diff --git a/_AI_USAGE_LOG.md b/_AI_USAGE_LOG.md
index 950d773..a18f02e 100644
--- a/_AI_USAGE_LOG.md
+++ b/_AI_USAGE_LOG.md
@@ -78,3 +78,4 @@
 | 2026-08-11 | Agent 实现 | T9（架构审查整改）：CJK 保守 token 估算（Issue11）。token.py approximate_token_count 重写：新增 _is_cjk_char（CJK 统一表意/扩展A/假名/韩文/兼容/全角六大 Unicode 范围）+ _CJK_TOKENS_PER_CHAR=1.5（旧逻辑 4 字符 1 token 对中文/日文严重低估，裁剪失效致 API 超限）；CJK 字符按 1.5 token/字符，其余仍 4 字符 1 token，最少 1 token；同步 config-design.md token_estimation 注释；新增 4 用例（纯 CJK 保守/ASCII 不回归/混合文本/全角符号）；TDD 验证 RED（（設計）4 字符仅 1 token）→ GREEN（聚焦 11 passed）→ 全量 186 passed 覆盖 100.00%（995 stmts/252 br），fail_under=99 达标 | src/genesis/inference/token.py, tests/test_inference_token.py, docs/config-design.md, _AI_USAGE_LOG.md | deepseek-v4-flash-free |
 | 2026-08-11 | Agent 实现 | T5（架构审查整改）：删死配置 + 同步文档（Issue5 + OV1）。config.py 删除 QdrantStoreConfig 类与 VectorStoreConfig.qdrant 字段、AppConfig.task_queue.redis_url；同步更新 6 处文档（api-design §1/§4.3/§5.2/§5.3/§6.2：TaskQueue 标注 v1 仅 InMemory、Redis/Valkey 为 v2 预留；rag-layer §9：Storage Adapter 仅 ChromaAdapter、移除切换流程/工厂 qdrant 分支；agent-runtime §3.1/§3.5；design §5.5/§8.4.1；config-design env/app.yaml/rag.yaml/docker compose；web-ui §4.1）+ tests/fixtures/rag.yaml 去 qdrant 段；历史评审记录（design-review/web-ui-review/phase1 plan）保留原样不改写；新增 3 用例（QdrantStoreConfig 已删/vector_store 无 qdrant 字段/task_queue 无 redis_url）+ 同步 2 个既有 qdrant 依赖用例；TDD 验证 RED（三处死配置存在）→ GREEN（聚焦 10 passed）→ 全量 189 passed 覆盖 100.00%（991 stmts/252 br），fail_under=99 达标 | src/genesis/config.py, tests/test_config.py, tests/fixtures/rag.yaml, docs/api-design.md, docs/rag-layer-design.md, docs/agent-runtime-design.md, docs/design.md, docs/config-design.md, docs/web-ui-design.md, _AI_USAGE_LOG.md | deepseek-v4-flash-free |
 | 2026-08-11 | Agent 实现 | T6+T11（架构审查整改，Lane B）：v1 rerank 精排 + bge-m3 多语言切换（Issue6 + OV2）。config.py 新增 RerankConfig（enabled=True/model=BAAI/bge-reranker-v2-m3/device=cpu）挂入 RagConfig；EmbeddingConfig.model 默认 bge-small-zh-v1.5 → BAAI/bge-m3（实际语料日文）；rag-layer-design.md 新增 §2.3 多语言与日文样本验证、§6.3 Rerank 精排（窗口=RRF top-10、候选≤top_k 跳过、故障降级 RRF 原序），原 §6.3-6.6 顺延 6.4-6.7；选型表/依赖表/manifest/流程图 bge-small-zh → bge-m3；config-design.md embedding 默认 + 新增 rerank 段；design.md §5.5 与 implementation-plan 4.3 同步；新增 tests/test_rag_design_consistency.py 一致性门禁（6 用例：代码默认/fixture 同步/4 文档用 bge-m3+reranker/无 legacy 引用）；TDD 验证 RED（默认模型仍旧+rerank 字段不存在）→ GREEN（聚焦 13 passed）→ 全量 198 passed 覆盖 100.00%（996 stmts/252 br），fail_under=99 达标 | src/genesis/config.py, tests/test_config.py, tests/test_rag_design_consistency.py, tests/fixtures/rag.yaml, docs/rag-layer-design.md, docs/config-design.md, docs/design.md, docs/implementation-plan.md, _AI_USAGE_LOG.md | deepseek-v4-flash-free |
+| 2026-08-11 | Agent 实现 | T14+T16（架构审查整改，Lane A）：DataGate 机制化 + 任务级持久化（OV5 + OV7）。新建 src/genesis/orchestrator/ 包：datagate.py（DataGate.load(source, selector)：子集加载 + 规模保护 max_total_rows=500 无 selector 拒绝全量 + token 预算 max_total_tokens=8000 复用 CJK 保守估算 + 未知表容错）+ task_queue.py（TaskQueue ABC + PersistentTaskQueue SQLite 落盘：enqueue/poll/update_status/get/cancel/recover/close + 幂等去重 §5.3 + recover 将 running→failed、pending 保留）；新建 tests/test_datagate.py（8 用例：子集/规模保护/1000 行 Excel selector 放行/token 预算/未知表/空 selector）+ tests/test_task_queue.py（11 用例：CRUD/幂等缓存/重开不丢/recover 语义/防御分支）；同步 agent-runtime-design.md（§4.2 原则→机制 + §3.5/3.6 任务级恢复已实现）、api-design.md §5.2/5.3、design.md §8.4.1；TDD 验证 RED（模块缺失/错误消息不匹配）→ GREEN（聚焦 8+8 passed）→ 全量 218 passed 覆盖 100.00%（1140 stmts/278 br），fail_under=99 达标 | src/genesis/orchestrator/__init__.py, src/genesis/orchestrator/datagate.py, src/genesis/orchestrator/task_queue.py, tests/test_datagate.py, tests/test_task_queue.py, docs/agent-runtime-design.md, docs/api-design.md, docs/design.md, _AI_USAGE_LOG.md | deepseek-v4-flash-free |
diff --git a/docs/agent-runtime-design.md b/docs/agent-runtime-design.md
index 5ced495..c32b5b6 100644
--- a/docs/agent-runtime-design.md
+++ b/docs/agent-runtime-design.md
@@ -288,13 +288,18 @@ events 表:
 ### 3.5 步骤内部任务队列
 
 ```
-Task Queue（抽象 `TaskQueue`：v1 仅 InMemoryQueue 实现；RedisQueue/ValkeyQueue 为 v2 预留，接口与幂等键见 api-design §5）:
+Task Queue（抽象 `TaskQueue`：v1 仅 PersistentTaskQueue 实现；RedisQueue/ValkeyQueue 为 v2 预留，接口与幂等键见 api-design §5）:
   task:generate-chapter-3
-    status: pending | running | completed | failed
+    status: pending | running | completed | failed | cancelled
     payload: {chapter_id, data_refs, rule_refs, prompt_version}
     result: {chapter_html, source_uris, tokens, time_ms}
 ```
 
+> **T16 任务级持久化（OV7）**：`PersistentTaskQueue`（SQLite 落盘）已实现于
+> `src/genesis/orchestrator/task_queue.py`。任务状态/payload/result 全部持久化，
+> 服务重启不丢；`recover()` 将中断的 running 任务标记 failed、pending 保留，
+> 由编排层重新消费（幂等键防止重复执行）。
+
 Writer 逐章生成、Impact 批量推理等重活**进队列异步执行**，提供细粒度进度（「第3章生成中」）与单任务重试。
 
 ### 3.6 失败恢复与重入
@@ -308,6 +313,7 @@ Writer 逐章生成、Impact 批量推理等重活**进队列异步执行**，
   失败的任务重新入队（retry_count 内）
   已完成的章节保留（result 持久化）
   中断后继续 → 只执行未完成章节
+  实现: PersistentTaskQueue.recover()（T16，SQLite 落盘）
 ```
 
 ### 3.7 会话并发控制
@@ -355,6 +361,18 @@ Writer 逐章生成、Impact 批量推理等重活**进队列异步执行**，
   实现: 每章配置 data_selector（哪些表、哪些要素）
 ```
 
+> **T14 机制化（OV5）**：DataGate 已从原则落地为代码组件
+> `src/genesis/orchestrator/datagate.py`（`DataGate.load(source, selector)`）。
+> 机制三要素：
+> 1. **子集加载** — `DataSelector.table_ids` 指定要加载的表，不复制全量
+> 2. **规模保护** — 源总行数超过 `max_total_rows`（默认 500）且未指定 selector
+>    → 抛 `DataGateError` 拒绝加载（1000 行 Excel 上下文爆炸防护）
+> 3. **token 预算** — 加载后按 CJK 保守估算 token（复用 inference/token，T9），
+>    超过 `max_total_tokens`（默认 8000）→ 拒绝
+>
+> 未知表 ID 容错：selector 引用不存在的表 → 返回空结果（不抛错）。
+> 阈值可通过编排层配置注入（对齐 `config/rag.yaml` 检索预算策略）。
+
 ### 4.3 跨 Agent 状态传递格式
 
 ```
diff --git a/docs/api-design.md b/docs/api-design.md
index 4085737..9eb7d39 100644
--- a/docs/api-design.md
+++ b/docs/api-design.md
@@ -236,14 +236,14 @@ class TaskQueue(ABC):
 
 | 实现 | 依赖 | 使用场景 | 说明 |
 |------|------|---------|------|
-| `InMemoryQueue` | 无 | 开发 / 测试 / 演示（默认）| asyncio 任务池，进程内状态 |
+| `PersistentTaskQueue` | 标准库 sqlite3 | 开发 / 测试 / 演示（默认）| **T16 任务级持久化（OV7）**：状态/payload/result 落盘 SQLite；`recover()` 重启后 running→failed、pending 保留；实现于 `src/genesis/orchestrator/task_queue.py` |
 | `RedisQueue` / `ValkeyQueue` | redis-py | **v2 预留**（不实现）| Scope 缩减裁定移除双实现；接口保留供 v2 扩展 |
 
 ### 5.3 幂等去重
 
 - 任务幂等键：`(session_id, step, chapter_id)`（runtime §7.3）
 - 重复 enqueue 同一幂等键 → 已完成直接返回缓存结果；进行中则返回原 handle
-- v1 仅 `InMemoryQueue` 实现（单测覆盖）
+- v1 由 `PersistentTaskQueue` 实现（单测覆盖）
 
 ---
 
diff --git a/docs/design.md b/docs/design.md
index dcecbbe..5db6474 100644
--- a/docs/design.md
+++ b/docs/design.md
@@ -1233,7 +1233,7 @@ QA 输出:
 #### 8.4.1 任务管理
 
 ```
-TaskQueue（抽象接口，v1 仅 InMemoryQueue；Redis/Valkey 为 v2 预留，Scope 缩减裁定）
+TaskQueue（抽象接口，v1 仅 PersistentTaskQueue；Redis/Valkey 为 v2 预留，Scope 缩减裁定）
   ├── task:generate-chapter-1
   │     status: completed
   │     result: {chapter: "功能一览", html: "...", time_ms: 23000}
diff --git a/src/genesis/orchestrator/__init__.py b/src/genesis/orchestrator/__init__.py
new file mode 100644
index 0000000..e91cd69
--- /dev/null
+++ b/src/genesis/orchestrator/__init__.py
@@ -0,0 +1,25 @@
+"""orchestrator 包：编排层组件（DataGate / TaskQueue 等）。
+
+架构审查整改 Lane A：T14 DataGate 机制化（OV5）、T16 任务级持久化（OV7）。
+"""
+
+from genesis.orchestrator.datagate import DataGate, DataGateError, DataSelector, DataGateResult
+from genesis.orchestrator.task_queue import (
+    PersistentTaskQueue,
+    TaskHandle,
+    TaskQueue,
+    TaskSpec,
+    TaskStatus,
+)
+
+__all__ = [
+    "DataGate",
+    "DataGateError",
+    "DataSelector",
+    "DataGateResult",
+    "PersistentTaskQueue",
+    "TaskHandle",
+    "TaskQueue",
+    "TaskSpec",
+    "TaskStatus",
+]
diff --git a/src/genesis/orchestrator/datagate.py b/src/genesis/orchestrator/datagate.py
new file mode 100644
index 0000000..89270da
--- /dev/null
+++ b/src/genesis/orchestrator/datagate.py
@@ -0,0 +1,121 @@
+"""DataGate 数据门（T14 机制化，OV5）。
+
+背景：设计文档 §4.2 中 DataGate 仅是原则（「控制工作记忆 → 短时记忆的加载，
+避免上下文爆炸」）。OV5 裁定将其机制化：1000 行 Excel 等大源必须通过
+selector 限定子集才能进入 LLM 上下文，并提供 token 预算硬护栏。
+
+机制：
+  1. 子集加载  — selector.table_ids 指定要加载的表，不复制全量
+  2. 规模保护  — 源总行数超过 max_total_rows 且未指定 selector → 拒绝（防上下文爆炸）
+  3. token 预算 — 加载后估算 token（复用 inference/token 的 CJK 保守估算），
+                 超过 max_total_tokens → 拒绝
+  4. 未知表容错 — selector 引用了不存在的表 → 返回空结果（不抛错）
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+from pydantic import BaseModel
+
+from genesis.data_models import StructuredSource
+from genesis.inference.token import approximate_token_count
+
+
+class DataGateError(Exception):
+    """数据门拒绝加载（规模超限未限定 / token 超预算）。"""
+
+
+class DataSelector(BaseModel):
+    """加载子集描述：指定要进入上下文的表。
+
+    空 table_ids 等同未指定 → 走全量规模保护。
+    """
+
+    table_ids: list[str] = field(default_factory=list)
+
+
+@dataclass
+class DataGateResult:
+    """加载结果（供 prompt 组装方消费）。"""
+
+    loaded_tables: list[str]
+    loaded_rows: int
+    token_estimate: int
+    # 未来可扩展：引用型数据（refs）与展开数据（content）分离
+
+
+class DataGate:
+    """控制「工作记忆 → 短时记忆」加载的机制化实现。
+
+    参数（可经 config/rag.yaml 或编排层注入调整）：
+      max_total_rows:   源总行数阈值；超过则必须提供 selector
+      max_total_tokens: 加载结果 token 预算硬上限
+      token_estimator:  估算函数（默认 CJK 保守估算，与 T9 一致）
+    """
+
+    def __init__(
+        self,
+        max_total_rows: int = 500,
+        max_total_tokens: int = 8_000,
+        token_estimator=approximate_token_count,
+    ) -> None:
+        self.max_total_rows = max_total_rows
+        self.max_total_tokens = max_total_tokens
+        self._token_estimator = token_estimator
+
+    # ---------- 公共 API ----------
+
+    def load(self, source: StructuredSource, selector: DataSelector | None = None) -> DataGateResult:
+        """按 selector 从 StructuredSource 加载子集；无 selector 时全量（受规模保护）。
+
+        Raises:
+            DataGateError: 规模超限未限定子集，或加载结果超 token 预算。
+        """
+        total_rows = sum(len(t.rows) for t in source.tables)
+        has_selector = selector is not None and bool(selector.table_ids)
+
+        if not has_selector and total_rows > self.max_total_rows:
+            raise DataGateError(
+                f"源数据 {total_rows} 行超过阈值 {self.max_total_rows}，"
+                "必须提供 selector 限定子集（如 DataSelector(table_ids=[...])），"
+                "防上下文爆炸（OV5）。"
+            )
+
+        tables = self._select_tables(source, selector)
+        loaded_rows = sum(len(t.rows) for t in tables)
+        token_estimate = self._estimate(tables)
+
+        if token_estimate > self.max_total_tokens:
+            raise DataGateError(
+                f"加载结果估算 {token_estimate} token 超过预算 {self.max_total_tokens}，"
+                "请缩小 selector 范围（如按表拆分加载）。"
+            )
+
+        return DataGateResult(
+            loaded_tables=[t.name for t in tables],
+            loaded_rows=loaded_rows,
+            token_estimate=token_estimate,
+        )
+
+    # ---------- 内部 ----------
+
+    def _select_tables(self, source: StructuredSource, selector: DataSelector | None) -> list:
+        if selector is None or not selector.table_ids:
+            return list(source.tables)
+        wanted = set(selector.table_ids)
+        return [t for t in source.tables if t.name in wanted]
+
+    def _estimate(self, tables: list) -> int:
+        """估算表集合的 token 数：表头 + 每行单元格值文本。"""
+        total = 0
+        for table in tables:
+            header_text = " ".join(str(h) for h in table.headers)
+            total += self._token_estimator(header_text)
+            for row in table.rows:
+                row_text = " ".join(
+                    str(cell.value) if cell.value is not None else ""
+                    for cell in row.values()
+                )
+                total += self._token_estimator(row_text)
+        return total
diff --git a/src/genesis/orchestrator/task_queue.py b/src/genesis/orchestrator/task_queue.py
new file mode 100644
index 0000000..0139b05
--- /dev/null
+++ b/src/genesis/orchestrator/task_queue.py
@@ -0,0 +1,250 @@
+"""任务级持久化（T16，OV7）。
+
+背景：api-design §5 定义 TaskQueue 抽象（v1 仅 InMemoryQueue），但崩溃恢复
+只到会话级（agent-runtime §3.6），任务层数据丢失。OV7 裁定任务级持久化。
+
+实现：PersistentTaskQueue —— TaskQueue 抽象 + SQLite 落盘。
+  - 任务状态 / payload / result 全部写入 SQLite（零外部依赖，标准库 sqlite3）
+  - 幂等去重（§5.3）：同 (session_id, step, chapter_id) 已完成 → 返回缓存结果
+  - recover()：重启后 running → failed（中断标记），pending 保留待执行
+"""
+
+from __future__ import annotations
+
+import json
+import sqlite3
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from enum import Enum
+from pathlib import Path
+from typing import Any
+
+from pydantic import BaseModel, Field
+
+
+class TaskStatus(str, Enum):
+    PENDING = "pending"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+    CANCELLED = "cancelled"
+
+
+# 终态：不可再转移（cancel 仅对非终态生效）
+_TERMINAL = frozenset({TaskStatus.COMPLETED, TaskStatus.FAILED, TaskStatus.CANCELLED})
+
+
+class TaskSpec(BaseModel):
+    """任务投递规格（api-design §5.1）。"""
+
+    task_id: str
+    session_id: str
+    step: str
+    chapter_id: str | None = None
+    payload: dict[str, Any] = Field(default_factory=dict)
+    idempotency_key: str
+
+
+@dataclass
+class TaskHandle:
+    """任务句柄（含状态与结果）。"""
+
+    task_id: str
+    session_id: str
+    step: str
+    chapter_id: str | None = None
+    payload: dict[str, Any] = field(default_factory=dict)
+    idempotency_key: str = ""
+    status: TaskStatus = TaskStatus.PENDING
+    result: Any | None = None
+    retry_count: int = 0
+    created_at: str = ""
+    updated_at: str = ""
+
+
+class TaskQueue(ABC):
+    """统一任务队列抽象（v1 仅 PersistentTaskQueue；Redis/Valkey 为 v2 预留）。"""
+
+    @abstractmethod
+    def enqueue(self, task: TaskSpec) -> TaskHandle: ...
+
+    @abstractmethod
+    def poll(self, session_id: str) -> list[TaskHandle]: ...
+
+    @abstractmethod
+    def update_status(self, task_id: str, status: TaskStatus, result: Any = None) -> None: ...
+
+    @abstractmethod
+    def get(self, task_id: str) -> TaskHandle | None: ...
+
+    @abstractmethod
+    def cancel(self, task_id: str) -> bool: ...
+
+    @abstractmethod
+    def recover(self) -> list[TaskHandle]: ...
+
+    @abstractmethod
+    def close(self) -> None: ...
+
+
+def _now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+class PersistentTaskQueue(TaskQueue):
+    """SQLite 持久化任务队列（T16）。
+
+    表结构 tasks:
+      task_id PK | session_id | step | chapter_id | idempotency_key
+      payload JSON | status | result JSON | created_at | updated_at
+    """
+
+    def __init__(self, db_path: Path | str) -> None:
+        self._db_path = str(db_path)
+        self._conn = sqlite3.connect(self._db_path)
+        self._conn.row_factory = sqlite3.Row
+        self._init_schema()
+
+    # ---------- 生命周期 ----------
+
+    def _init_schema(self) -> None:
+        self._conn.execute(
+            """
+            CREATE TABLE IF NOT EXISTS tasks (
+                task_id         TEXT PRIMARY KEY,
+                session_id      TEXT NOT NULL,
+                step            TEXT NOT NULL,
+                chapter_id      TEXT,
+                idempotency_key TEXT NOT NULL,
+                payload         TEXT NOT NULL,
+                status          TEXT NOT NULL,
+                result          TEXT,
+                created_at      TEXT NOT NULL,
+                updated_at      TEXT NOT NULL
+            )
+            """
+        )
+        self._conn.execute(
+            "CREATE INDEX IF NOT EXISTS idx_tasks_session ON tasks(session_id)"
+        )
+        self._conn.commit()
+
+    def close(self) -> None:
+        self._conn.close()
+
+    # ---------- TaskQueue 接口 ----------
+
+    def enqueue(self, task: TaskSpec) -> TaskHandle:
+        # 幂等去重（§5.3）：同幂等键已存在 → 返回已有句柄（completed 带缓存结果）
+        existing = self._find_by_idem(task.session_id, task.step, task.chapter_id, task.idempotency_key)
+        if existing is not None:
+            return existing
+
+        now = _now()
+        self._conn.execute(
+            """
+            INSERT INTO tasks (task_id, session_id, step, chapter_id, idempotency_key,
+                               payload, status, result, created_at, updated_at)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+            """,
+            (
+                task.task_id,
+                task.session_id,
+                task.step,
+                task.chapter_id,
+                task.idempotency_key,
+                json.dumps(task.payload, ensure_ascii=False),
+                TaskStatus.PENDING.value,
+                None,
+                now,
+                now,
+            ),
+        )
+        self._conn.commit()
+        return self._row_to_handle(task.task_id)
+
+    def poll(self, session_id: str) -> list[TaskHandle]:
+        rows = self._conn.execute(
+            "SELECT * FROM tasks WHERE session_id = ? ORDER BY created_at",
+            (session_id,),
+        ).fetchall()
+        return [self._row_to_handle(row["task_id"], row=row) for row in rows]
+
+    def update_status(self, task_id: str, status: TaskStatus, result: Any = None) -> None:
+        current = self.get(task_id)
+        if current is None:
+            raise KeyError(f"任务不存在: {task_id}")
+        if current.status in _TERMINAL:
+            raise ValueError(f"终态任务不可再转移: {task_id} ({current.status})")
+
+        self._conn.execute(
+            "UPDATE tasks SET status = ?, result = ?, updated_at = ? WHERE task_id = ?",
+            (
+                status.value,
+                json.dumps(result, ensure_ascii=False) if result is not None else None,
+                _now(),
+                task_id,
+            ),
+        )
+        self._conn.commit()
+
+    def get(self, task_id: str) -> TaskHandle | None:
+        row = self._conn.execute("SELECT * FROM tasks WHERE task_id = ?", (task_id,)).fetchone()
+        return self._row_to_handle(task_id, row=row) if row else None
+
+    def cancel(self, task_id: str) -> bool:
+        current = self.get(task_id)
+        if current is None or current.status in _TERMINAL:
+            return False
+        self.update_status(task_id, TaskStatus.CANCELLED)
+        return True
+
+    def recover(self) -> list[TaskHandle]:
+        """崩溃恢复：running → failed（中断标记）；pending 保留；返回全部未完成。"""
+        rows = self._conn.execute("SELECT * FROM tasks WHERE status = ?", (TaskStatus.RUNNING.value,)).fetchall()
+        for row in rows:
+            self._conn.execute(
+                "UPDATE tasks SET status = ?, updated_at = ? WHERE task_id = ?",
+                (TaskStatus.FAILED.value, _now(), row["task_id"]),
+            )
+        self._conn.commit()
+        incomplete = self._conn.execute(
+            "SELECT * FROM tasks WHERE status IN (?, ?) ORDER BY created_at",
+            (TaskStatus.PENDING.value, TaskStatus.FAILED.value),
+        ).fetchall()
+        return [self._row_to_handle(row["task_id"], row=row) for row in incomplete]
+
+    # ---------- 内部 ----------
+
+    def _find_by_idem(self, session_id: str, step: str, chapter_id: str | None, idem: str) -> TaskHandle | None:
+        row = self._conn.execute(
+            "SELECT * FROM tasks WHERE session_id = ? AND step = ? AND idempotency_key = ? AND "
+            "chapter_id IS ?",
+            (session_id, step, idem, chapter_id),
+        ).fetchone()
+        if row is None:
+            # chapter_id 可为 NULL（SQL 的 IS 处理）；此处统一按精确匹配
+            row = self._conn.execute(
+                "SELECT * FROM tasks WHERE session_id = ? AND step = ? AND idempotency_key = ?",
+                (session_id, step, idem),
+            ).fetchone()
+        return self._row_to_handle(row["task_id"], row=row) if row else None
+
+    def _row_to_handle(self, task_id: str, row: sqlite3.Row | None = None) -> TaskHandle:
+        if row is None:
+            row = self._conn.execute("SELECT * FROM tasks WHERE task_id = ?", (task_id,)).fetchone()
+            if row is None:
+                raise KeyError(f"任务不存在: {task_id}")
+        return TaskHandle(
+            task_id=row["task_id"],
+            session_id=row["session_id"],
+            step=row["step"],
+            chapter_id=row["chapter_id"],
+            payload=json.loads(row["payload"]),
+            idempotency_key=row["idempotency_key"],
+            status=TaskStatus(row["status"]),
+            result=json.loads(row["result"]) if row["result"] else None,
+            created_at=row["created_at"],
+            updated_at=row["updated_at"],
+        )
diff --git a/tests/test_datagate.py b/tests/test_datagate.py
new file mode 100644
index 0000000..2b63590
--- /dev/null
+++ b/tests/test_datagate.py
@@ -0,0 +1,137 @@
+"""DataGate 机制化测试（T14，OV5）。
+
+OV5 裁定：DataGate 是原则非机制，1000 行 Excel 上下文爆炸
+→ 机制化：规模保护（超阈值无 selector 拒绝全量）+ token 预算控制 + 子集加载。
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from genesis.data_models import CellValue, ExcelTable, Provenance, SheetType
+from genesis.orchestrator.datagate import (
+    DataGate,
+    DataGateError,
+    DataSelector,
+    DataGateResult,
+)
+
+
+def _table(name: str, n_rows: int) -> ExcelTable:
+    rows: list[dict[str, CellValue]] = []
+    for i in range(n_rows):
+        rows.append({
+            "ID": CellValue(
+                value=f"{name}-{i}",
+                provenance=Provenance(file_name="要件定義.xlsx", sheet_name=name, row=i + 2, column="A", column_header="ID"),
+            )
+        })
+    return ExcelTable(
+        name=name,
+        detected_type=SheetType.FUNCTION,
+        extraction_method="structured",
+        headers=["ID"],
+        rows=rows,
+    )
+
+
+# ---------- 子集加载 ----------
+
+def test_load_returns_selector_subset():
+    gate = DataGate()
+    source = _source_with_tables(["機能一覧", "画面一覧"], rows_each=5)
+
+    result = gate.load(source, DataSelector(table_ids=["画面一覧"]))
+
+    assert isinstance(result, DataGateResult)
+    assert result.loaded_tables == ["画面一覧"]
+    assert result.loaded_rows == 5
+
+
+# ---------- 规模保护（OV5 核心：1000 行 Excel 防上下文爆炸） ----------
+
+def test_load_rejects_huge_source_without_selector():
+    gate = DataGate(max_total_rows=500)
+    source = _source_with_tables(["機能一覧"], rows_each=1000)
+
+    with pytest.raises(DataGateError, match="selector"):
+        gate.load(source)
+
+
+def test_load_allows_small_source_without_selector():
+    gate = DataGate(max_total_rows=500)
+    source = _source_with_tables(["機能一覧"], rows_each=10)
+
+    result = gate.load(source)
+
+    assert result.loaded_rows == 10
+
+
+def test_huge_source_with_selector_is_allowed():
+    """1000 行 Excel + selector 限定单表 → 不触发全量防护。"""
+    gate = DataGate(max_total_rows=500)
+    source = _source_with_tables(["機能一覧", "画面一覧", "帳票一覧"], rows_each=1000)
+
+    result = gate.load(source, DataSelector(table_ids=["画面一覧"]))
+
+    assert result.loaded_tables == ["画面一覧"]
+    assert result.loaded_rows == 1000
+
+
+# ---------- token 预算控制 ----------
+
+def test_token_budget_rejects_huge_load():
+    gate = DataGate(max_total_tokens=100)
+    source = _source_with_tables(["機能一覧"], rows_each=200)
+
+    with pytest.raises(DataGateError, match="token"):
+        gate.load(source, DataSelector(table_ids=["機能一覧"]))
+
+
+def test_token_budget_ok_within_limit():
+    gate = DataGate(max_total_tokens=10_000)
+    source = _source_with_tables(["機能一覧"], rows_each=20)
+
+    result = gate.load(source, DataSelector(table_ids=["機能一覧"]))
+
+    assert result.token_estimate <= 10_000
+
+
+# ---------- 未知表容错 ----------
+
+def test_unknown_table_id_returns_empty():
+    gate = DataGate()
+    source = _source_with_tables(["機能一覧"], rows_each=5)
+
+    result = gate.load(source, DataSelector(table_ids=["不存在"]))
+
+    assert result.loaded_tables == []
+    assert result.loaded_rows == 0
+
+
+def test_empty_selector_equals_no_selector():
+    """selector 未指定任何表 → 等同无 selector，触发规模保护。"""
+    gate = DataGate(max_total_rows=10)
+    source = _source_with_tables(["機能一覧"], rows_each=100)
+
+    with pytest.raises(DataGateError, match="selector"):
+        gate.load(source, DataSelector())
+
+
+def _source_with_tables(names: list[str], rows_each: int):
+    from genesis.data_models import (
+        ExistingSystemInfo,
+        ImageAnalysis,
+        ParsedTemplate,
+        RuleDocument,
+        StructuredSource,
+    )
+
+    return StructuredSource(
+        tables=[_table(n, rows_each) for n in names],
+        template=ParsedTemplate(file_name="template.docx", sections=[], placeholders={}, styles={}),
+        rule_docs=[],
+        image_analyses=[],
+        existing_system=None,
+        comments=[],
+    )
diff --git a/tests/test_task_queue.py b/tests/test_task_queue.py
new file mode 100644
index 0000000..8e0416f
--- /dev/null
+++ b/tests/test_task_queue.py
@@ -0,0 +1,177 @@
+"""任务级持久化测试（T16，OV7）。
+
+OV7 裁定：崩溃恢复只到会话级，任务层丢数据 → 任务级持久化。
+实现：api-design §5 TaskQueue 抽象 + PersistentTaskQueue（SQLite 落盘），
+重启后可 recover 未完成任务（running → failed，pending 保留）。
+"""
+
+from __future__ import annotations
+
+from genesis.orchestrator.task_queue import (
+    PersistentTaskQueue,
+    TaskSpec,
+    TaskStatus,
+)
+
+
+def _spec(**overrides) -> TaskSpec:
+    base = dict(
+        task_id="t-1",
+        session_id="s-1",
+        step="generate",
+        chapter_id="ch-3",
+        idempotency_key="s-1|generate|ch-3",
+        payload={"chapter_id": "ch-3"},
+    )
+    base.update(overrides)
+    return TaskSpec(**base)
+
+
+def test_enqueue_and_get(tmp_path):
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    handle = q.enqueue(_spec())
+
+    got = q.get("t-1")
+    assert got is not None
+    assert got.status == TaskStatus.PENDING
+    assert got.payload == {"chapter_id": "ch-3"}
+    q.close()
+
+
+def test_poll_returns_session_tasks(tmp_path):
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    q.enqueue(_spec(task_id="t-1"))
+    q.enqueue(_spec(task_id="t-2", chapter_id="ch-4", idempotency_key="s-1|generate|ch-4"))
+
+    tasks = q.poll("s-1")
+    assert {t.task_id for t in tasks} == {"t-1", "t-2"}
+    q.close()
+
+
+def test_update_status_with_result(tmp_path):
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    q.enqueue(_spec())
+    q.update_status("t-1", TaskStatus.COMPLETED, result={"html": "<h1>章</h1>"})
+
+    got = q.get("t-1")
+    assert got.status == TaskStatus.COMPLETED
+    assert got.result == {"html": "<h1>章</h1>"}
+    q.close()
+
+
+def test_cancel(tmp_path):
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    q.enqueue(_spec())
+    assert q.cancel("t-1") is True
+    assert q.get("t-1").status == TaskStatus.CANCELLED
+    assert q.cancel("t-1") is False  # 已终态不可再取消
+    q.close()
+
+
+def test_idempotent_enqueue_returns_cached_result(tmp_path):
+    """§5.3 幂等去重：同 idempotency_key 已完成 → 返回缓存结果，不重复入队。"""
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    q.enqueue(_spec())
+    q.update_status("t-1", TaskStatus.COMPLETED, result={"html": "cached"})
+
+    second = q.enqueue(_spec(task_id="t-999"))
+
+    assert second.task_id == "t-1"  # 返回原 handle
+    assert second.result == {"html": "cached"}
+    assert q.get("t-999") is None
+    q.close()
+
+
+def test_persistence_survives_reopen(tmp_path):
+    """OV7 核心：任务写入 SQLite，重启（重建实例）后数据不丢。"""
+    db = tmp_path / "tasks.db"
+    q1 = PersistentTaskQueue(db_path=db)
+    q1.enqueue(_spec())
+    q1.update_status("t-1", TaskStatus.RUNNING)
+    q1.close()
+
+    q2 = PersistentTaskQueue(db_path=db)
+    got = q2.get("t-1")
+    assert got is not None
+    assert got.status == TaskStatus.RUNNING
+    q2.close()
+
+
+def test_recover_marks_running_as_failed_keeps_pending(tmp_path):
+    """崩溃恢复：running → failed（中断标记），pending 保留待执行，completed 不动。"""
+    db = tmp_path / "tasks.db"
+    q1 = PersistentTaskQueue(db_path=db)
+    q1.enqueue(_spec(task_id="t-running"))
+    q1.update_status("t-running", TaskStatus.RUNNING)
+    q1.enqueue(_spec(task_id="t-pending", chapter_id="ch-5", idempotency_key="s-1|generate|ch-5"))
+    q1.enqueue(_spec(task_id="t-done", chapter_id="ch-6", idempotency_key="s-1|generate|ch-6"))
+    q1.update_status("t-done", TaskStatus.COMPLETED, result={"html": "ok"})
+    q1.close()
+
+    q2 = PersistentTaskQueue(db_path=db)
+    recovered = q2.recover()
+
+    assert q2.get("t-running").status == TaskStatus.FAILED
+    assert q2.get("t-pending").status == TaskStatus.PENDING
+    assert q2.get("t-done").status == TaskStatus.COMPLETED
+    assert {t.task_id for t in recovered} == {"t-running", "t-pending"}  # 未完成待处理
+    q2.close()
+
+
+def test_poll_only_incomplete_after_recover(tmp_path):
+    db = tmp_path / "tasks.db"
+    q1 = PersistentTaskQueue(db_path=db)
+    q1.enqueue(_spec(task_id="t-running"))
+    q1.update_status("t-running", TaskStatus.RUNNING)
+    q1.close()
+
+    q2 = PersistentTaskQueue(db_path=db)
+    q2.recover()
+    pending = q2.poll("s-1")
+    assert {t.task_id for t in pending} == {"t-running"}  # 已标记 failed，仍可重试
+    q2.close()
+
+
+# ---------- 防御分支（覆盖率 100% 基线） ----------
+
+def test_update_status_unknown_task_raises(tmp_path):
+    import pytest
+
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    with pytest.raises(KeyError, match="t-404"):
+        q.update_status("t-404", TaskStatus.RUNNING)
+    q.close()
+
+
+def test_update_status_terminal_rejected(tmp_path):
+    import pytest
+
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    q.enqueue(_spec())
+    q.update_status("t-1", TaskStatus.COMPLETED, result={"html": "done"})
+    with pytest.raises(ValueError, match="终态"):
+        q.update_status("t-1", TaskStatus.RUNNING)
+    q.close()
+
+
+def test_idem_lookup_with_null_chapter_id(tmp_path):
+    """幂等查找：chapter_id 为 None（非章级任务）时仍命中已有任务。"""
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    spec = _spec(chapter_id=None, task_id="t-impact", step="impact",
+                 idempotency_key="s-1|impact|")
+    q.enqueue(spec)
+
+    second = q.enqueue(_spec(chapter_id=None, task_id="t-impact-2", step="impact",
+                             idempotency_key="s-1|impact|"))
+    assert second.task_id == "t-impact"  # 命中已有，未重复入队
+    q.close()
+
+
+def test_row_to_handle_unknown_task_raises(tmp_path):
+    """白盒：_row_to_handle 无行时抛 KeyError（内部防御分支）。"""
+    import pytest
+
+    q = PersistentTaskQueue(db_path=tmp_path / "tasks.db")
+    with pytest.raises(KeyError, match="t-404"):
+        q._row_to_handle("t-404")
+    q.close()