等功能静默失效。
+
+6. **智能评估可见性接缝**：逻辑删除（`deleted` 状态）语义只住在 `intelligent_eval/repository.py`——查询走 `IntelligentEvalRepository.visible()` 谓词，存在性校验走 `require_live_eval`（路由层翻译为 404）。禁止在其他模块直查 `IntelligentEvalDB` 判断删除态，否则删除语义会在多处漂移（Phase 1 收敛前的教训：快照、决策日志、任务监控各自实现了一遍"已删即 404"）。

backend/agenteval/intelligent_eval/config_snapshot.py

@@ -70,6 +70,23 @@ def get_snapshot(snapshot_id: str, session: Session) -> Optional[IntelligentEval
     return session.get(IntelligentEvalConfigSnapshotDB, snapshot_id)
 
 
+def snapshot_to_dict(s: IntelligentEvalConfigSnapshotDB) -> dict[str, Any]:
+    """快照的稳定序列化形状（列表与详情共用的单一出口）。"""
+    return {
+        "id": s.id,
+        "eval_id": s.eval_id,
+        "snapshot_type": s.snapshot_type,
+        "goal": s.goal,
+        "seeds": s.get_seeds(),
+        "intent": s.intent,
+        "role_description": s.role_description,
+        "time_window_hours": s.time_window_hours,
+        "plan": s.get_plan(),
+        "created_at": s.created_at.isoformat() if s.created_at else None,
+        "created_by": s.created_by,
+    }
+
+
 def compare_snapshots(
     snapshot1: IntelligentEvalConfigSnapshotDB,
     snapshot2: IntelligentEvalConfigSnapshotDB,

backend/agenteval/intelligent_eval/decision_logs.py

@@ -3,11 +3,14 @@
 Pulled out of ``web/routers/intelligent_evals.py`` so the router only handles
 HTTP validation and error translation. The ORM writes and reads now live here.
 """
+
 from typing import Any
 
+from sqlalchemy import func
 from sqlmodel import Session, select
 
 from agenteval.intelligent_eval.models import IntelligentEvalStatus
+from agenteval.intelligent_eval.repository import IntelligentEvalRepository
 from agenteval.storage.db import (
     IntelligentEvalDB,
     IntelligentEvalDecisionLogDB,
@@ -27,12 +30,6 @@ def _log_to_dict(log: IntelligentEvalDecisionLogDB) -> dict:
     }
 
 
-def _require_eval(eval_id: str, session: Session) -> None:
-    eval_db = session.get(IntelligentEvalDB, eval_id)
-    if eval_db is None or eval_db.status == IntelligentEvalStatus.DELETED.value:
-        raise LookupError(f"intelligent eval {eval_id} not found")
-
-
 def _append_row(
     eval_id: str,
     decision_type: str,
@@ -71,7 +68,7 @@ def create_decision_log(
 
     Raises ``LookupError`` if eval not found.
     """
-    _require_eval(eval_id, session)
+    IntelligentEvalRepository(session).require_live_eval(eval_id)
 
     # Dedupe: same (eval, decision_type, context) → return existing.
     # Limit to 100 records to avoid loading too many into memory; in practice,
@@ -105,25 +102,25 @@ def append_decision_log(
 
     Raises ``LookupError`` if eval not found.
     """
-    _require_eval(eval_id, session)
+    IntelligentEvalRepository(session).require_live_eval(eval_id)
     return _append_row(eval_id, decision_type, reason, cron_id, context, session)
 
 
 def count_decisions(eval_id: str, decision_type: str, session: Session) -> int:
     """该评估某类型决策日志的条数（平台闸门计数与 attempt 落账的计数原语）。"""
-    return len(
+    return session.exec(
-        session.exec(
+        select(func.count())
-            select(IntelligentEvalDecisionLogDB).where(
+        .select_from(IntelligentEvalDecisionLogDB)
-                IntelligentEvalDecisionLogDB.eval_id == eval_id,
+        .where(
-                IntelligentEvalDecisionLogDB.decision_type == decision_type,
+            IntelligentEvalDecisionLogDB.eval_id == eval_id,
-            )
+            IntelligentEvalDecisionLogDB.decision_type == decision_type,
-        ).all()
+        )
-    )
+    ).one()
 
 
 def list_decision_logs(eval_id: str, session: Session) -> list[dict]:
     """List decision logs for an eval. Raises ``LookupError`` if eval not found."""
-    _require_eval(eval_id, session)
+    IntelligentEvalRepository(session).require_live_eval(eval_id)
 
     logs = session.exec(
         select(IntelligentEvalDecisionLogDB)
@@ -252,9 +249,7 @@ def supplement_decision_logs(session: Session) -> int:
     for ev in evals:
         plan = ev.get_plan() if ev.plan else {}
         estimated = plan.get("estimated_sessions", 0)
-        sessions = session.exec(
+        sessions = session.exec(select(IntelligentEvalSessionDB).where(IntelligentEvalSessionDB.eval_id == ev.id)).all()
-            select(IntelligentEvalSessionDB).where(IntelligentEvalSessionDB.eval_id == ev.id)
-        ).all()
         completed = sum(1 for s in sessions if s.status == "completed")
         types = {
             x.decision_type

backend/agenteval/intelligent_eval/read_model.py

@@ -22,6 +22,7 @@ from agenteval.intelligent_eval.models import (
     IntelligentEvalSessionStatus,
     IntelligentEvalStatus,
 )
+from agenteval.intelligent_eval.report import render_report_markdown
 from agenteval.intelligent_eval.repository import (
     IntelligentEvalMessageRepository,
     IntelligentEvalRepository,
@@ -345,3 +346,19 @@ class IntelligentEvalReadModel:
         if eval_obj is None or eval_obj.report is None:
             return None
         return eval_obj.name, eval_obj.report
+
+    def report_markdown_by_eval(self, eval_id: str) -> str | None:
+        """报告的 Markdown 投影（含 expired 会话的不完整证据标注，ADR-0011）。"""
+
+        report = self.report_by_eval(eval_id)
+        if report is None:
+            return None
+        name, payload = report
+        markdown = render_report_markdown(payload, name=name, eval_id=eval_id)
+        expired = [s for s in self._sessions.list_by_eval(eval_id) if s.status is IntelligentEvalSessionStatus.EXPIRED]
+        if expired:
+            lines = ["", "## 不完整证据会话", ""]
+            for s in expired:
+                lines.append(f"- 会话 `{s.id}`（{s.turn_count} 轮）：60 分钟无新轮次过期，证据不完整")
+            markdown = markdown.rstrip() + "\n" + "\n".join(lines) + "\n"
+        return markdown

backend/agenteval/intelligent_eval/repository.py

@@ -97,26 +97,25 @@ class IntelligentEvalRepository:
             completed_at=db.completed_at,
         )
 
+    @staticmethod
+    def visible():
+        """可见性谓词：逻辑删除的评估默认不可见（查询位点共用接缝）。"""
+        return IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value
+
+    @staticmethod
+    def _is_visible(db: IntelligentEvalDB) -> bool:
+        return db.status != IntelligentEvalStatus.DELETED.value
+
     def list_all(self) -> list[IntelligentEval]:
-        statement = (
+        statement = select(IntelligentEvalDB).where(self.visible()).order_by(IntelligentEvalDB.created_at.desc())
-            select(IntelligentEvalDB)
-            .where(IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value)
-            .order_by(IntelligentEvalDB.created_at.desc())
-        )
         return [self._from_db(r) for r in self.session.exec(statement).all()]
 
-    def list_page(
+    def list_page(self, offset: int, limit: int, status: Optional[str] = None) -> list[IntelligentEval]:
-        self, offset: int, limit: int, status: Optional[str] = None
-    ) -> list[IntelligentEval]:
         """Return one page of evals (created_at desc), optionally filtered by status.
 
         Deleted evals are always excluded — the list API has no recycle bin.
         """
-        statement = (
+        statement = select(IntelligentEvalDB).where(self.visible()).order_by(IntelligentEvalDB.created_at.desc())
-            select(IntelligentEvalDB)
-            .where(IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value)
-            .order_by(IntelligentEvalDB.created_at.desc())
-        )
         if status:
             statement = statement.where(IntelligentEvalDB.status == status)
         statement = statement.offset(offset).limit(limit)
@@ -124,17 +123,13 @@ class IntelligentEvalRepository:
 
     def count(self) -> int:
         """Total number of evaluations (for pagination metadata)."""
-        return self.session.exec(
+        return self.session.exec(select(func.count()).select_from(IntelligentEvalDB).where(self.visible())).one()
-            select(func.count())
-            .select_from(IntelligentEvalDB)
-            .where(IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value)
-        ).one()
 
     def count_by_status(self) -> dict[str, int]:
         """Count evaluations per status (for the list page stat bar)."""
         rows = self.session.exec(
             select(IntelligentEvalDB.status, func.count(IntelligentEvalDB.id))
-            .where(IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value)
+            .where(self.visible())
             .group_by(IntelligentEvalDB.status)
         ).all()
         stats = {status.value: 0 for status in IntelligentEvalStatus if status != IntelligentEvalStatus.DELETED}
@@ -144,10 +139,17 @@ class IntelligentEvalRepository:
 
     def get(self, eval_id: str) -> Optional[IntelligentEval]:
         db = self.session.get(IntelligentEvalDB, eval_id)
-        if db is None or db.status == IntelligentEvalStatus.DELETED.value:
+        if db is None or not self._is_visible(db):
             return None
         return self._from_db(db)
 
+    def require_live_eval(self, eval_id: str) -> IntelligentEval:
+        """可见性单一接缝：存在且未删 → 返回实体，否则 ``LookupError``。"""
+        ev = self.get(eval_id)
+        if ev is None:
+            raise LookupError(f"intelligent eval {eval_id} not found")
+        return ev
+
     def get_including_deleted(self, eval_id: str) -> Optional[IntelligentEval]:
         """包含已删除评估的原样读取（供 delete 幂等检查使用）。"""
         db = self.session.get(IntelligentEvalDB, eval_id)
@@ -168,7 +170,7 @@ class IntelligentEvalRepository:
             )
             .where(
                 IntelligentEvalDB.id == eval_id,
-                IntelligentEvalDB.status != IntelligentEvalStatus.DELETED.value,
+                self.visible(),
             )
             .order_by(IntelligentEvalSessionDB.created_at.asc())
         )
@@ -285,6 +287,7 @@ class IntelligentEvalRepository:
             },
         )
 
+
 class IntelligentEvalSessionRepository:
     """CRUD for intelligent eval sessions."""
 
@@ -443,6 +446,7 @@ class IntelligentEvalSessionRepository:
             return CompareAndSetStatus.NOT_FOUND, None
         return CompareAndSetStatus.APPLIED, self._from_db(db)
 
+
 class IntelligentEvalMessageRepository:
     """CRUD for intelligent eval session messages."""
 

backend/agenteval/intelligent_eval/task_queue.py

@@ -13,6 +13,7 @@ from typing import Optional
 from sqlmodel import Session, func, select, update
 
 from agenteval.intelligent_eval.models import IntelligentEvalStatus
+from agenteval.intelligent_eval.repository import IntelligentEvalRepository
 from agenteval.storage.db import (
     IntelligentEvalDB,
     IntelligentEvalTaskQueueDB,
@@ -259,9 +260,7 @@ def settle_tasks_for_finished_evals(session: Session) -> int:
     from agenteval.storage.db import IntelligentEvalTaskQueueDB
 
     tasks = session.exec(
-        select(IntelligentEvalTaskQueueDB).where(
+        select(IntelligentEvalTaskQueueDB).where(IntelligentEvalTaskQueueDB.status.in_(["pending", "assigned"]))
-            IntelligentEvalTaskQueueDB.status.in_(["pending", "assigned"])
-        )
     ).all()
 
     settled = 0
@@ -305,7 +304,14 @@ def list_tasks(
     Returns:
         {"tasks": [...], "stats": {pending, assigned, completed, failed, unresolved}}
     """
-    stmt = select(IntelligentEvalTaskQueueDB).order_by(IntelligentEvalTaskQueueDB.created_at.desc())
+    # 可见性接缝：已删评估的任务不出现在监控列表与统计中
+    live_eval_ids = select(IntelligentEvalDB.id).where(IntelligentEvalRepository.visible())
+
+    stmt = (
+        select(IntelligentEvalTaskQueueDB)
+        .where(IntelligentEvalTaskQueueDB.eval_id.in_(live_eval_ids))
+        .order_by(IntelligentEvalTaskQueueDB.created_at.desc())
+    )
     if status:
         stmt = stmt.where(IntelligentEvalTaskQueueDB.status == status)
     if eval_id:
@@ -318,7 +324,9 @@ def list_tasks(
         select(
             IntelligentEvalTaskQueueDB.status,
             func.count(IntelligentEvalTaskQueueDB.id),
-        ).group_by(IntelligentEvalTaskQueueDB.status)
+        )
+        .where(IntelligentEvalTaskQueueDB.eval_id.in_(live_eval_ids))
+        .group_by(IntelligentEvalTaskQueueDB.status)
     ).all()
     for status_val, cnt in rows:
         if status_val in stats:

backend/agenteval/web/routers/intelligent_evals.py

@@ -17,10 +17,9 @@ from agenteval.intelligent_eval.lifecycle import (
     IntelligentEvalNotFoundError,
     IntelligentEvalTransitionError,
 )
-from agenteval.intelligent_eval.models import IntelligentEvalSessionStatus
 from agenteval.intelligent_eval.read_model import IntelligentEvalReadModel
-from agenteval.intelligent_eval.report import ReportModel, render_report_markdown
+from agenteval.intelligent_eval.report import ReportModel
-from agenteval.intelligent_eval.repository import IntelligentEvalSessionRepository
+from agenteval.intelligent_eval.repository import IntelligentEvalRepository
 from agenteval.web.deps import get_db
 
 router = APIRouter()
@@ -70,14 +69,12 @@ def _translate(exc: Exception) -> HTTPException:
     return HTTPException(status_code=409, detail=exc.reason)
 
 
-def _require_eval_exists(session: Session, eval_id: str) -> None:
+def _require_live_eval(session: Session, eval_id: str) -> None:
-    """404 when the eval is missing or logically deleted."""
+    """把可见性接缝翻译成 HTTP：缺失或已删 → 404。"""
-    from agenteval.intelligent_eval.models import IntelligentEvalStatus
+    try:
-    from agenteval.storage.db import IntelligentEvalDB
+        IntelligentEvalRepository(session).require_live_eval(eval_id)
-
+    except LookupError as exc:
-    eval_db = session.get(IntelligentEvalDB, eval_id)
+        raise HTTPException(status_code=404, detail=str(exc)) from exc
-    if eval_db is None or eval_db.status == IntelligentEvalStatus.DELETED.value:
-        raise HTTPException(status_code=404, detail=f"intelligent eval {eval_id} not found")
 
 
 def _eval_response(ev, session: Session) -> dict:
@@ -234,22 +231,9 @@ async def get_report(eval_id: str, session: Session = Depends(get_db)) -> dict:
 
 @router.get("/{eval_id}/report/markdown", response_class=PlainTextResponse)
 async def get_report_markdown(eval_id: str, session: Session = Depends(get_db)) -> PlainTextResponse:
-    report = IntelligentEvalReadModel(session).report_by_eval(eval_id)
+    markdown = IntelligentEvalReadModel(session).report_markdown_by_eval(eval_id)
-    if report is None:
+    if markdown is None:
         raise HTTPException(status_code=404, detail="report not submitted yet")
-    name, payload = report
-    markdown = render_report_markdown(payload, name=name, eval_id=eval_id)
-    # ADR-0011：expired 会话按不完整证据标注，提醒读者结论证据不全
-    expired_sessions = [
-        s
-        for s in IntelligentEvalSessionRepository(session).list_by_eval(eval_id)
-        if s.status == IntelligentEvalSessionStatus.EXPIRED
-    ]
-    if expired_sessions:
-        lines = ["", "## 不完整证据会话", ""]
-        for s in expired_sessions:
-            lines.append(f"- 会话 `{s.id}`（{s.turn_count} 轮）：60 分钟无新轮次过期，证据不完整")
-        markdown = markdown.rstrip() + "\n" + "\n".join(lines) + "\n"
     return PlainTextResponse(markdown, media_type="text/markdown; charset=utf-8")
 
 
@@ -393,27 +377,10 @@ async def list_config_snapshots(eval_id: str, session: Session = Depends(get_db)
     """List all config snapshots for an evaluation."""
     from agenteval.intelligent_eval import config_snapshot
 
-    _require_eval_exists(session, eval_id)
+    _require_live_eval(session, eval_id)
 
     snapshots = config_snapshot.list_snapshots(eval_id, session)
-    return {
+    return {"snapshots": [config_snapshot.snapshot_to_dict(s) for s in snapshots]}
-        "snapshots": [
-            {
-                "id": s.id,
-                "eval_id": s.eval_id,
-                "snapshot_type": s.snapshot_type,
-                "goal": s.goal,
-                "seeds": s.get_seeds(),
-                "intent": s.intent,
-                "role_description": s.role_description,
-                "time_window_hours": s.time_window_hours,
-                "plan": s.get_plan(),
-                "created_at": s.created_at.isoformat() if s.created_at else None,
-                "created_by": s.created_by,
-            }
-            for s in snapshots
-        ]
-    }
 
 
 @router.get("/{eval_id}/config-snapshots/{snapshot_id}")
@@ -421,25 +388,13 @@ async def get_config_snapshot(eval_id: str, snapshot_id: str, session: Session =
     """Get a single config snapshot."""
     from agenteval.intelligent_eval import config_snapshot
 
-    _require_eval_exists(session, eval_id)
+    _require_live_eval(session, eval_id)
 
     snapshot = config_snapshot.get_snapshot(snapshot_id, session)
     if snapshot is None or snapshot.eval_id != eval_id:
         raise HTTPException(status_code=404, detail=f"snapshot {snapshot_id} not found")
 
-    return {
+    return config_snapshot.snapshot_to_dict(snapshot)
-        "id": snapshot.id,
-        "eval_id": snapshot.eval_id,
-        "snapshot_type": snapshot.snapshot_type,
-        "goal": snapshot.goal,
-        "seeds": snapshot.get_seeds(),
-        "intent": snapshot.intent,
-        "role_description": snapshot.role_description,
-        "time_window_hours": snapshot.time_window_hours,
-        "plan": snapshot.get_plan(),
-        "created_at": snapshot.created_at.isoformat() if snapshot.created_at else None,
-        "created_by": snapshot.created_by,
-    }
 
 
 class CompareSnapshotsRequest(BaseModel):
@@ -456,7 +411,7 @@ async def compare_config_snapshots(
     """Compare two config snapshots and return differences."""
     from agenteval.intelligent_eval import config_snapshot
 
-    _require_eval_exists(session, eval_id)
+    _require_live_eval(session, eval_id)
 
     # Get both snapshots
     snapshot1 = config_snapshot.get_snapshot(request.snapshot_id_1, session)

tests/conftest.py

@@ -37,6 +37,7 @@ def db_session(tmp_db_path: Path) -> Session:
         EvalTargetDB,
         ExplorationMessageDB,
         ExplorationSessionDB,
+        IntelligentEvalConfigSnapshotDB,
         IntelligentEvalDB,
         IntelligentEvalDecisionLogDB,
         IntelligentEvalMessageDB,

tests/integration/conftest.py

@@ -0,0 +1,8 @@
+"""Integration-level shared fixtures.
+
+Re-exports the in-process app fixtures from ``test_intelligent_evals_api``
+so newer integration test modules can consume them without importing
+fixture objects into test-module namespaces (which ruff flags as F811).
+"""
+
+from .test_intelligent_evals_api import client, seeded_db  # noqa: F401

tests/integration/test_intelligent_eval_task_queue_api.py

@@ -257,8 +257,9 @@ def test_list_tasks(client: TestClient, db_session: Session):
 
 def test_list_tasks_status_filter(client: TestClient, db_session: Session):
     """Status filter narrows the task list."""
-    db_session.add(IntelligentEvalTaskQueueDB(eval_id="eval1", status="pending", priority=1, reason="slot_due"))
+    eval_db = _make_eval(db_session, "status-filter-eval", days_ago=0)
-    db_session.add(IntelligentEvalTaskQueueDB(eval_id="eval1", status="failed", priority=1, reason="slot_due"))
+    db_session.add(IntelligentEvalTaskQueueDB(eval_id=eval_db.id, status="pending", priority=1, reason="slot_due"))
+    db_session.add(IntelligentEvalTaskQueueDB(eval_id=eval_db.id, status="failed", priority=1, reason="slot_due"))
     db_session.commit()
 
     response = client.get("/api/intelligent-evals/tasks?status=failed")

tests/integration/test_intelligent_eval_visibility_characterization.py

@@ -0,0 +1,204 @@
+"""Characterization tests: intelligent-eval visibility semantics (Phase 1).
+
+锁定逻辑删除可见性收敛（require_live_eval 接缝）后的行为契约。覆盖四处：
+
+1. 「已删即 404」的统一裁决：config-snapshots 详情/对比端点与
+   decision_logs 服务层都经 require_live_eval 接缝。
+2. 任务队列监控**隐藏**已删评估的任务（Phase 1.6 唯一的刻意行为变化，
+   锁定变更后的新契约）。
+3. config-snapshots 端点的响应形状（11 字段序列化，snapshot_to_dict 单一出口）。
+4. expired 会话在 Markdown 报告中的「不完整证据」标注文本（ADR-0011）。
+"""
+
+import pytest
+from agenteval.intelligent_eval.models import IntelligentEvalStatus
+from agenteval.storage.db import (
+    IntelligentEvalSessionDB,
+    IntelligentEvalTaskQueueDB,
+    utc_now,
+)
+from sqlmodel import select
+
+from .test_intelligent_evals_api import (
+    _create_eval,
+    _create_executing_eval,
+    _create_session,
+    _report_payload,
+    _submit_plan,
+)
+
+
+async def _complete_eval(client, eval_id: str) -> None:
+    resp = await client.put(f"/api/intelligent-evals/{eval_id}/report", json={"report": _report_payload()})
+    assert resp.status_code == 200, resp.text
+
+
+class TestSnapshotEndpointDeletedSemantics:
+    """路由 _require_eval_exists 路径：config-snapshots 详情与对比。"""
+
+    async def test_deleted_eval_snapshot_detail_returns_404(self, client, seeded_db):
+        eval_id = await _create_executing_eval(client)
+        await _complete_eval(client, eval_id)
+        snapshots = (await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots")).json()["snapshots"]
+        assert snapshots
+        await client.delete(f"/api/intelligent-evals/{eval_id}")
+
+        resp = await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots/{snapshots[0]['id']}")
+        assert resp.status_code == 404
+
+    async def test_deleted_eval_snapshot_compare_returns_404(self, client, seeded_db):
+        eval_id = (await _create_eval(client))["id"]
+        await _submit_plan(client, eval_id)  # second snapshot (plan_submitted)
+        await client.post(f"/api/intelligent-evals/{eval_id}/approve")
+        await _complete_eval(client, eval_id)
+        snapshots = (await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots")).json()["snapshots"]
+        assert len(snapshots) >= 2
+        await client.delete(f"/api/intelligent-evals/{eval_id}")
+
+        resp = await client.post(
+            f"/api/intelligent-evals/{eval_id}/config-snapshots/compare",
+            json={"snapshot_id_1": snapshots[0]["id"], "snapshot_id_2": snapshots[1]["id"]},
+        )
+        assert resp.status_code == 404
+
+
+class TestDecisionLogServiceDeletedSemantics:
+    """decision_logs 经 require_live_eval 接缝：已删评估一律 LookupError。"""
+
+    async def test_deleted_eval_service_calls_raise_lookup_error(self, client, seeded_db):
+        from agenteval.intelligent_eval import decision_logs
+
+        eval_id = await _create_executing_eval(client)
+        await _complete_eval(client, eval_id)
+        await client.delete(f"/api/intelligent-evals/{eval_id}")
+
+        with pytest.raises(LookupError):
+            decision_logs.list_decision_logs(eval_id, seeded_db)
+        with pytest.raises(LookupError):
+            decision_logs.create_decision_log(eval_id, "wait", "late report", "w-1", {}, seeded_db)
+        with pytest.raises(LookupError):
+            decision_logs.append_decision_log(eval_id, "wait", "late bookkeeping", "platform", {}, seeded_db)
+
+    async def test_repo_get_hides_deleted_but_including_deleted_returns_it(self, client, seeded_db):
+        from agenteval.intelligent_eval.repository import IntelligentEvalRepository
+
+        eval_id = await _create_executing_eval(client)
+        await _complete_eval(client, eval_id)
+        await client.delete(f"/api/intelligent-evals/{eval_id}")
+
+        repo = IntelligentEvalRepository(seeded_db)
+        assert repo.get(eval_id) is None
+        hidden = repo.get_including_deleted(eval_id)
+        assert hidden is not None
+        assert hidden.status is IntelligentEvalStatus.DELETED
+
+
+class TestTaskMonitorHidesDeletedEvals:
+    """Phase 1.6 刻意行为变化：任务监控不再泄漏已删评估的任务。"""
+
+    async def test_list_tasks_hides_tasks_of_deleted_eval(self, client, seeded_db):
+        eval_id = await _create_executing_eval(client)
+        seeded_db.add(
+            IntelligentEvalTaskQueueDB(
+                eval_id=eval_id,
+                status="completed",
+                priority=1,
+                reason="slot_due",
+                completed_at=utc_now(),
+            )
+        )
+        seeded_db.commit()
+
+        await _complete_eval(client, eval_id)
+        await client.delete(f"/api/intelligent-evals/{eval_id}")
+
+        resp = await client.get("/api/intelligent-evals/tasks")
+        assert resp.status_code == 200
+        data = resp.json()
+
+        assert [t for t in data["tasks"] if t["eval_id"] == eval_id] == []
+        assert data["stats"]["completed"] == 0
+
+        by_eval = await client.get("/api/intelligent-evals/tasks", params={"eval_id": eval_id})
+        assert by_eval.json()["tasks"] == []
+
+
+class TestSnapshotResponseShape:
+    """锁定 config-snapshots 列表/详情的 11 字段序列化形状。"""
+
+    SNAPSHOT_KEYS = {
+        "id",
+        "eval_id",
+        "snapshot_type",
+        "goal",
+        "seeds",
+        "intent",
+        "role_description",
+        "time_window_hours",
+        "plan",
+        "created_at",
+        "created_by",
+    }
+
+    async def test_list_shape_and_snapshot_types(self, client, seeded_db):
+        eval_id = (await _create_eval(client))["id"]
+        await _submit_plan(client, eval_id)
+
+        snapshots = (await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots")).json()["snapshots"]
+        assert len(snapshots) == 2
+        assert {s["snapshot_type"] for s in snapshots} == {"created", "plan_submitted"}
+        for s in snapshots:
+            assert set(s.keys()) == self.SNAPSHOT_KEYS
+            assert s["eval_id"] == eval_id
+        created = next(s for s in snapshots if s["snapshot_type"] == "created")
+        assert created["created_by"] == "user"
+        assert created["goal"] == "评估退货流程处理能力"
+        assert created["time_window_hours"] == 24
+        assert created["plan"] is None
+
+    async def test_detail_shape_matches_list_item(self, client, seeded_db):
+        eval_id = (await _create_eval(client))["id"]
+        listed = (await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots")).json()["snapshots"][0]
+
+        detail = (await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots/{listed['id']}")).json()
+        assert set(detail.keys()) == self.SNAPSHOT_KEYS
+        assert detail == listed
+
+    async def test_snapshot_of_other_eval_returns_404(self, client, seeded_db):
+        eval_id = (await _create_eval(client))["id"]
+        other_id = (await _create_eval(client, name="另一个评估"))["id"]
+        foreign = (await client.get(f"/api/intelligent-evals/{other_id}/config-snapshots")).json()["snapshots"][0]
+
+        resp = await client.get(f"/api/intelligent-evals/{eval_id}/config-snapshots/{foreign['id']}")
+        assert resp.status_code == 404
+
+
+class TestExpiredSessionMarkdownAnnotation:
+    """ADR-0011：expired 会话在 Markdown 报告尾部标注不完整证据。"""
+
+    async def test_markdown_annotates_expired_sessions(self, client, seeded_db):
+        eval_id = await _create_executing_eval(client)
+        session = await _create_session(client, eval_id)
+
+        row = seeded_db.exec(select(IntelligentEvalSessionDB).where(IntelligentEvalSessionDB.id == session["id"])).one()
+        row.status = "expired"
+        row.turn_count = 3
+        row.closed_at = utc_now()
+        seeded_db.add(row)
+        seeded_db.commit()
+
+        await _complete_eval(client, eval_id)
+
+        resp = await client.get(f"/api/intelligent-evals/{eval_id}/report/markdown")
+        assert resp.status_code == 200
+        text = resp.text
+        assert "## 不完整证据会话" in text
+        assert f"- 会话 `{session['id']}`（3 轮）：60 分钟无新轮次过期，证据不完整" in text
+
+    async def test_markdown_without_expired_sessions_has_no_annotation(self, client, seeded_db):
+        eval_id = await _create_executing_eval(client)
+        await _complete_eval(client, eval_id)
+
+        resp = await client.get(f"/api/intelligent-evals/{eval_id}/report/markdown")
+        assert resp.status_code == 200
+        assert "不完整证据会话" not in resp.text