Merge pull request 'fix(scheduler): no false skip for full snapshot; quiet window + retry' (#7) from auto/sched-no-false-skip-4821 into main

Reviewed-on: #7
Reviewed-by: kairee <1+kairee@users.noreply.github.com>

DEVELOPMENT.md

@@ -456,6 +456,9 @@ P1 实现说明：
 
 - 时区使用固定 UTC+8（中国无夏令时），避免 Windows 缺少 `tzdata`；
 - 午夜 `seed_at`（默认 `00:00`）把 `snapshot:last_completed` 的 ok=1 行克隆为当日壳，写入 `snapshot:seeded:<date>=1` / `fans:seeded:<date>=1`，**不**推进完成日；
+- 也可手动：`python app.py snapshot-seed [--date YYYY-MM-DD]`（Debian / `docker compose exec re-xxm python app.py snapshot-seed`）；
+- 目标日已有 `snapshot:completed:<date>` 时 seed 记 `already_completed` 成功（即使 `last_completed` 已等于当日），**不得**误判 `no_source`；
+- 若全站 `snapshot` 正在跑，或目标日未完成且 `last_completed` 暂不可用（`no_source`），seed **不记成功**，约 10 分钟后重试直到真正克隆成功；
 - 晚间 `snapshot_at`（默认 `23:15`）对 seeded 日 force 重采；成功后写 `snapshot:completed:<date>=1`、推进 `last_completed` 并清除 seeded；
 - 当日全部视频与粉丝采集成功后在 `kv` 写入 `snapshot:completed:<date>=1`；启动时未完成会补跑（纯 seed 日不误标完成）；
 - 完成标记不是唯一门槛：每次检查 `pending_snapshot_aids(date)`，新入库或从 `dropped` 恢复的视频会立即补采当天快照；
@@ -466,6 +469,15 @@ P1 实现说明：
 - Cookie 登录态通过 `x/web-interface/nav` 检查，失效只告警，不阻塞匿名 `view`；
 - Telegram 摘要推迟到 P3，P1 先写结构化 JSON 日志和 `kv` 状态。
 
+调度避让与补跑：
+
+- 常驻启动时日更任务（`snapshot_seed` / `snapshot` / `fans`）按墙钟下一档 `seed_at` / `snapshot_at` 排期（可参考上次成功/失败记录），**不得** `next_run=0` 重启即跑；
+- 仅到达当日 `snapshot_at` 后（或 CLI `--force`）才对 seed 壳 `force` 打穿；白天重启不因 `snapshot:seeded` 全量强采；
+- 全站 `snapshot` **不因**其它任务 running 而 skip；到期/强制必须执行；
+- 上海时区 **23:00–01:00** 为全站快照窗：`discovery` / `snapshot_recent` / `candidate_cleanup` 不执行，延后到 01:00 后；`fans` 短耗时，不避开任何任务；
+- 任务 skip（含静默窗延后）写入 `ok=false` + `skipped=true`，**不得**记成成功或推进「当日已完成」；
+- skip / fail 后短间隔补跑（约 5 分钟；`snapshot_seed` 避让/无源为约 10 分钟）；日更成功才推到次日 `snapshot_at` / `seed_at`。
+
 ---
 
 ## 8. Telegram 与入库操作
@@ -608,7 +620,7 @@ interaction_rate = (like + coin + favorite) / max(view, 1)
 
 - 查询前一有效快照而不是填补缺失日期；
 - 缺失日期在图表中断开；
-- 增速榜要求播放基数至少 200；
+- 增速榜要求播放基数至少 1000；
 - 互动率榜要求播放至少 100；
 - 样本不足两天的视频明确标注，不伪造增长数据。
 
@@ -756,7 +768,7 @@ uv run python tests/test_moves.py
 - 粉丝采集对每条 `relation/stat` 加 45s 硬超时看门狗（防 DNS/传输层挂起），每 25 条输出 `up_profile.progress` 进度；单条失败不阻塞整批；
 - `app.py backfill-followers`：一次性回填历史视频粉丝数，支持 `--limit` / `--dry-run`；
 - 新增内容效率/互动效率榜与 UP 内容效率（Gamma–Poisson EB 收缩），过滤 < 600 秒与未满 2 天的视频；
-- `analytics` 配置段：`min_duration_seconds`、`maturity_days`、`min_view_for_ranking`、`video_shrink_minutes`、`shrink_minutes`；
+- `analytics` 配置段：`min_duration_seconds`、`maturity_days`、`min_view_for_ranking`、`video_shrink_minutes`、`shrink_minutes`、`growth_rate_prior`；
 - 主导航拆分：`/videos`（视频）与 `/ups`（UP主）独立；旧 `/rank` 保留兼容；
 - 新增 `/about` 数据说明页与 `/api/about`；
 - 卡片精简：列表只保留主指标与互动率，明细移到详情页；

config.docker.example.jsonc

@@ -108,5 +108,6 @@
     "video_shrink_minutes": 60,
     "shrink_minutes": null,
     "home_growth_days": "auto",
+    "growth_rate_prior": 1000,
   },
 }

config.example.jsonc

@@ -123,5 +123,7 @@
     "shrink_minutes": null,
     // "auto" = 7 日有数据用 7，否则 1
     "home_growth_days": "auto",
+    // 增速榜排序：Δ / (基线 + prior)。卡片上的百分比仍是 Δ / 基线。
+    "growth_rate_prior": 1000,
   },
 }

project.md

@@ -380,7 +380,7 @@ for folder in 配置的年度夹（2025、2026、…）:
 - 两个夹共 831 条，一次性导入 `video` 表；
 - **历史快照无法回填**：B 站不提供任意日期历史数据，只能从部署当天开始积累；
 - **开发联调阶段产生的数据一律保留**（不清库、不重建），正式上线直接续用同一个 `app.db`；
-- 新发现/新入库的视频从"入库当天"开始积累快照，第一天没有增量是正常的，看板对"仅 1 个数据点"的视频标注"样本不足"。
+- 新发现/新入库的视频从"入库当天"开始积累快照。近 1 日：当日发布按 0 起算（一次成功采样即可）；更早发布的新入库按入库时播放量起算，第二次采样走晚间全量，不进小时快照。近 7/30 日仍要跨日快照。
 
 ---
 
@@ -473,6 +473,8 @@ CREATE TABLE snapshot (
   ok          INTEGER NOT NULL DEFAULT 1,
   err         TEXT,
   src         TEXT DEFAULT 'view',      -- view | infos（降级时标记）
+  view_open   INTEGER,                  -- 同日第一次成功采样的播放量
+  open_captured_at INTEGER,             -- 开盘采样时刻
   PRIMARY KEY (stat_date, aid)
 );
 CREATE INDEX idx_snap_aid ON snapshot(aid, stat_date);
@@ -672,23 +674,26 @@ w = {view:0.30, like:0.20, coin:0.20, favorite:0.15,
 
 ### 8.2 增长榜
 
-> 看板 UI 中“增长”标注为**播放增量**（区间内绝对增量），“增速”标注为**播放增速**（增量 / 基线播放）。两者基线都取不早于 `最新快照 - N 天` 的最近一条成功快照。
+> 看板 UI 中“增长”标注为**播放增量**（区间内绝对增量），“增速”标注为**播放增速**（增量 / 基线播放）。近 7/30 日基线取不早于 `最新快照 - N 天` 的最近一条成功快照。近 1 日：有更早日期快照则用前一日；否则当日发布基准为 0，更早发布的新入库用当天第一次成功采样（`view_open`）。
 
 ```
 d7 = (view_today - view_{today-7}) / 实际间隔天数
 r7 = (view_today - view_{today-7}) / max(view_{today-7}, 1)
+d1_published_today = view_today - 0          # 只进增量榜，不计算增速
+d1_imported_today  = view_today - view_open   # 需同日第二次采样
+sort_rate = Δ / (基线 + analytics.growth_rate_prior)  # 默认 prior=1000，只用于增速榜排序
 ```
 
 | 榜 | 口径 | 基数门槛 |
 |---|---|---|
 | 累计热度 | §8.1 | — |
 | 近 7 天播放增量 | `d7(view)` | — |
-| 近 7 天增速 | `r7(view)` | `view >= 200` |
+| 近 7 天增速 | `r7(view)` | 基线播放 `>= 1000` |
 | 互动率 | `(like+coin+fav)/view` | `view >= 100` |
 | 单项榜 | 7 项直接排序 | — |
 | UP 榜 | 按 `up_mid` 聚合 Σ/AVG | — |
 
-**冷启动**：不足 2 天数据的视频用"发布后 24h 增量"，并标注"样本不足"；缺失日不插值。
+**冷启动**：近 1 日不要求跨日快照。当日发布相对 0，只进播放增量、不进播放增速；更早发布且当天入库的片子相对 `view_open`（第二次采样走晚间全量，不进小时快照）。近 7/30 日仍要跨日快照。缺失日不插值。
 
 ### 8.3 趋势
 
@@ -795,7 +800,7 @@ secrets.example.json            凭据样例（auth 回写目标）
 **进程结构**
 
 ```
-scheduler 线程：discovery(每 60min) / snapshot_seed(每日 00:00) / snapshot+fans(每日 23:15) / snapshot_recent(小时) / candidate_cleanup(每日)
+scheduler 线程：discovery(每 discover_minutes，避开 23:00–01:00) / snapshot_seed(每日 00:00；snapshot 占用或 no_source 则 10 分钟重试，CLI: snapshot-seed) / snapshot(每日 23:15，不因其它任务 skip) / fans(每日 23:15，不避开) / snapshot_recent(小时，避开 23:00–01:00) / candidate_cleanup(每日，避开 23:00–01:00)
 web 线程：ThreadingHTTPServer :8080
 telegram：getUpdates long polling（回调 + 私聊文本入库）
 storage：SQLite WAL，单写线程

reaction_board/analytics.py

@@ -34,7 +34,7 @@ DEFAULT_HEAT_WEIGHTS: dict[str, float] = {
 }
 
 MIN_VIEW_FOR_INTERACTION = 100
-MIN_VIEW_FOR_GROWTH_RATE = 200
+MIN_VIEW_FOR_GROWTH_RATE = 1000
 MIN_EFFICIENCY_DURATION = 600
 DEFAULT_MATURITY_DAYS = 2
 DEFAULT_SHRINK_MINUTES = 120.0
@@ -313,15 +313,20 @@ def compute_growth(
     current: int | float | None,
     previous: int | float | None,
     gap_days: int | float | None,
+    *,
+    allow_zero_gap: bool = False,
 ) -> Growth:
     """计算累计差值、日均差值和相对增速。
 
-    缺失基线或间隔不足 1 天时返回不可用，不把 0 当作真实增长。
+    缺失基线或间隔为负时返回不可用。默认间隔不足 1 天也不可用，不把 0 当成真实增长。
+    ``allow_zero_gap=True`` 时允许同一天的两次采样（或相对 0）计算增量，``per_day`` 为 None。
     """
     if current is None or previous is None or gap_days is None:
         return Growth(available=False)
     gap = int(gap_days)
-    if gap <= 0:
+    if gap < 0:
+        return Growth(available=False)
+    if gap == 0 and not allow_zero_gap:
         return Growth(available=False)
     current_value = max(0, int(current))
     previous_value = max(0, int(previous))
@@ -329,12 +334,35 @@ def compute_growth(
     return Growth(
         available=True,
         delta=delta,
-        per_day=delta / gap,
+        per_day=None if gap == 0 else delta / gap,
         rate=delta / max(previous_value, 1),
         gap_days=gap,
     )
 
 
+def smoothed_growth_rate(
+    delta: int | float | None,
+    previous: int | float | None,
+    prior: int | float,
+) -> float | None:
+    """增速榜排序分：``Δ / (基线 + prior)``。
+
+    ``rate`` 仍是真实相对增速。``prior`` 只削弱小基数在排序里的优势。
+    """
+    if delta is None or previous is None or isinstance(delta, bool) or isinstance(previous, bool):
+        return None
+    if isinstance(prior, bool):
+        return None
+    try:
+        strength = float(prior)
+        base = max(0, int(previous)) + strength
+    except (TypeError, ValueError):
+        return None
+    if strength < 0 or base <= 0:
+        return None
+    return int(delta) / base
+
+
 def normalize_cover(url: object) -> str | None:
     """把 B 站封面统一为 https；空值返回 None。"""
     if not isinstance(url, str):

reaction_board/config.py

@@ -52,6 +52,7 @@ DEFAULT_ANALYTICS: dict[str, Any] = {
     "video_shrink_minutes": 60,
     "shrink_minutes": None,
     "home_growth_days": "auto",
+    "growth_rate_prior": 1000,
 }
 
 DEFAULT_CONFIG: dict[str, Any] = {
@@ -210,6 +211,7 @@ class AnalyticsConfig:
     video_shrink_minutes: int
     shrink_minutes: int | None
     home_growth_days: str | int
+    growth_rate_prior: int
 
 
 @dataclass(frozen=True)
@@ -592,6 +594,11 @@ def _build_config(data: Mapping[str, Any], path: Path, *, secrets_path: Path) ->
             )
         ),
         home_growth_days=home_growth_days,
+        growth_rate_prior=_as_int(
+            analytics_raw.get("growth_rate_prior", 1000),
+            "analytics.growth_rate_prior",
+            minimum=0,
+        ),
     )
 
     return AppConfig(

reaction_board/migrations.py

@@ -244,6 +244,18 @@ MIGRATIONS: tuple[Migration, ...] = (
             "CREATE INDEX idx_visit_date ON visit(stat_date)",
         ),
     ),
+    Migration(
+        version=8,
+        name="snapshot open view",
+        statements=(
+            # 同日第一次成功采样的播放量，供近 1 日「入库后增量」做基准。
+            # 已有成功行无法找回被覆盖的入库值，只能把当前 view 当作开盘。
+            'ALTER TABLE snapshot ADD COLUMN view_open INTEGER',
+            "ALTER TABLE snapshot ADD COLUMN open_captured_at INTEGER",
+            'UPDATE snapshot SET view_open = "view", open_captured_at = captured_at '
+            "WHERE ok = 1 AND view_open IS NULL",
+        ),
+    ),
 )
 
 

reaction_board/repository.py

@@ -690,12 +690,17 @@ class Repository:
         src: str = "view",
     ) -> None:
         values = metrics or {}
+        ok_flag = 1 if ok else 0
+        view_value = values.get("view")
+        open_view = view_value if ok else None
+        open_at = int(captured_at) if ok else None
         def run(conn: sqlite3.Connection) -> None:
             conn.execute(
                 """
                 INSERT INTO snapshot(stat_date, aid, "view", "like", favorite, coin,
-                                     share, reply, danmaku, captured_at, ok, err, src)
-                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                                     share, reply, danmaku, captured_at, ok, err, src,
+                                     view_open, open_captured_at)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
                 ON CONFLICT(stat_date, aid) DO UPDATE SET
                   "view" = excluded."view",
                   "like" = excluded."like",
@@ -707,12 +712,22 @@ class Repository:
                   captured_at = excluded.captured_at,
                   ok = excluded.ok,
                   err = excluded.err,
-                  src = excluded.src
+                  src = excluded.src,
+                  view_open = CASE
+                    WHEN excluded.ok = 1 AND snapshot.view_open IS NULL
+                    THEN excluded."view"
+                    ELSE snapshot.view_open
+                  END,
+                  open_captured_at = CASE
+                    WHEN excluded.ok = 1 AND snapshot.open_captured_at IS NULL
+                    THEN excluded.captured_at
+                    ELSE snapshot.open_captured_at
+                  END
                 """,
                 (
                     stat_date,
                     int(aid),
-                    values.get("view"),
+                    view_value,
                     values.get("like"),
                     values.get("favorite"),
                     values.get("coin"),
@@ -720,9 +735,11 @@ class Repository:
                     values.get("reply"),
                     values.get("danmaku"),
                     int(captured_at),
-                    1 if ok else 0,
+                    ok_flag,
                     err,
                     src,
+                    open_view,
+                    open_at,
                 ),
             )
         self.db.write(run)
@@ -741,18 +758,23 @@ class Repository:
                 """
                 INSERT INTO snapshot(
                   stat_date, aid, "view", "like", favorite, coin,
-                  share, reply, danmaku, captured_at, ok, err, src
+                  share, reply, danmaku, captured_at, ok, err, src,
+                  view_open, open_captured_at
                 )
                 SELECT
                   ?, aid, "view", "like", favorite, coin,
-                  share, reply, danmaku, ?, 1, NULL, ?
+                  share, reply, danmaku, ?, 1, NULL, ?,
+                  "view", ?
                 FROM snapshot
                 WHERE stat_date = ? AND ok = 1
                   AND aid NOT IN (
                     SELECT aid FROM snapshot WHERE stat_date = ?
                   )
                 """,
-                (target_date, int(captured_at), src, source_date, target_date),
+                (
+                    target_date, int(captured_at), src, int(captured_at),
+                    source_date, target_date,
+                ),
             )
             return int(cur.rowcount or 0)
         return self.db.write(run)

reaction_board/scheduler.py

@@ -3,6 +3,9 @@
 
 P0 只注册收藏夹同步任务；P1/P3 会在同一调度器上追加快照、粉丝和发现任务。
 调度只负责“什么时候运行”，具体任务保持幂等，并在自己的服务层处理重试。
+
+任务 ``fn`` 可返回 :class:`TaskControl` 以覆盖 ``ok`` / ``next_run``
+（用于静默窗延后、失败补跑等）；未返回时仍按 ``next_run_fn`` / ``interval`` 排期。
 """
 
 from __future__ import annotations
@@ -12,6 +15,15 @@ from dataclasses import dataclass
 from typing import Any, Callable, Sequence
 
 
+@dataclass
+class TaskControl:
+    """任务可选返回值：覆盖成功标记与下次运行时间。"""
+
+    ok: bool = True
+    next_run: float | None = None
+    error: str | None = None
+
+
 @dataclass
 class ScheduledTask:
     name: str
@@ -47,6 +59,7 @@ class Scheduler:
         clock=time.time,
         sleeper=time.sleep,
         logger=None,
+        due_reader: Callable[[str], float | None] | None = None,
     ) -> None:
         if not tasks:
             raise ValueError("至少需要一个调度任务")
@@ -57,6 +70,8 @@ class Scheduler:
         self._clock = clock
         self._sleeper = sleeper
         self.logger = logger
+        # 每轮从落盘 due_at 刷新 next_run（seed 挂当晚 seal 等同进程可见）
+        self._due_reader = due_reader
 
     def _log(self, level: str, event: str, **fields) -> None:
         if self.logger is not None:
@@ -66,27 +81,56 @@ class Scheduler:
         current = self._clock() if now is None else now
         results: list[TaskResult] = []
         for task in self.tasks:
+            if self._due_reader is not None:
+                disk_due = self._due_reader(task.name)
+                if disk_due is not None:
+                    task.next_run = float(disk_due)
             if task.next_run > current:
                 continue
             started = current
             error: str | None = None
+            control: TaskControl | None = None
             try:
-                task.fn()
+                raw = task.fn()
             except Exception as exc:  # noqa: BLE001  单任务失败不终止调度
                 error = str(exc)
-                self._log("error", "scheduler.task_error", task=task.name, error=error)
             else:
-                self._log("info", "scheduler.task_ok", task=task.name)
+                if isinstance(raw, TaskControl):
+                    control = raw
             finished = self._clock()
-            if task.next_run_fn is not None:
+            if control is not None and control.next_run is not None:
+                task.next_run = float(control.next_run)
+            elif task.next_run_fn is not None:
                 task.next_run = task.next_run_fn(started)
             else:
                 task.next_run = started + task.interval
+            ok = error is None
+            if control is not None:
+                ok = bool(control.ok) and error is None
+                if control.error and error is None:
+                    error = control.error
+            # Skip/fail ≠ success：先定 ok/error，再记 scheduler 日志。
+            if ok:
+                self._log("info", "scheduler.task_ok", task=task.name)
+            elif control is not None:
+                self._log(
+                    "warning",
+                    "scheduler.task_fail",
+                    task=task.name,
+                    error=error,
+                )
+            else:
+                self._log(
+                    "error",
+                    "scheduler.task_error",
+                    task=task.name,
+                    error=error,
+                )
             result = TaskResult(
                 name=task.name,
                 started_at=started,
                 finished_at=finished,
-                ok=error is None,
+                ok=ok,
                 error=error,
             )
             results.append(result)

reaction_board/services/snapshot.py

@@ -51,6 +51,7 @@ class SnapshotSummary:
     completed: bool = False
     already_completed: bool = False
     finished_at: int = 0
+    reason: str | None = None
     errors: list[str] = field(default_factory=list)
 
     def as_dict(self) -> dict[str, Any]:
@@ -73,6 +74,7 @@ class SnapshotSummary:
             "backup_error": self.backup_error,
             "completed": self.completed,
             "already_completed": self.already_completed,
+            "reason": self.reason,
             "errors": self.errors,
         }
 
@@ -140,12 +142,20 @@ class SnapshotService:
         return f"fans:seeded:{stat_date}"
 
     def run_if_needed(
-        self, *, date: str | None = None, force: bool = False
+        self,
+        *,
+        date: str | None = None,
+        force: bool = False,
+        keep_scraped: bool | None = None,
     ) -> SnapshotSummary:
         """视频快照（含 UP 头像补齐）；粉丝采集请用 ``run_fans_if_needed``。
 
         当日若仅有午夜 seed 壳且未 ``force``，直接返回等待晚间落盘，
         避免把「全员 skip」误标成 completed。
+
+        ``force`` + ``keep_scraped``：
+        - 晚间 seal / 断点续跑：``keep_scraped=True``（默认）保留已采 ok=1；
+        - CLI ``--force``：显式 ``keep_scraped=False`` 全量重采。
         """
         now = self._now()
         stat_date = date or shanghai_date(now)
@@ -153,7 +163,10 @@ class SnapshotService:
             raise ValueError(f"日期格式必须是 YYYY-MM-DD: {stat_date!r}")
         seeded = self.repo.kv_get(self.seeded_key(stat_date)) == "1"
         if force:
-            return self.run_videos(date=stat_date, force=True)
+            ks = True if keep_scraped is None else bool(keep_scraped)
+            return self.run_videos(
+                date=stat_date, force=True, keep_scraped=ks,
+            )
 
         completed = self.repo.kv_get(self.completed_key(stat_date)) == "1"
         if seeded and not completed:
@@ -167,6 +180,7 @@ class SnapshotService:
                 finished_at=now,
                 completed=False,
                 already_completed=False,
+                reason="seeded_pending_seal",
             )
         pending = self.repo.pending_snapshot_aids(stat_date)
         if completed and not pending:
@@ -203,8 +217,19 @@ class SnapshotService:
         summary = self.run_videos(date=date, force=force)
         return self.run_fans(date=summary.stat_date, force=force, summary=summary)
 
-    def run_videos(self, *, date: str | None = None, force: bool = False) -> SnapshotSummary:
-        """采集视频七项指标并在全部成功后标记完成日 / 备份。"""
+    def run_videos(
+        self,
+        *,
+        date: str | None = None,
+        force: bool = False,
+        keep_scraped: bool = False,
+    ) -> SnapshotSummary:
+        """采集视频七项指标并在全部成功后标记完成日 / 备份。
+
+        ``force``：忽略完成标记，重跑。
+        ``keep_scraped``：force 时仍跳过非 seed 的 ok=1（seal 断点续跑）；
+        CLI ``--force`` 不传此标志，会重采全部成功行。
+        """
         cfg = self.config_store.get()
         client = self._client()
         now = self._now()
@@ -225,11 +250,19 @@ class SnapshotService:
         summary.total = len(videos)
         for index, aid in enumerate(videos, 1):
             existing = self.repo.get_snapshot(stat_date, aid)
-            if existing is not None and int(existing.get("ok") or 0) == 1 and not force:
-                # 断点续跑：同日已成功则跳过。小时任务刷新过的行也是 ok=1，
-                # 自然覆盖「1 小时内 captured_at 已更新则不再次更新」的要求。
-                summary.skipped += 1
-                continue
+            if existing is not None and int(existing.get("ok") or 0) == 1:
+                src = str(existing.get("src") or "")
+                if keep_scraped and src == "seed":
+                    # seal 续跑：必须打穿克隆壳
+                    pass
+                elif keep_scraped and src != "seed":
+                    summary.skipped += 1
+                    continue
+                elif not force:
+                    # 普通续跑：同日已成功则跳过（含小时任务刚刷新的）
+                    summary.skipped += 1
+                    continue
+                # force 且非 keep_scraped：CLI 全量重采，不 skip
 
             stat, error, unavailable = self._fetch_view(client, aid)
             if unavailable:
@@ -310,7 +343,8 @@ class SnapshotService:
         """刷新近 N 天新视频的当日快照；不碰完成日标记与日备份。
 
         「近 N 天」按 pubdate 墙钟窗口：``now - recent_video_days * 86400``，
-        不是上海日历日差。允许覆盖同日已有 ``ok=1`` 行以便小时级刷新。
+        不是上海日历日差。当天入库但发布更早的旧片不进小时任务。
+        允许覆盖同日已有 ``ok=1`` 行以便小时级刷新。
         """
         cfg = self.config_store.get()
         client = self._client()
@@ -422,6 +456,7 @@ class SnapshotService:
                     completed=False,
                     already_completed=False,
                     fans_skipped=len(cfg.track_fans),
+                    reason="seeded_pending_seal",
                 )
             return self.run_fans(date=stat_date, force=True)
         if not self._pending_fan_mids(stat_date, cfg.track_fans):
@@ -535,15 +570,8 @@ class SnapshotService:
             raise ValueError(f"日期格式必须是 YYYY-MM-DD: {stat_date!r}")
 
         summary = SnapshotSummary(stat_date=stat_date, started_at=now)
-        source = self.repo.kv_get("snapshot:last_completed")
-        if not source or source >= stat_date:
-            self._log(
-                "info", "snapshot.seed_skip",
-                stat_date=stat_date, reason="no_source", source=source,
-            )
-            summary.finished_at = self._now()
-            summary.completed = True
-            return summary
+        # 目标日已完成：先于 no_source。全站快照当天完成后
+        # last_completed == stat_date，不可误判为无源再 10 分钟重试。
         if self.repo.kv_get(self.completed_key(stat_date)) == "1":
             self._log(
                 "info", "snapshot.seed_skip",
@@ -552,6 +580,17 @@ class SnapshotService:
             summary.finished_at = self._now()
             summary.completed = True
             summary.already_completed = True
+            summary.reason = "already_completed"
+            return summary
+        source = self.repo.kv_get("snapshot:last_completed")
+        if not source or source >= stat_date:
+            self._log(
+                "info", "snapshot.seed_skip",
+                stat_date=stat_date, reason="no_source", source=source,
+            )
+            summary.finished_at = self._now()
+            summary.completed = False
+            summary.reason = "no_source"
             return summary
         if self.repo.kv_get(self.seeded_key(stat_date)) == "1":
             self._log(
@@ -561,6 +600,7 @@ class SnapshotService:
             summary.finished_at = self._now()
             summary.completed = True
             summary.already_completed = True
+            summary.reason = "already_seeded"
             return summary
 
         copied = self.repo.copy_ok_snapshots(
@@ -582,6 +622,7 @@ class SnapshotService:
                 self.repo.bump_cache_version("fans")
         summary.finished_at = self._now()
         summary.completed = True  # seed 任务本身成功；不等于快照落盘完成
+        summary.reason = "seeded"
         self._log(
             "info", "snapshot.seed_done",
             stat_date=stat_date, source=source,

reaction_board/task_state.py

@@ -0,0 +1,438 @@
+# -*- coding: utf-8 -*-
+"""调度任务状态持久化（跨进程重启 / 容器重建）。
+
+每条任务一条 kv 记录 ``task:state:{name}``，字段见 :class:`TaskState`。
+调度器只认落盘的 ``due_at`` / ``status`` / ``stat_date`` / ``progress``，
+不再靠启动时把 ``next_run=0`` 猜「要不要立刻跑」。
+"""
+
+from __future__ import annotations
+
+import json
+import time
+from dataclasses import asdict, dataclass
+from datetime import timedelta
+from typing import Any, Mapping
+
+from .repository import Repository
+from .timeutil import parse_hhmm, shanghai_date, shanghai_datetime
+
+STATUS_PENDING = "pending"
+STATUS_RUNNING = "running"
+STATUS_DONE = "done"
+STATUS_FAILED = "failed"
+STATUS_DEFERRED = "deferred"
+
+_ACTIVE = frozenset({STATUS_PENDING, STATUS_RUNNING, STATUS_FAILED, STATUS_DEFERRED})
+
+
+@dataclass
+class TaskState:
+    name: str
+    status: str
+    due_at: float
+    stat_date: str | None = None
+    progress: dict[str, Any] | None = None
+    updated_at: float = 0.0
+    error: str | None = None
+
+    def as_dict(self) -> dict[str, Any]:
+        payload = asdict(self)
+        # JSON 友好：去掉 None progress 也可保留
+        return payload
+
+
+def state_key(name: str) -> str:
+    return f"task:state:{name}"
+
+
+def load_state(repo: Repository, name: str) -> TaskState | None:
+    raw = repo.kv_get(state_key(name))
+    if not raw:
+        return None
+    try:
+        data = json.loads(raw)
+    except (TypeError, ValueError, json.JSONDecodeError):
+        return None
+    if not isinstance(data, dict) or not data.get("name"):
+        return None
+    try:
+        due_at = float(data.get("due_at") or 0)
+        updated_at = float(data.get("updated_at") or 0)
+    except (TypeError, ValueError):
+        return None
+    progress = data.get("progress")
+    if progress is not None and not isinstance(progress, dict):
+        progress = None
+    return TaskState(
+        name=str(data.get("name") or name),
+        status=str(data.get("status") or STATUS_PENDING),
+        due_at=due_at,
+        stat_date=(str(data["stat_date"]) if data.get("stat_date") else None),
+        progress=progress,
+        updated_at=updated_at,
+        error=(str(data["error"]) if data.get("error") else None),
+    )
+
+
+def save_state(repo: Repository, state: TaskState) -> None:
+    repo.kv_set(state_key(state.name), json.dumps(state.as_dict(), ensure_ascii=False))
+
+
+def _now(now: float | None) -> float:
+    return time.time() if now is None else float(now)
+
+
+def recover_stale_running(
+    repo: Repository,
+    name: str,
+    *,
+    now: float | None = None,
+) -> TaskState | None:
+    """进程已死但仍为 ``running``：改为 ``pending``，保留 progress / stat_date。
+
+    ``due_at`` 取 ``min(原 due_at, now)``，使过期/中断任务可立即续跑，
+    未到期的（极少见）仍尊重原 due_at。
+    """
+    current = _now(now)
+    state = load_state(repo, name)
+    if state is None or state.status != STATUS_RUNNING:
+        return state
+    state.status = STATUS_PENDING
+    state.due_at = min(float(state.due_at or current), current)
+    state.updated_at = current
+    save_state(repo, state)
+    return state
+
+
+def mark_running(
+    repo: Repository,
+    name: str,
+    *,
+    stat_date: str | None,
+    now: float | None = None,
+    progress: Mapping[str, Any] | None = None,
+    due_at: float | None = None,
+) -> TaskState:
+    current = _now(now)
+    prev = load_state(repo, name)
+    state = TaskState(
+        name=name,
+        status=STATUS_RUNNING,
+        due_at=float(due_at if due_at is not None else (prev.due_at if prev else current)),
+        stat_date=stat_date if stat_date is not None else (prev.stat_date if prev else None),
+        progress=dict(progress) if progress is not None else (prev.progress if prev else None),
+        updated_at=current,
+        error=None,
+    )
+    save_state(repo, state)
+    return state
+
+
+def mark_done(
+    repo: Repository,
+    name: str,
+    *,
+    due_at: float,
+    now: float | None = None,
+    stat_date: str | None = None,
+    progress: Mapping[str, Any] | None = None,
+) -> TaskState:
+    """成功：记 done，并写入下一次 ``due_at``（下一条墙钟/间隔）。"""
+    current = _now(now)
+    state = TaskState(
+        name=name,
+        status=STATUS_DONE,
+        due_at=float(due_at),
+        stat_date=stat_date,
+        progress=dict(progress) if progress is not None else None,
+        updated_at=current,
+        error=None,
+    )
+    save_state(repo, state)
+    return state
+
+
+def mark_failed(
+    repo: Repository,
+    name: str,
+    *,
+    due_at: float,
+    now: float | None = None,
+    stat_date: str | None = None,
+    progress: Mapping[str, Any] | None = None,
+    error: str | None = None,
+) -> TaskState:
+    current = _now(now)
+    prev = load_state(repo, name)
+    state = TaskState(
+        name=name,
+        status=STATUS_FAILED,
+        due_at=float(due_at),
+        stat_date=stat_date if stat_date is not None else (prev.stat_date if prev else None),
+        progress=(
+            dict(progress) if progress is not None
+            else (prev.progress if prev else None)
+        ),
+        updated_at=current,
+        error=error,
+    )
+    save_state(repo, state)
+    return state
+
+
+def mark_deferred(
+    repo: Repository,
+    name: str,
+    *,
+    due_at: float,
+    now: float | None = None,
+    stat_date: str | None = None,
+    progress: Mapping[str, Any] | None = None,
+    error: str | None = None,
+) -> TaskState:
+    """主动延后（静默窗 / seed 避让等）：保留进度，写未来 due_at。"""
+    current = _now(now)
+    prev = load_state(repo, name)
+    state = TaskState(
+        name=name,
+        status=STATUS_DEFERRED,
+        due_at=float(due_at),
+        stat_date=stat_date if stat_date is not None else (prev.stat_date if prev else None),
+        progress=(
+            dict(progress) if progress is not None
+            else (prev.progress if prev else None)
+        ),
+        updated_at=current,
+        error=error,
+    )
+    save_state(repo, state)
+    return state
+
+
+def ensure_pending(
+    repo: Repository,
+    name: str,
+    *,
+    due_at: float,
+    stat_date: str | None,
+    now: float | None = None,
+    overwrite_done: bool = True,
+) -> TaskState:
+    """保证存在一条 pending（如 seed 成功后挂上当晚 seal）。
+
+    同日未完成：取更早的 due_at。
+    跨日：不得把新一天的 due 粘到昨天的 ``stat_date`` 上——
+    若已有更早日期的未完成记录，保留它（跨天续跑优先）；
+    若已有更新日期的未完成，则重置为调用方的 due/stat_date。
+    ``overwrite_done`` 时覆盖已完成记录。
+    """
+    current = _now(now)
+    prev = load_state(repo, name)
+    if prev is not None and prev.status in _ACTIVE and prev.status != STATUS_RUNNING:
+        if (
+            stat_date is not None
+            and prev.stat_date
+            and prev.stat_date != stat_date
+        ):
+            # 未完成的更早归属日优先保留，避免 seed 挂今晚时冲掉昨天续跑。
+            if prev.stat_date < stat_date:
+                return prev
+            # prev 归属日更新：按调用方重置，禁止跨日字段拼接。
+            state = TaskState(
+                name=name,
+                status=STATUS_PENDING,
+                due_at=float(due_at),
+                stat_date=stat_date,
+                progress=None,
+                updated_at=current,
+                error=None,
+            )
+            save_state(repo, state)
+            return state
+        # 同日（或缺一边 stat_date）：取更早 due_at，补上缺失的 stat_date
+        due = min(float(prev.due_at or due_at), float(due_at))
+        sd = prev.stat_date or stat_date
+        if due == prev.due_at and sd == prev.stat_date and prev.status == STATUS_PENDING:
+            return prev
+        prev.due_at = due
+        prev.stat_date = sd
+        prev.status = STATUS_PENDING
+        prev.updated_at = current
+        save_state(repo, prev)
+        return prev
+    if prev is not None and prev.status == STATUS_RUNNING:
+        return prev
+    if prev is not None and prev.status == STATUS_DONE and not overwrite_done:
+        return prev
+    state = TaskState(
+        name=name,
+        status=STATUS_PENDING,
+        due_at=float(due_at),
+        stat_date=stat_date,
+        progress=None,
+        updated_at=current,
+        error=None,
+    )
+    save_state(repo, state)
+    return state
+
+
+def bootstrap_from_last(
+    repo: Repository,
+    name: str,
+    *,
+    wall_due_at: float,
+    now: float | None = None,
+    retry_seconds: float = 300.0,
+) -> TaskState | None:
+    """升级/冷启动：无 ``task:state`` 时从 ``task:last`` 回填。
+
+    - 成功 → ``done`` + ``due_at=wall_due_at``
+    - 失败/跳过 → ``failed``/``deferred``，``due_at`` 为补跑点（已过则立刻）
+    """
+    current = _now(now)
+    existing = load_state(repo, name)
+    if existing is not None:
+        return existing
+    raw = repo.kv_get(f"task:last:{name}")
+    if not raw:
+        return None
+    try:
+        data = json.loads(raw)
+    except (TypeError, ValueError, json.JSONDecodeError):
+        return None
+    if not isinstance(data, dict):
+        return None
+    ok = bool(data.get("ok"))
+    skipped = bool(data.get("skipped"))
+    try:
+        finished = float(data.get("finished_at") or data.get("started_at") or current)
+    except (TypeError, ValueError):
+        finished = current
+    summary = data.get("summary")
+    stat_date = None
+    if isinstance(summary, dict) and summary.get("stat_date"):
+        stat_date = str(summary["stat_date"])
+    error = data.get("error")
+    error_s = str(error) if error else None
+    if ok and not skipped:
+        state = TaskState(
+            name=name,
+            status=STATUS_DONE,
+            due_at=float(wall_due_at),
+            stat_date=stat_date,
+            updated_at=current,
+            error=None,
+        )
+        save_state(repo, state)
+        return state
+    due = finished + float(retry_seconds)
+    if due > current:
+        pass
+    else:
+        due = current
+    state = TaskState(
+        name=name,
+        status=STATUS_DEFERRED if skipped else STATUS_FAILED,
+        due_at=float(due),
+        stat_date=stat_date,
+        updated_at=current,
+        error=error_s,
+    )
+    save_state(repo, state)
+    return state
+
+
+def bootstrap_seed_markers(
+    repo: Repository,
+    name: str,
+    *,
+    slot_hhmm: str,
+    wall_due_at: float,
+    now: float | None = None,
+) -> TaskState | None:
+    """无 state 时：若存在未 seal 的 seed 壳，挂 pending（跨天优先昨天）。
+
+    已过当日 ``slot_hhmm``（或归属日早于今天）→ ``due_at=now`` 尽快补跑。
+    """
+    del wall_due_at  # 未完成 seed 不走纯墙钟
+    current = _now(now)
+    existing = load_state(repo, name)
+    if existing is not None:
+        return existing
+    if name not in {"snapshot", "fans"}:
+        return None
+    today = shanghai_date(current)
+    yday = (shanghai_datetime(current) - timedelta(days=1)).strftime("%Y-%m-%d")
+    hour, minute = parse_hhmm(slot_hhmm)
+    today_slot = shanghai_datetime(current).replace(
+        hour=hour, minute=minute, second=0, microsecond=0,
+    ).timestamp()
+
+    def _seeded_unfinished(day: str) -> bool:
+        if name == "snapshot":
+            if repo.kv_get(f"snapshot:seeded:{day}") != "1":
+                return False
+            return repo.kv_get(f"snapshot:completed:{day}") != "1"
+        # fans：seed 键在成功 seal 后删除，有键即未完成
+        return repo.kv_get(f"fans:seeded:{day}") == "1"
+
+    for day in (yday, today):
+        if not _seeded_unfinished(day):
+            continue
+        if day < today or current >= today_slot:
+            due = current
+        else:
+            due = float(today_slot)
+        return ensure_pending(
+            repo, name, due_at=due, stat_date=day, now=current,
+        )
+    return None
+
+
+def startup_next_run(
+    repo: Repository,
+    name: str,
+    *,
+    wall_due_at: float,
+    now: float | None = None,
+    slot_hhmm: str | None = None,
+    retry_seconds: float = 300.0,
+) -> float:
+    """常驻启动时的 ``ScheduledTask.next_run``：只认落盘状态。
+
+    - 无状态时先 ``bootstrap_from_last`` / seed 标记回填
+    - 恢复残留 ``running`` → ``pending``
+    - ``due_at`` 是唯一时间门闩（失败延时不会被提前）
+    """
+    current = _now(now)
+    if load_state(repo, name) is None:
+        bootstrap_from_last(
+            repo, name, wall_due_at=wall_due_at, now=current,
+            retry_seconds=retry_seconds,
+        )
+        if slot_hhmm:
+            bootstrap_seed_markers(
+                repo, name, slot_hhmm=slot_hhmm,
+                wall_due_at=wall_due_at, now=current,
+            )
+    state = recover_stale_running(repo, name, now=current)
+    if state is None:
+        return float(wall_due_at)
+    due = float(state.due_at or 0)
+    if due <= 0:
+        return float(wall_due_at)
+    # 无论 status：due_at 是唯一时间门闩
+    if due > current:
+        return due
+    # 已到期：未完成要续跑；done 也表示下一轮已到期
+    return due
+
+
+def resume_stat_date(repo: Repository, name: str, *, fallback: str) -> str:
+    """续跑时优先用状态里的归属日，避免跨日重启跑错天。"""
+    state = load_state(repo, name)
+    if state is not None and state.stat_date and state.status in _ACTIVE:
+        return state.stat_date
+    return fallback

reaction_board/timeutil.py

@@ -57,3 +57,39 @@ def day_bounds(date: str) -> tuple[float, float]:
         raise ValueError(f"日期格式必须是 YYYY-MM-DD: {date!r}")
     start = datetime.strptime(date, "%Y-%m-%d").replace(tzinfo=SHANGHAI_TZ)
     return start.timestamp(), (start + timedelta(days=1)).timestamp()
+
+
+# 全站快照静默窗：上海时区 [23:00, 01:00)。此间 discovery / recent / cleanup 不执行。
+SNAPSHOT_QUIET_START_HOUR = 23
+SNAPSHOT_QUIET_END_HOUR = 1
+
+
+def in_snapshot_quiet_window(ts: float | None = None) -> bool:
+    """是否处于全站快照静默窗 [23:00, 01:00)（上海时区，跨日）。"""
+    dt = shanghai_datetime(ts)
+    return dt.hour >= SNAPSHOT_QUIET_START_HOUR or dt.hour < SNAPSHOT_QUIET_END_HOUR
+
+
+def next_after_snapshot_quiet_window(ts: float) -> float:
+    """若 ``ts`` 落在静默窗内，推到窗结束后的 01:00；否则原样返回。"""
+    if not in_snapshot_quiet_window(ts):
+        return ts
+    dt = shanghai_datetime(ts)
+    if dt.hour >= SNAPSHOT_QUIET_START_HOUR:
+        target = (dt + timedelta(days=1)).replace(
+            hour=SNAPSHOT_QUIET_END_HOUR, minute=0, second=0, microsecond=0,
+        )
+    else:
+        target = dt.replace(
+            hour=SNAPSHOT_QUIET_END_HOUR, minute=0, second=0, microsecond=0,
+        )
+    if target.timestamp() <= ts:
+        target += timedelta(days=1)
+    return target.timestamp()
+
+
+def interval_next_run(now: float, interval: float) -> float:
+    """``now + interval``，若落入静默窗则推到 01:00 之后。"""
+    if interval <= 0:
+        raise ValueError("interval 必须大于 0")
+    return next_after_snapshot_quiet_window(now + interval)

reaction_board/web/queries.py

@@ -20,6 +20,7 @@ from ..analytics import (
     METRICS,
     MIN_VIEW_FOR_GROWTH_RATE,
     compute_growth,
+    smoothed_growth_rate,
     content_hours,
     eb_rate,
     effective_shrink_minutes,
@@ -33,7 +34,14 @@ from ..analytics import (
 )
 from ..config import ConfigStore
 from ..database import Database
-from ..timeutil import day_bounds, next_daily_run, shanghai_date, shanghai_datetime
+from ..timeutil import (
+    day_bounds,
+    interval_next_run,
+    next_after_snapshot_quiet_window,
+    next_daily_run,
+    shanghai_date,
+    shanghai_datetime,
+)
 from .cache import DASHBOARD_CACHE, TTL_FANS, TTL_VIDEOS, build_cache_key
 
 TRACKED_STATUSES = ("active", "unfaved")
@@ -58,6 +66,19 @@ def _chunks(values: Sequence[int], size: int = 400) -> Iterable[Sequence[int]]:
         yield values[index:index + size]
 
 
+def _growth_baseline_view(item: Mapping[str, Any]) -> int | None:
+    """增速分母：当前播放减去区间增量。zero 基线的增量等于当前播放，结果为 0。"""
+    growth = item.get("growth") or {}
+    current = (item.get("metrics") or {}).get("view")
+    delta = growth.get("delta")
+    if current is None or delta is None:
+        return None
+    try:
+        return int(current) - int(delta)
+    except (TypeError, ValueError):
+        return None
+
+
 def _date_minus(day: str, days: int) -> str:
     return (date.fromisoformat(day) - timedelta(days=int(days))).isoformat()
 
@@ -385,6 +406,18 @@ class DashboardQueries:
         percentiles = score_percentiles(scores)
         return videos, snapshots, scores, percentiles
 
+    @staticmethod
+    def _publish_date(pubdate: object) -> str | None:
+        if pubdate is None or isinstance(pubdate, bool):
+            return None
+        try:
+            ts = int(pubdate)
+        except (TypeError, ValueError):
+            return None
+        if ts <= 0:
+            return None
+        return shanghai_date(ts)
+
     @staticmethod
     def _growth_for(
         snapshot: Mapping[str, Any] | None,
@@ -406,6 +439,70 @@ class DashboardQueries:
                 gap = None
         return compute_growth(current, before, gap)
 
+    @classmethod
+    def _same_day_growth(
+        cls,
+        snapshot: Mapping[str, Any] | None,
+        video: Mapping[str, Any],
+        metric: str = "view",
+    ) -> dict[str, Any] | None:
+        """近 1 日、没有更早快照时：当日发布相对 0；否则用 view_open。"""
+        if snapshot is None:
+            return None
+        current = snapshot.get(metric)
+        if current is None:
+            return None
+        stat_date = str(snapshot.get("stat_date") or "")
+        published_on = cls._publish_date(video.get("pubdate"))
+        if published_on is not None and published_on == stat_date:
+            growth = compute_growth(current, 0, 0, allow_zero_gap=True)
+            payload = growth.as_dict()
+            payload["baseline"] = "zero"
+            # previous=0 时 rate = view/1，会按原始播放碾压真日环比。
+            payload["rate"] = None
+            return payload
+        try:
+            open_at = snapshot.get("open_captured_at")
+            captured_at = snapshot.get("captured_at")
+            view_open = snapshot.get("view_open")
+            open_ready = (
+                open_at is not None
+                and captured_at is not None
+                and int(open_at) < int(captured_at)
+                and view_open is not None
+            )
+        except (TypeError, ValueError):
+            open_ready = False
+        if not open_ready:
+            return None
+        growth = compute_growth(current, view_open, 0, allow_zero_gap=True)
+        payload = growth.as_dict()
+        payload["baseline"] = "open"
+        return payload
+
+    @classmethod
+    def _growth_result(
+        cls,
+        video: Mapping[str, Any],
+        snapshot: Mapping[str, Any] | None,
+        previous: Mapping[str, Any] | None,
+        *,
+        days_back: int,
+        metric: str = "view",
+    ) -> dict[str, Any]:
+        if previous is not None:
+            growth = cls._growth_for(snapshot, previous, metric=metric)
+            payload = growth.as_dict()
+            payload["baseline"] = "snapshot" if growth.available else None
+            return payload
+        if days_back == 1:
+            fallback = cls._same_day_growth(snapshot, video, metric=metric)
+            if fallback is not None:
+                return fallback
+        payload = compute_growth(None, None, None).as_dict()
+        payload["baseline"] = None
+        return payload
+
     def _item(
         self,
         video: Mapping[str, Any],
@@ -415,6 +512,7 @@ class DashboardQueries:
         percentile: float | None = None,
         previous: Mapping[str, Any] | None = None,
         include_growth: bool = False,
+        growth_days: int | None = None,
     ) -> dict[str, Any]:
         metrics = _metric_map(snapshot)
         duration = video.get("duration")
@@ -455,7 +553,9 @@ class DashboardQueries:
             ),
         }
         if include_growth:
-            item["growth"] = self._growth_for(snapshot, previous).as_dict()
+            item["growth"] = self._growth_result(
+                video, snapshot, previous, days_back=int(growth_days or 1),
+            )
         return item
 
     # ------------------------------------------------------------ overview
@@ -474,7 +574,7 @@ class DashboardQueries:
             videos = self._videos(conn)
             aids = [int(video["aid"]) for video in videos]
             snapshots = self._snapshot_map(conn, aids)
-            # 播放增量：按片相对当前点的相邻成功日，避免 seed/recent 拉到 T-2。
+            # 播放增量：近 1 日。有更早日用相邻成功日；当日发布相对 0；否则用 view_open。
             previous = self._snapshot_previous_relative(conn, snapshots, days_back=1)
             totals = {metric: 0 for metric in METRICS}
             for video in videos:
@@ -489,12 +589,13 @@ class DashboardQueries:
             for video in videos:
                 aid = int(video["aid"])
                 snapshot = snapshots.get(aid)
-                prior = previous.get(aid)
-                if snapshot is None or prior is None:
+                if snapshot is None:
                     continue
-                growth = self._growth_for(snapshot, prior)
-                if growth.available and growth.delta is not None:
-                    view_delta_sum += growth.delta
+                growth = self._growth_result(
+                    video, snapshot, previous.get(aid), days_back=1,
+                )
+                if growth.get("available") and growth.get("delta") is not None:
+                    view_delta_sum += int(growth["delta"])
                     available_deltas += 1
             if available_deltas:
                 view_delta = view_delta_sum
@@ -881,6 +982,7 @@ class DashboardQueries:
                     percentile=percentiles.get(int(video["aid"])),
                     previous=previous.get(int(video["aid"])),
                     include_growth=sort in {"growth", "growth_rate"},
+                    growth_days=growth_days,
                 )
                 for video in videos
             ]
@@ -952,9 +1054,13 @@ class DashboardQueries:
                 if sort == "growth_rate":
                     items = [
                         item for item in items
-                        if (item["metrics"] or {}).get("view", 0) >= MIN_VIEW_FOR_GROWTH_RATE
+                        if item.get("growth", {}).get("baseline") != "zero"
+                        and item.get("growth", {}).get("rate") is not None
+                        and (_growth_baseline_view(item) or 0) >= MIN_VIEW_FOR_GROWTH_RATE
                     ]
 
+            rate_prior = max(0, int(analytics.growth_rate_prior))
+
             def sort_key(item: Mapping[str, Any]):
                 if sort == "added":
                     return (int(item["first_seen"]), int(item["pubdate"]))
@@ -971,9 +1077,18 @@ class DashboardQueries:
                     )
                     value = item.get(field)
                     return value if value is not None else float("-inf")
-                if sort in {"growth", "growth_rate"}:
-                    field = "delta" if sort == "growth" else "rate"
-                    value = item.get("growth", {}).get(field)
+                if sort == "growth":
+                    value = item.get("growth", {}).get("delta")
+                    return value if value is not None else float("-inf")
+                if sort == "growth_rate":
+                    growth = item.get("growth") or {}
+                    delta = growth.get("delta")
+                    current = (item.get("metrics") or {}).get("view")
+                    if delta is None or current is None:
+                        return float("-inf")
+                    value = smoothed_growth_rate(
+                        delta, int(current) - int(delta), rate_prior,
+                    )
                     return value if value is not None else float("-inf")
                 value = (item["metrics"] or {}).get(sort)
                 return value if value is not None else -1
@@ -990,9 +1105,20 @@ class DashboardQueries:
             reason = None
             if sort in {"growth", "growth_rate"} and not growth_available:
                 available = False
-                reason = (
-                    f"近 {growth_days} 日增长需要至少 {growth_days + 1} 天快照"
-                )
+                if growth_days == 1 and sort == "growth_rate":
+                    reason = (
+                        "近 1 日增速需要基线播放至少 "
+                        f"{MIN_VIEW_FOR_GROWTH_RATE}；当日发布不计入增速"
+                    )
+                elif growth_days == 1:
+                    reason = (
+                        "近 1 日增长需要前一日快照；当日发布有一次成功采样即可；"
+                        "更早发布的新入库需要当天至少两次成功采样"
+                    )
+                else:
+                    reason = (
+                        f"近 {growth_days} 日增长需要至少 {growth_days + 1} 天快照"
+                    )
             duration_filter = None
             if sort in duration_sorts:
                 duration_filter = {
@@ -1023,6 +1149,7 @@ class DashboardQueries:
                 "available": available,
                 "reason": reason,
                 "duration_filter": duration_filter,
+                "growth_rate_prior": rate_prior if sort == "growth_rate" else None,
             }
         return self.db.read(run)
 
@@ -1044,16 +1171,19 @@ class DashboardQueries:
                 video = dict(row)
             snapshots = self._snapshot_map(conn, [aid])
             snapshot = snapshots.get(aid)
-            previous = self._snapshot_map(conn, [aid], before=snapshot["stat_date"] if snapshot else None).get(aid)
+            previous = self._snapshot_map(
+                conn, [aid], before=snapshot["stat_date"] if snapshot else None,
+            ).get(aid)
             _, _, heat, percentiles = self._global_heat(conn)
-            growth = self._growth_for(snapshot, previous)
             item = self._item(
                 video,
                 snapshot,
                 heat=heat.get(aid),
                 percentile=percentiles.get(aid),
             )
-            item["growth"] = growth.as_dict()
+            item["growth"] = self._growth_result(
+                video, snapshot, previous, days_back=1,
+            )
             item["snapshot_count"] = int(conn.execute(
                 "SELECT COUNT(*) FROM snapshot WHERE aid = ?", (aid,)
             ).fetchone()[0])
@@ -1666,6 +1796,7 @@ class DashboardQueries:
                     "video_shrink_minutes": cfg.analytics.video_shrink_minutes,
                     "shrink_minutes": cfg.analytics.shrink_minutes,
                     "home_growth_days": cfg.analytics.home_growth_days,
+                    "growth_rate_prior": cfg.analytics.growth_rate_prior,
                 },
                 "heat_weights": {
                     metric: float(value)
@@ -1847,9 +1978,27 @@ class DashboardQueries:
                 "fans": int(next_daily_run(now, cfg.schedule.snapshot_at)),
                 "snapshot": int(next_daily_run(now, cfg.schedule.snapshot_at)),
                 "snapshot_seed": int(next_daily_run(now, cfg.schedule.seed_at)),
-                "discover": int(now + discover_interval),
-                "snapshot_recent": int(now + recent_interval),
+                "discover": int(interval_next_run(now, float(discover_interval))),
+                "snapshot_recent": int(interval_next_run(now, float(recent_interval))),
             }
+            retry_seconds = 300
+            seed_retry_seconds = 600
+            # 优先 task:state.due_at（跨重启唯一时间门闩）；无则回退 task:last 推导。
+            state_rows = conn.execute(
+                "SELECT k, v FROM kv WHERE k LIKE 'task:state:%'"
+            ).fetchall()
+            for row in state_rows:
+                key = str(row["k"]).removeprefix("task:state:")
+                ui_name = "discover" if key == "discovery" else key
+                if ui_name not in next_runs:
+                    continue
+                try:
+                    state = json.loads(str(row["v"]))
+                    due = float(state.get("due_at") or 0)
+                except (TypeError, ValueError, json.JSONDecodeError):
+                    continue
+                if due > 0:
+                    next_runs[ui_name] = int(due)
             for name, key, at in (
                 ("fans", "fans", cfg.schedule.snapshot_at),
                 ("snapshot", "snapshot", cfg.schedule.snapshot_at),
@@ -1857,23 +2006,58 @@ class DashboardQueries:
                 ("discover", "discovery", None),
                 ("snapshot_recent", "snapshot_recent", None),
             ):
+                # 已有落盘 due_at 则不再用 last 覆盖
+                state_raw = self._kv(conn, f"task:state:{key}")
+                if state_raw:
+                    try:
+                        if float(json.loads(state_raw).get("due_at") or 0) > 0:
+                            continue
+                    except (TypeError, ValueError, json.JSONDecodeError):
+                        pass
                 record = tasks.get(key)
                 if not (isinstance(record, dict) and record.get("finished_at")):
                     continue
                 finished_at = int(record["finished_at"])
+                summary = record.get("summary") if isinstance(record.get("summary"), dict) else {}
+                skipped = bool(record.get("skipped") or summary.get("skipped"))
+                failed = record.get("ok") is False
                 if name in {"discover", "snapshot_recent"}:
-                    # 只按上次完成推下一档；进行中由 running 展示，不按墙钟抢跑。
+                    # 上次完成推下一档；skip/失败短补跑；静默窗推到 01:00 后。
                     interval = (
                         recent_interval if name == "snapshot_recent"
                         else discover_interval
                     )
-                    next_runs[name] = finished_at + interval
+                    if skipped:
+                        next_runs[name] = int(
+                            next_after_snapshot_quiet_window(float(finished_at))
+                        )
+                    elif failed:
+                        next_runs[name] = int(finished_at + retry_seconds)
+                    else:
+                        next_runs[name] = int(
+                            interval_next_run(float(finished_at), float(interval))
+                        )
                 else:
-                    next_runs[name] = int(next_daily_run(finished_at, at))
+                    if skipped or failed:
+                        # 日更任务失败/skip：短间隔补跑，绝不按成功推到次日。
+                        # snapshot_seed：全站快照占用 / no_source 时 10 分钟补跑。
+                        gap = (
+                            seed_retry_seconds if name == "snapshot_seed"
+                            else retry_seconds
+                        )
+                        next_runs[name] = int(finished_at + gap)
+                    else:
+                        next_runs[name] = int(next_daily_run(finished_at, at))
             errors = [
                 {"task": name, **record}
                 for name, record in tasks.items()
-                if isinstance(record, dict) and record.get("ok") is False
+                if isinstance(record, dict)
+                and record.get("ok") is False
+                and not record.get("skipped")
+                and not (
+                    isinstance(record.get("summary"), dict)
+                    and record["summary"].get("skipped")
+                )
             ]
             counts = {str(row["status"]): int(row["n"]) for row in conn.execute(
                 "SELECT status, COUNT(*) AS n FROM video GROUP BY status"

reaction_board/web/static/components.js

@@ -1,6 +1,6 @@
 import {
   coverUrl, escapeAttr, escapeHtml, formatCount, formatDate,
-  formatDelta, formatDuration, formatPercent, metricLabel, statusLabel,
+  formatDelta, formatDuration, formatPercent, formatSignedPercent, metricLabel, statusLabel,
 } from "./format.js";
 
 export function cover(item, className = "thumb") {
@@ -30,9 +30,15 @@ export function avatar(item, className = "avatar") {
     loading="lazy" referrerpolicy="no-referrer"></div>`;
 }
 
+function growthBaselineTag(growth) {
+  if (growth?.baseline === "zero") return "<small>今日</small>";
+  if (growth?.baseline === "open") return "<small>入库后</small>";
+  return "";
+}
+
 export function videoCard(item, {
   rank = null, primary = "view", primaryLabel = "", primaryText = null,
-  showStatus = false,
+  showStatus = false, growthField = "delta",
 } = {}) {
   const metrics = item.metrics || {};
   const hasMetric = Object.prototype.hasOwnProperty.call(metrics, primary);
@@ -43,7 +49,13 @@ export function videoCard(item, {
   const primaryValue = typeof raw === "number" ? formatCount(raw) : raw;
   const label = primaryLabel || metricLabel(primary);
   const growth = item.growth;
-  const delta = growth?.available ? growth.delta : null;
+  const showRate = growthField === "rate";
+  const rate = showRate && growth?.available ? growth.rate : null;
+  const delta = !showRate && growth?.available ? growth.delta : null;
+  const change = showRate
+    ? (rate === null || rate === undefined ? null : formatSignedPercent(rate))
+    : (delta === null || delta === undefined ? null : formatDelta(delta));
+  const changeUp = showRate ? Number(rate) >= 0 : Number(delta) >= 0;
   const title = escapeHtml(item.title || "(无标题)");
   const up = escapeHtml(item.up_name || "未知 UP");
   const upMid = item.up_mid;
@@ -73,7 +85,7 @@ export function videoCard(item, {
           <span class="primary-stat">
             <b>${escapeHtml(primaryValue ?? "—")}</b><small>${escapeHtml(label)}</small>
           </span>
-          ${delta !== null ? `<span class="delta ${delta >= 0 ? "up" : "down"}">${formatDelta(delta)}</span>` : ""}
+          ${change !== null ? `<span class="delta ${changeUp ? "up" : "down"}">${change}${growthBaselineTag(growth)}</span>` : ""}
           <span class="secondary-stat">互动 ${formatPercent(item.interaction_rate)}</span>
         </div>
       </div>
@@ -118,7 +130,7 @@ export function metricGrid(item) {
   return `<div class="metric-grid">
     ${["view", "like", "coin", "favorite", "share", "reply", "danmaku"].map((metric) => {
       const extra = metric === "view" && growth?.available
-        ? `<span class="metric-delta ${growth.delta >= 0 ? "up" : "down"}">${formatDelta(growth.delta)}</span>`
+        ? `<span class="metric-delta ${growth.delta >= 0 ? "up" : "down"}">${formatDelta(growth.delta)}${growthBaselineTag(growth)}</span>`
         : "";
       return `<div class="metric-cell">
         <span class="metric-label">${escapeHtml(metricLabel(metric))}</span>

reaction_board/web/static/format.js

@@ -46,6 +46,12 @@ export function formatPercent(value, digits = 2) {
   return `${(Number(value) * 100).toFixed(digits)}%`;
 }
 
+export function formatSignedPercent(value, digits = 2) {
+  const text = formatPercent(value, digits);
+  if (text === "—" || Number(value) <= 0) return text;
+  return `+${text}`;
+}
+
 export function formatDate(timestamp) {
   if (!timestamp) return "—";
   const date = new Date(Number(timestamp) * 1000);

reaction_board/web/static/style.css

@@ -353,7 +353,7 @@ a.live-chip:hover { border-color: color-mix(in srgb, var(--accent) 55%, var(--bo
   transition: border-color 0.15s cubic-bezier(0.22, 1, 0.36, 1),
     background-color 0.15s cubic-bezier(0.22, 1, 0.36, 1);
 }
-.video-card { grid-template-columns: 128px minmax(0, 1fr); }
+.video-card { grid-template-columns: 160px minmax(0, 1fr); }
 .up-card { grid-template-columns: 64px minmax(0, 1fr); align-items: center; }
 .video-card:hover, .up-card:hover {
   border-color: color-mix(in srgb, var(--accent) 45%, var(--border));
@@ -995,7 +995,7 @@ a.live-chip:hover { border-color: color-mix(in srgb, var(--accent) 55%, var(--bo
   .up-stats { grid-template-columns: repeat(5, minmax(0, 1fr)); }
   .detail-hero { flex-direction: row; align-items: flex-start; }
   .hero-cover { width: 300px; flex: 0 0 300px; }
-  .video-card { grid-template-columns: 144px minmax(0, 1fr); }
+  .video-card { grid-template-columns: 160px minmax(0, 1fr); }
   .secondary-stat.optional { display: inline; }
 }
 
@@ -1048,7 +1048,7 @@ a.live-chip:hover { border-color: color-mix(in srgb, var(--accent) 55%, var(--bo
 }
 
 @media (max-width: 380px) {
-  .video-card { grid-template-columns: 104px minmax(0, 1fr); gap: 8px; }
+  .video-card { gap: 8px; }
   .metric-grid { grid-template-columns: repeat(2, minmax(0, 1fr)); }
   .candidate-card { grid-template-columns: 80px minmax(0, 1fr); }
   .candidate-thumb { width: 80px; }

reaction_board/web/static/views/about.js

@@ -50,6 +50,7 @@ export async function renderAbout(root) {
         <li><strong>播放 / 投币 / 收藏 / 评论</strong>：直接取当日快照；弹幕与转发不单列，已并入“互动”。</li>
         <li><strong>互动</strong>：点赞 + 投币 + 收藏。综合热度也使用全部七项指标，按配置权重加权。</li>
         <li><strong>互动率</strong>：(点赞 + 投币 + 收藏) / 播放；播放不足 100 不计算。</li>
+        <li><strong>播放增速</strong>：卡片上的百分比是增量 / 基线播放。基线播放不足 1000 不进增速榜。排序用增量 / (基线 + ${analytics.growth_rate_prior})。当日发布没有基线，不进增速榜。</li>
         <li><strong>综合热度</strong>：每项先 <code>log1p</code>，再按中位数与 IQR 做稳健标准化，最后加权求和。分数只用于同一批数据内排序，不能跨日期比较。</li>
         <li><strong>内容效率</strong>：播放 / 内容小时。内容小时 = 视频总时长（多 P 为各 P 之和）÷ 3600。榜单里的单条视频用收缩后的估计：<code>(播放 + κ·μ₀) / (小时 + κ)</code>，κ 默认 ${analytics.video_shrink_minutes} 分钟、μ₀ 为全库播放/小时；详情页保留原始播放/小时。</li>
         <li><strong>互动效率</strong>：互动 / 内容小时，互动口径与收缩方式同上。</li>
@@ -92,7 +93,7 @@ export async function renderAbout(root) {
         <li>B 站不公开完播率与单 P 播放，内容小时是“供给时长”，不是“实际观看时长”。</li>
         <li>多 P 可能让同一用户产生多次播放计数，本站不单独校正，属于已知偏差。</li>
         <li>粉丝数可能包含不活跃账号；粉丝快照暂不参与任何榜单。</li>
-        <li>样本不足（少于 2 天快照、单视频 UP）会显示提示，不伪造 0。</li>
+        <li>近 7/30 日增长仍要跨日快照。近 1 日：当日发布按 0 起算；更早发布的新入库按入库时播放量起算，需当天至少两次采样。样本不足时显示提示，不伪造 0。</li>
       </ul>
     </section>
 

reaction_board/web/static/views/home.js

@@ -69,7 +69,7 @@ export async function renderHome(root) {
   const growthEmptyTitle = growth.reason
     || `近 ${growthDays} 日增长暂不可用`;
   const growthEmptyHint = growthDays <= 1
-    ? "当前快照不足，完成至少 2 天快照后即可显示播放增量。"
+    ? "当日发布按 0 起算；更早发布的新入库需要当天至少两次采样。"
     : `需要至少 ${growthDays + 1} 天快照才能计算近 ${growthDays} 日增长。`;
   const updatedLabel = formatClock(overview.updated_at);
   const dataDateLabel = formatDataDate(overview.data_date);

reaction_board/web/static/views/rankings.js

@@ -166,6 +166,7 @@ function listHtml(data, scope, sort, startRank) {
     return videoCard(item, {
       rank: startRank + index, primary: primary.primary,
       primaryLabel: primary.label, primaryText: primary.text,
+      growthField: sort === "growth_rate" ? "rate" : "delta",
     });
   }).join("")}</div>`;
 }
@@ -179,6 +180,21 @@ function durationNote(data) {
   return `<div class="section-note">内容效率与互动效率：时长 ≥ ${filter.min_duration_seconds} 秒、发布满 ${filter.maturity_days} 天、播放 ≥ ${filter.min_view}${shrink}；时长缺失不参与。</div>`;
 }
 
+function growthNote(sort, growthDays, prior) {
+  if (sort === "growth_rate") {
+    const strength = Number(prior);
+    const smooth = Number.isFinite(strength)
+      ? `百分比是真实相对增量；排序按平滑增速（基线 + ${strength}）。`
+      : "百分比是真实相对增量；排序按平滑增速。";
+    const cold = Number(growthDays) === 1
+      ? "基线播放不足 1000、以及当日发布，不进增速榜。"
+      : "基线播放不足 1000 不进增速榜。";
+    return `<div class="section-note">${smooth}${cold}</div>`;
+  }
+  if (sort !== "growth" || Number(growthDays) !== 1) return "";
+  return `<div class="section-note">近 1 日：当日发布按 0 起算；更早发布的新入库按入库时播放量起算，需要当天至少两次采样。</div>`;
+}
+
 function bindControls(root) {
   root.querySelector("#rank-sort")?.addEventListener("change", (event) => {
     replaceQuery({ sort: event.target.value, page: null });
@@ -245,6 +261,7 @@ async function renderScope(root, { query }, scope) {
     </header>
     ${controls({ scope, sort, within, year, growthDays, q, years })}
     ${durationNote(data)}
+    ${growthNote(sort, growthDays, data.growth_rate_prior)}
     ${data.available === false ? `<div class="notice-row warning">
       <strong>${sort.startsWith("growth") ? "增长数据尚不可用" : "榜单筛选后为空"}</strong>
       <span>${escapeHtml(data.reason || "")}</span>

reaction_board/web/static/views/system.js

@@ -5,6 +5,15 @@ import {
   escapeHtml, formatCount, formatDateTime, formatNumber,
 } from "../format.js";
 
+function taskResultLabel(record) {
+  const summary = record.summary && typeof record.summary === "object" ? record.summary : {};
+  if (record.skipped || summary.skipped) {
+    return `<span class="muted">跳过</span>`;
+  }
+  if (record.ok) return "成功";
+  return `<span class="danger">失败</span>`;
+}
+
 function taskRows(tasks) {
   const entries = Object.entries(tasks || {});
   if (!entries.length) return emptyState("暂无任务记录", "运行一次同步、快照或发现后会写入最近结果。");
@@ -15,12 +24,16 @@ function taskRows(tasks) {
         ${entries.map(([name, record]) => {
           const summary = record.summary && typeof record.summary === "object"
             ? Object.entries(record.summary)
-              .filter(([key]) => ["added", "removed", "candidates", "notified", "ok", "failed", "total", "completed", "backup_error"].includes(key))
+              .filter(([key]) => [
+                "added", "removed", "candidates", "notified", "ok", "failed",
+                "total", "completed", "backup_error", "skipped", "already_completed",
+                "reason", "elapsed", "success",
+              ].includes(key))
               .map(([key, value]) => `${key}=${value}`).join(" ")
             : "";
           return `<tr>
             <td>${escapeHtml(name)}</td>
-            <td>${record.ok ? "成功" : `<span class="danger">失败</span>`}</td>
+            <td>${taskResultLabel(record)}</td>
             <td>${escapeHtml(formatDateTime(record.finished_at))}</td>
             <td class="summary-cell">${escapeHtml(summary || record.error || "—")}</td>
           </tr>`;
@@ -100,7 +113,7 @@ export async function renderSystem(root) {
         <div><span>新视频小时同步</span><strong>${running.snapshot_recent ? "运行中" : escapeHtml(formatDateTime(next.snapshot_recent))}</strong></div>
         <div><span>新视频搜索</span><strong>${running.discover || running.discovery ? "运行中" : escapeHtml(formatDateTime(next.discover))}</strong></div>
       </div>
-      <p class="section-note">进行中显示「运行中」；空闲时按该任务上次完成时间推算下次运行，不按墙钟抢跑。新视频小时同步与全站快照互斥。</p>
+      <p class="section-note">进行中显示「运行中」；空闲时按该任务上次完成时间推算下次运行。23:00–01:00（上海）为全站快照窗：发现 / 小时同步 / 候选清理会延后到 01:00 后；粉丝更新不避开。跳过或失败不会记成成功，会短间隔补跑。</p>
     </section>
 
     <section class="panel">

reaction_board/web/static/views/video.js

@@ -62,7 +62,7 @@ export async function renderVideo(root, { params, query }) {
           <span>${escapeHtml(formatDuration(detail.duration))}</span>
           ${detail.pages ? `<span>${formatNumber(detail.pages)} P</span>` : ""}
           ${badge(statusLabel(detail.status), `status-${detail.status}`)}
-          ${samples < 2 ? badge("样本不足", "warning") : ""}
+          ${samples < 2 && !detail.growth?.available ? badge("样本不足", "warning") : ""}
         </div>
         ${detail.blacklisted ? `<p class="section-note warning-text">该 UP 在黑名单中；黑名单只影响未来发现，不影响当前视频追踪。</p>` : ""}
         ${detail.up_follower_backfilled ? `<p class="section-note">UP 粉丝数为回填值，不是真实的入库时点。</p>` : ""}

tests/test_analytics.py

@@ -12,6 +12,7 @@ from reaction_board.analytics import (  # noqa: E402
     METRICS,
     age_days,
     compute_growth,
+    smoothed_growth_rate,
     content_hours,
     eb_rate,
     effective_shrink_minutes,
@@ -131,6 +132,18 @@ def test_compute_growth():
     no_gap = compute_growth(100, 90, 0)
     assert no_gap.available is False
 
+    same_day = compute_growth(100, 90, 0, allow_zero_gap=True)
+    assert same_day.available is True
+    assert same_day.delta == 10
+    assert same_day.per_day is None
+    assert same_day.gap_days == 0
+    assert abs(same_day.rate - 10 / 90) < 1e-12
+
+    from_zero_same_day = compute_growth(50, 0, 0, allow_zero_gap=True)
+    assert from_zero_same_day.delta == 50
+    assert from_zero_same_day.rate == 50.0
+    assert from_zero_same_day.per_day is None
+
     growth = compute_growth(1000, 800, 7)
     assert growth.available is True
     assert growth.delta == 200
@@ -149,6 +162,19 @@ def test_compute_growth():
     assert negative.per_day == -5.0
 
 
+def test_smoothed_growth_rate():
+    # 80→512：真实增速 5.4；prior=500 时排序分 432/580
+    assert abs(smoothed_growth_rate(432, 80, 500) - 432 / 580) < 1e-12
+    assert abs(smoothed_growth_rate(432, 5000, 500) - 432 / 5500) < 1e-12
+    assert abs(smoothed_growth_rate(432, 80, 1000) - 432 / 1080) < 1e-12
+    # prior=0 时排序分退回真实相对增速（基线 ≥ 1）
+    assert smoothed_growth_rate(432, 80, 0) == 432 / 80
+    assert smoothed_growth_rate(None, 80, 500) is None
+    assert smoothed_growth_rate(432, None, 500) is None
+    assert smoothed_growth_rate(10, 0, 0) is None
+    assert smoothed_growth_rate(10, 80, -1) is None
+
+
 def test_normalize_cover_and_hosts():
     assert normalize_cover(None) is None
     assert normalize_cover("") is None
@@ -239,6 +265,7 @@ def main_test():
     test_score_percentiles()
     test_interaction_rate_thresholds()
     test_compute_growth()
+    test_smoothed_growth_rate()
     test_normalize_cover_and_hosts()
     test_content_efficiency_helpers()
     test_age_and_maturity()

tests/test_config.py

@@ -146,6 +146,7 @@ def test_analytics_defaults_and_validation():
         assert cfg.analytics.video_shrink_minutes == 60
         assert cfg.analytics.shrink_minutes is None
         assert cfg.analytics.home_growth_days == "auto"
+        assert cfg.analytics.growth_rate_prior == 1000
         assert cfg.schedule.recent_snapshot_minutes == 60
         assert cfg.schedule.recent_video_days == 2
         assert cfg.schedule.snapshot_at == "23:15"
@@ -159,6 +160,7 @@ def test_analytics_defaults_and_validation():
             "video_shrink_minutes": 30,
             "shrink_minutes": 90,
             "home_growth_days": 1,
+            "growth_rate_prior": 2000,
         }
         data["schedule"] = {
             "discover_minutes": 30,
@@ -176,6 +178,7 @@ def test_analytics_defaults_and_validation():
         assert cfg.analytics.video_shrink_minutes == 30
         assert cfg.analytics.shrink_minutes == 90
         assert cfg.analytics.home_growth_days == 1
+        assert cfg.analytics.growth_rate_prior == 2000
         assert cfg.schedule.recent_snapshot_minutes == 15
         assert cfg.schedule.recent_video_days == 3
 
@@ -194,6 +197,7 @@ def test_analytics_defaults_and_validation():
             ({"home_growth_days": 3}, "home_growth_days"),
             ({"home_growth_days": "7"}, "home_growth_days"),
             ({"home_growth_days": 0}, "home_growth_days"),
+            ({"growth_rate_prior": -1}, "growth_rate_prior"),
         ):
             data = base_config()
             data["analytics"] = bad

tests/test_database.py

@@ -24,7 +24,7 @@ def test_write_read_transaction():
     with tempfile.TemporaryDirectory() as tmp:
         db = make_db(tmp)
         try:
-            assert db.schema_version == 7
+            assert db.schema_version == 8
             db.write(lambda c: c.execute("INSERT INTO kv(k, v) VALUES ('a', '1')"))
             assert db.read(lambda c: c.execute("SELECT v FROM kv WHERE k='a'").fetchone()[0]) == "1"
 
@@ -94,8 +94,8 @@ def test_close_and_reinitialize():
 
         db2 = make_db(tmp)
         try:
-            assert db2.schema_version == 7
-            assert db2.read(lambda c: c.execute("SELECT MAX(version) FROM schema_version").fetchone()[0]) == 7
+            assert db2.schema_version == 8
+            assert db2.read(lambda c: c.execute("SELECT MAX(version) FROM schema_version").fetchone()[0]) == 8
         finally:
             db2.close()
 
@@ -123,7 +123,7 @@ def test_migration_backup_on_existing_db():
         db = Database(path, backup_dir=Path(tmp) / "backup")
         try:
             applied = db.initialize()
-            assert applied == [1, 2, 3, 4, 5, 6, 7]
+            assert applied == [1, 2, 3, 4, 5, 6, 7, 8]
             assert len(list((Path(tmp) / "backup").glob("app-pre-v0-*.db"))) == 1
         finally:
             db.close()

tests/test_migrations.py

@@ -38,8 +38,8 @@ def test_fresh_and_idempotent():
         conn = sqlite3.connect(Path(tmp) / "app.db")
         try:
             applied = run_migrations(conn)
-            assert applied == [1, 2, 3, 4, 5, 6, 7]
-            assert current_version(conn) == 7
+            assert applied == [1, 2, 3, 4, 5, 6, 7, 8]
+            assert current_version(conn) == 8
             assert EXPECTED_TABLES <= table_names(conn)
             assert "schema_version" in table_names(conn)
             columns = {
@@ -70,10 +70,14 @@ def test_fresh_and_idempotent():
             assert {
                 "stat_date", "ip", "path", "hits", "first_at", "last_at",
             } <= visit_columns
+            snapshot_columns = {
+                row[1] for row in conn.execute("PRAGMA table_info(snapshot)")
+            }
+            assert {"view_open", "open_captured_at"} <= snapshot_columns
             # 再跑一次不应重复应用
             assert run_migrations(conn) == []
             row = conn.execute("SELECT COUNT(*) FROM schema_version").fetchone()
-            assert row[0] == 7
+            assert row[0] == 8
         finally:
             conn.close()
 
@@ -88,15 +92,15 @@ def test_backup_before_upgrade():
             conn.execute("INSERT INTO kv(k, v) VALUES ('hello', 'world')")
             conn.commit()
 
-            v8 = Migration(8, "add probe", ("CREATE TABLE probe (id INTEGER PRIMARY KEY)",))
+            v9 = Migration(9, "add probe", ("CREATE TABLE probe (id INTEGER PRIMARY KEY)",))
             applied = run_migrations(
                 conn,
                 backup_dir=backup_dir,
                 now=1_800_000_000,
-                migrations=MIGRATIONS + (v8,),
+                migrations=MIGRATIONS + (v9,),
             )
-            assert applied == [8]
-            backups = list(backup_dir.glob("app-pre-v7-*.db"))
+            assert applied == [9]
+            backups = list(backup_dir.glob("app-pre-v8-*.db"))
             assert len(backups) == 1
             check = sqlite3.connect(backups[0])
             try:
@@ -181,13 +185,46 @@ def test_v5_normalizes_unfaved():
             )
             conn.commit()
             applied = run_migrations(conn)
-            assert applied == [5, 6, 7]
+            assert applied == [5, 6, 7, 8]
             rows = dict(conn.execute("SELECT aid, status FROM video").fetchall())
             assert rows == {1: "active", 2: "dropped"}
         finally:
             conn.close()
 
 
+def test_v8_backfills_view_open():
+    with tempfile.TemporaryDirectory() as tmp:
+        conn = sqlite3.connect(Path(tmp) / "app.db")
+        try:
+            run_migrations(conn, migrations=MIGRATIONS[:7])
+            conn.execute(
+                "INSERT INTO video(aid, title, pubdate, status, first_seen) "
+                "VALUES (1, 't', 1, 'active', 1)"
+            )
+            conn.execute(
+                'INSERT INTO snapshot(stat_date, aid, "view", captured_at, ok) '
+                "VALUES ('2026-10-31', 1, 500, 100, 1)"
+            )
+            conn.execute(
+                'INSERT INTO snapshot(stat_date, aid, "view", captured_at, ok) '
+                "VALUES ('2026-10-30', 1, NULL, 90, 0)"
+            )
+            conn.commit()
+            assert run_migrations(conn) == [8]
+            ok_row = conn.execute(
+                "SELECT view_open, open_captured_at FROM snapshot "
+                "WHERE stat_date='2026-10-31'"
+            ).fetchone()
+            assert ok_row[0] == 500 and ok_row[1] == 100
+            fail_row = conn.execute(
+                "SELECT view_open, open_captured_at FROM snapshot "
+                "WHERE stat_date='2026-10-30'"
+            ).fetchone()
+            assert fail_row[0] is None and fail_row[1] is None
+        finally:
+            conn.close()
+
+
 def main():
     test_fresh_and_idempotent()
     test_backup_before_upgrade()
@@ -195,6 +232,7 @@ def main():
     test_prune_backups()
     test_validate_migration_order()
     test_v5_normalizes_unfaved()
+    test_v8_backfills_view_open()
     print("OK: SQLite 迁移、备份与回滚测试全部通过")
 
 

tests/test_repository.py

@@ -296,6 +296,59 @@ def test_pending_snapshot_aids():
             db.close()
 
 
+def test_upsert_snapshot_preserves_view_open():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = setup(tmp)
+        try:
+            metrics = {
+                "view": 10, "like": 1, "favorite": 0, "coin": 0,
+                "share": 0, "reply": 0, "danmaku": 0,
+            }
+            repo.upsert_snapshot(
+                "2026-10-31", 1, metrics=metrics, ok=True, err=None, captured_at=10,
+            )
+            first = repo.get_snapshot("2026-10-31", 1)
+            assert first["view"] == 10
+            assert first["view_open"] == 10
+            assert first["open_captured_at"] == 10
+
+            metrics["view"] = 40
+            repo.upsert_snapshot(
+                "2026-10-31", 1, metrics=metrics, ok=True, err=None, captured_at=20,
+            )
+            second = repo.get_snapshot("2026-10-31", 1)
+            assert second["view"] == 40
+            assert second["view_open"] == 10
+            assert second["open_captured_at"] == 10
+            assert second["captured_at"] == 20
+
+            repo.upsert_snapshot(
+                "2026-10-31", 2, metrics=None, ok=False, err="x", captured_at=1,
+            )
+            failed = repo.get_snapshot("2026-10-31", 2)
+            assert failed["view_open"] is None
+            metrics["view"] = 7
+            repo.upsert_snapshot(
+                "2026-10-31", 2, metrics=metrics, ok=True, err=None, captured_at=5,
+            )
+            recovered = repo.get_snapshot("2026-10-31", 2)
+            assert recovered["view"] == 7
+            assert recovered["view_open"] == 7
+            assert recovered["open_captured_at"] == 5
+
+            copied = repo.copy_ok_snapshots(
+                "2026-10-31", "2026-11-01", captured_at=30, src="seed",
+            )
+            assert copied == 2
+            seeded = repo.get_snapshot("2026-11-01", 1)
+            assert seeded["view"] == 40
+            assert seeded["view_open"] == 40
+            assert seeded["open_captured_at"] == 30
+            assert seeded["src"] == "seed"
+        finally:
+            db.close()
+
+
 def test_up_follower_snapshot_and_backfill():
     with tempfile.TemporaryDirectory() as tmp:
         db, repo = setup(tmp)
@@ -367,6 +420,7 @@ def main():
     test_primary_folder_is_order_independent()
     test_dropped_invalid_preserves_original_title()
     test_pending_snapshot_aids()
+    test_upsert_snapshot_preserves_view_open()
     test_up_follower_snapshot_and_backfill()
     print("OK: 收藏夹/视频状态机与黑名单测试全部通过")
 

tests/test_scheduler.py

@@ -8,7 +8,7 @@ import threading
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 sys.stdout.reconfigure(encoding="utf-8")
 
-from reaction_board.scheduler import ScheduledTask, Scheduler  # noqa: E402
+from reaction_board.scheduler import ScheduledTask, Scheduler, TaskControl  # noqa: E402
 
 
 def test_run_when_due_and_reschedule():
@@ -128,9 +128,111 @@ def test_next_run_fn_controls_reschedule():
     assert calls == [1_000.0, 1_123.0]
 
 
+
+def test_task_control_overrides_ok_and_next_run():
+    clock = [1_000.0]
+    task = ScheduledTask(
+        "defer",
+        60,
+        lambda: TaskControl(ok=False, next_run=1_500.0, error="quiet"),
+    )
+    scheduler = Scheduler([task], clock=lambda: clock[0])
+    results = scheduler.run_due()
+    assert len(results) == 1
+    assert results[0].ok is False
+    assert results[0].error == "quiet"
+    assert task.next_run == 1_500.0
+    # 未到 next_run 不重跑
+    clock[0] = 1_499.0
+    assert scheduler.run_due() == []
+    clock[0] = 1_500.0
+    task.fn = lambda: TaskControl(ok=True, next_run=2_000.0)
+    results = scheduler.run_due()
+    assert results[0].ok is True
+    assert task.next_run == 2_000.0
+
+
+
+
+class _CaptureLogger:
+    def __init__(self):
+        self.events = []
+
+    def info(self, event, **fields):
+        self.events.append(("info", event, fields))
+
+    def warning(self, event, **fields):
+        self.events.append(("warning", event, fields))
+
+    def error(self, event, **fields):
+        self.events.append(("error", event, fields))
+
+
+def test_task_control_ok_false_logs_fail_not_ok():
+    """TaskControl(ok=False) 不得记 scheduler.task_ok（Skip ≠ success）。"""
+    clock = [1_000.0]
+    logger = _CaptureLogger()
+    task = ScheduledTask(
+        "quiet",
+        60,
+        lambda: TaskControl(ok=False, next_run=1_500.0, error="snapshot_quiet_window"),
+    )
+    scheduler = Scheduler([task], clock=lambda: clock[0], logger=logger)
+    results = scheduler.run_due()
+    assert results[0].ok is False
+    assert results[0].error == "snapshot_quiet_window"
+    assert ("info", "scheduler.task_ok", {"task": "quiet"}) not in logger.events
+    fail_events = [e for e in logger.events if e[1] == "scheduler.task_fail"]
+    assert len(fail_events) == 1
+    assert fail_events[0][0] == "warning"
+    assert fail_events[0][2]["task"] == "quiet"
+    assert fail_events[0][2]["error"] == "snapshot_quiet_window"
+
+
+def test_task_control_ok_true_still_logs_task_ok():
+    clock = [1_000.0]
+    logger = _CaptureLogger()
+    task = ScheduledTask(
+        "ok_task",
+        60,
+        lambda: TaskControl(ok=True, next_run=1_200.0),
+    )
+    scheduler = Scheduler([task], clock=lambda: clock[0], logger=logger)
+    results = scheduler.run_due()
+    assert results[0].ok is True
+    assert ("info", "scheduler.task_ok", {"task": "ok_task"}) in logger.events
+    assert not any(e[1] == "scheduler.task_fail" for e in logger.events)
+
+
+def test_raised_exception_logs_task_fail():
+    clock = [100.0]
+    logger = _CaptureLogger()
+
+    def boom():
+        raise RuntimeError("db down")
+
+    scheduler = Scheduler(
+        [ScheduledTask("bad", 10, boom)],
+        clock=lambda: clock[0],
+        logger=logger,
+    )
+    results = scheduler.run_due()
+    assert results[0].ok is False
+    assert results[0].error == "db down"
+    assert not any(e[1] == "scheduler.task_ok" for e in logger.events)
+    assert not any(e[1] == "scheduler.task_fail" for e in logger.events)
+    err_events = [e for e in logger.events if e[1] == "scheduler.task_error"]
+    assert len(err_events) == 1
+    assert err_events[0][0] == "error"
+    assert err_events[0][2]["error"] == "db down"
+
 def main():
     test_run_when_due_and_reschedule()
     test_next_run_fn_controls_reschedule()
+    test_task_control_overrides_ok_and_next_run()
+    test_task_control_ok_false_logs_fail_not_ok()
+    test_task_control_ok_true_still_logs_task_ok()
+    test_raised_exception_logs_task_fail()
     test_task_error_does_not_block_other_tasks()
     test_run_forever_with_stop_event()
     test_max_iterations_and_validation()

tests/test_snapshot.py

@@ -444,19 +444,22 @@ def test_run_recent_videos_refreshes_without_completion_markers():
             1: view_for(1, view=111),
             2: view_for(2, view=222),
             3: view_for(3, view=333),
+            4: view_for(4, view=444),
         })
         db, repo, service = make_service(tmp, transport, track_fans=())
         try:
-            # 1/2 在墙钟近 2 天内；3 为旧片
+            # 1/2 在墙钟近 2 天内；3 为旧片；4 发布已久但当天入库，不进小时任务
             add_video(repo, 1)
             add_video(repo, 2)
             add_video(repo, 3)
+            add_video(repo, 4)
             repo.db.write(lambda conn: conn.executemany(
                 "UPDATE video SET pubdate = ? WHERE aid = ?",
                 [
                     (clock - 3600, 1),
                     (clock - 86400, 2),
                     (clock - 10 * 86400, 3),
+                    (clock - 10 * 86400, 4),
                 ],
             ))
             # 预置同日 ok=1，验证小时任务可覆盖刷新
@@ -477,23 +480,27 @@ def test_run_recent_videos_refreshes_without_completion_markers():
             )) == {1, 2}
 
             row = repo.db.read(lambda conn: dict(conn.execute(
-                "SELECT view, ok, captured_at FROM snapshot WHERE stat_date=? AND aid=1",
+                "SELECT view, ok, captured_at, view_open FROM snapshot "
+                "WHERE stat_date=? AND aid=1",
                 (DATE,),
             ).fetchone()))
             assert row["view"] == 111
             assert row["ok"] == 1
             assert row["captured_at"] == clock
+            assert row["view_open"] == 1
 
             # 完成日 / 备份标记保持原样
             assert repo.kv_get(service.completed_key(DATE)) is None
             assert repo.kv_get("snapshot:last_completed") is None
             assert repo.kv_get(service.backup_key(DATE)) is None
             assert transport.count("/web-interface/view") == 2
+            assert repo.get_snapshot(DATE, 4) is None
 
             # 全量任务随后才写完成日；且已 ok=1 的跳过
             full = service.run_videos(date=DATE)
             assert full.skipped >= 2  # 1、2 已被小时任务写过
-            assert full.success >= 1  # 至少旧片 3
+            assert full.success >= 1  # 至少旧片 3、4
+            assert repo.get_snapshot(DATE, 4) is not None
             assert repo.kv_get(service.completed_key(DATE)) == "1"
             assert repo.kv_get("snapshot:last_completed") == DATE
         finally:
@@ -530,6 +537,36 @@ def test_full_snapshot_skips_ok_rows_including_hourly_refresh():
 
 
 
+def test_seed_same_day_completed_is_already_completed_not_no_source():
+    """全站完成后 last_completed==target：应 already_completed，不误判 no_source。"""
+    clock = 1_800_000_000
+    with tempfile.TemporaryDirectory() as tmp:
+        transport = SnapshotTransport({1: view_for(1)})
+        db, repo, service = make_service(tmp, transport, track_fans=())
+        try:
+            add_video(repo, 1)
+            today = "2026-10-31"
+            repo.upsert_snapshot(
+                today, 1,
+                metrics={"view": 10, "like": 1, "coin": 0, "favorite": 0,
+                         "share": 0, "reply": 0, "danmaku": 0},
+                ok=True, err=None, captured_at=clock,
+            )
+            # 正常收尾态：last_completed 已推进到当日，且 completed 标记在位
+            repo.kv_set("snapshot:last_completed", today)
+            repo.kv_set(service.completed_key(today), "1")
+
+            summary = service.run_seed_from_last_completed(date=today)
+            assert summary.reason == "already_completed"
+            assert summary.already_completed is True
+            assert summary.completed is True
+            assert summary.reason != "no_source"
+            assert transport.count("/web-interface/view") == 0
+        finally:
+            db.close()
+
+
+
 def test_seed_from_last_completed_does_not_mark_completed():
     clock = 1_800_000_000
     with tempfile.TemporaryDirectory() as tmp:
@@ -562,6 +599,7 @@ def test_seed_from_last_completed_does_not_mark_completed():
             row = repo.get_snapshot(today, 1)
             assert row is not None
             assert int(row["view"]) == 10
+            assert int(row["view_open"]) == 10
             assert row["src"] == "seed"
             fan = repo.get_up_stat(today, 37754047)
             assert fan is not None and int(fan["follower"]) == 100
@@ -623,6 +661,117 @@ def test_seeded_day_skips_without_force_and_seals_with_force():
             db.close()
 
 
+def test_seal_keep_scraped_skips_view_progress_punches_seed():
+    """晚间 seal（keep_scraped）：已 view 的 ok=1 作进度跳过，seed 壳仍打穿。"""
+    clock = 1_800_000_000
+    with tempfile.TemporaryDirectory() as tmp:
+        transport = SnapshotTransport({
+            1: view_for(1, view=111),
+            2: view_for(2, view=222),
+        })
+        db, repo, service = make_service(tmp, transport)
+        try:
+            add_video(repo, 1)
+            add_video(repo, 2)
+            today = DATE
+            # 两条 seed 壳
+            for aid, view in ((1, 10), (2, 20)):
+                repo.upsert_snapshot(
+                    today, aid,
+                    metrics={"view": view, "like": 1, "coin": 0, "favorite": 0,
+                             "share": 0, "reply": 0, "danmaku": 0},
+                    ok=True, err=None, captured_at=clock, src="seed",
+                )
+            # 模拟 seal 中途：aid=1 已真实采集
+            repo.upsert_snapshot(
+                today, 1,
+                metrics={"view": 111, "like": 1, "coin": 0, "favorite": 0,
+                         "share": 0, "reply": 0, "danmaku": 0},
+                ok=True, err=None, captured_at=clock + 1, src="view",
+            )
+            repo.kv_set(service.seeded_key(today), "1")
+
+            sealed = service.run_if_needed(date=today, force=True)
+            assert sealed.completed is True
+            assert sealed.skipped >= 1  # aid=1 进度跳过
+            assert sealed.success >= 1  # aid=2 seed 打穿
+            # 只应请求 aid=2 的 view（aid=1 已是 view 进度）
+            assert transport.count("/web-interface/view") == 1
+            assert int(repo.get_snapshot(today, 1)["view"]) == 111
+            assert repo.get_snapshot(today, 1)["src"] == "view"
+            assert int(repo.get_snapshot(today, 2)["view"]) == 222
+            assert repo.get_snapshot(today, 2)["src"] == "view"
+        finally:
+            db.close()
+
+
+
+def test_force_keep_scraped_false_rescrapes_view_progress():
+    """C: CLI --force 路径 keep_scraped=False 须重采已有 view 进度行。"""
+    clock = 1_800_000_000
+    with tempfile.TemporaryDirectory() as tmp:
+        transport = SnapshotTransport({
+            1: view_for(1, view=999),
+            2: view_for(2, view=222),
+        })
+        db, repo, service = make_service(tmp, transport)
+        try:
+            add_video(repo, 1)
+            add_video(repo, 2)
+            today = DATE
+            for aid, view, src in ((1, 111, "view"), (2, 20, "seed")):
+                repo.upsert_snapshot(
+                    today, aid,
+                    metrics={"view": view, "like": 1, "coin": 0, "favorite": 0,
+                             "share": 0, "reply": 0, "danmaku": 0},
+                    ok=True, err=None, captured_at=clock, src=src,
+                )
+            repo.kv_set(service.seeded_key(today), "1")
+            forced = service.run_if_needed(
+                date=today, force=True, keep_scraped=False,
+            )
+            assert forced.completed is True
+            # 两条都重采
+            assert transport.count("/web-interface/view") == 2
+            assert int(repo.get_snapshot(today, 1)["view"]) == 999
+        finally:
+            db.close()
+
+
+def test_force_default_keep_scraped_skips_view_progress():
+    """C: force 默认 keep_scraped=True（seal）跳过已 view 行。"""
+    clock = 1_800_000_000
+    with tempfile.TemporaryDirectory() as tmp:
+        transport = SnapshotTransport({
+            1: view_for(1, view=999),
+            2: view_for(2, view=222),
+        })
+        db, repo, service = make_service(tmp, transport)
+        try:
+            add_video(repo, 1)
+            add_video(repo, 2)
+            today = DATE
+            repo.upsert_snapshot(
+                today, 1,
+                metrics={"view": 111, "like": 1, "coin": 0, "favorite": 0,
+                         "share": 0, "reply": 0, "danmaku": 0},
+                ok=True, err=None, captured_at=clock, src="view",
+            )
+            repo.upsert_snapshot(
+                today, 2,
+                metrics={"view": 20, "like": 1, "coin": 0, "favorite": 0,
+                         "share": 0, "reply": 0, "danmaku": 0},
+                ok=True, err=None, captured_at=clock, src="seed",
+            )
+            repo.kv_set(service.seeded_key(today), "1")
+            sealed = service.run_if_needed(date=today, force=True)
+            assert sealed.completed is True
+            assert transport.count("/web-interface/view") == 1
+            assert int(repo.get_snapshot(today, 1)["view"]) == 111
+        finally:
+            db.close()
+
+
 def main():
     test_snapshot_success_resume_and_force()
     test_resume_skips_ok_and_retries_failed()
@@ -638,8 +787,12 @@ def main():
     test_backup_failure_retries_without_api()
     test_run_recent_videos_refreshes_without_completion_markers()
     test_full_snapshot_skips_ok_rows_including_hourly_refresh()
+    test_seed_same_day_completed_is_already_completed_not_no_source()
     test_seed_from_last_completed_does_not_mark_completed()
     test_seeded_day_skips_without_force_and_seals_with_force()
+    test_seal_keep_scraped_skips_view_progress_punches_seed()
+    test_force_keep_scraped_false_rescrapes_view_progress()
+    test_force_default_keep_scraped_skips_view_progress()
     print("OK: 每日快照、断点续跑、失败占位与粉丝采集测试全部通过")
 
 

tests/test_task_state.py

@@ -0,0 +1,398 @@
+# -*- coding: utf-8 -*-
+"""task:state 持久化与启动续跑: uv run python tests/test_task_state.py"""
+
+import json
+import os
+import sys
+import tempfile
+from datetime import datetime
+from pathlib import Path
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+sys.stdout.reconfigure(encoding="utf-8")
+
+from reaction_board.database import Database  # noqa: E402
+from reaction_board.repository import Repository  # noqa: E402
+from reaction_board.scheduler import ScheduledTask, Scheduler  # noqa: E402
+from reaction_board.task_state import (  # noqa: E402
+    STATUS_DEFERRED,
+    STATUS_DONE,
+    STATUS_FAILED,
+    STATUS_PENDING,
+    STATUS_RUNNING,
+    bootstrap_from_last,
+    ensure_pending,
+    load_state,
+    mark_deferred,
+    mark_done,
+    mark_failed,
+    mark_running,
+    recover_stale_running,
+    resume_stat_date,
+    startup_next_run,
+)
+from reaction_board.timeutil import SHANGHAI_TZ, next_daily_run  # noqa: E402
+
+
+def _repo(tmp: str) -> tuple[Database, Repository]:
+    db = Database(Path(tmp) / "app.db")
+    db.initialize()
+    return db, Repository(db)
+
+
+def test_restart_mid_run_resumes_same_stat_date():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 23, 40, tzinfo=SHANGHAI_TZ).timestamp()
+            mark_running(
+                repo, "snapshot", stat_date="2026-11-03", now=now,
+                progress={"done": 10, "total": 100},
+            )
+            # 模拟进程崩溃：running 残留
+            recovered = recover_stale_running(repo, "snapshot", now=now + 60)
+            assert recovered is not None
+            assert recovered.status == STATUS_PENDING
+            assert recovered.stat_date == "2026-11-03"
+            assert recovered.progress == {"done": 10, "total": 100}
+            assert recovered.due_at <= now + 60
+            assert resume_stat_date(repo, "snapshot", fallback="2026-11-04") == "2026-11-03"
+            nxt = startup_next_run(
+                repo, "snapshot",
+                wall_due_at=next_daily_run(now + 60, "23:15"),
+                now=now + 60,
+            )
+            assert nxt <= now + 60
+        finally:
+            db.close()
+
+
+def test_seeded_before_snapshot_at_does_not_force_run():
+    """seed 后挂 pending；白天重启认落盘 due_at（刻意 ≠ 墙钟）且未到不跑。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 10, 30, tzinfo=SHANGHAI_TZ).timestamp()
+            # 刻意偏离墙钟 23:15，证明认落盘而非 fallback
+            slot = datetime(2026, 11, 3, 22, 0, tzinfo=SHANGHAI_TZ).timestamp()
+            wall = next_daily_run(now, "23:15")
+            assert slot != wall
+            ensure_pending(
+                repo, "snapshot", due_at=slot, stat_date="2026-11-03", now=now,
+            )
+            repo.kv_set("snapshot:seeded:2026-11-03", "1")
+            nxt = startup_next_run(
+                repo, "snapshot",
+                wall_due_at=wall,
+                now=now,
+            )
+            assert nxt == slot
+            assert nxt != wall
+            assert nxt > now
+            calls = []
+
+            def run_snapshot():
+                calls.append("ran")
+                return None
+
+            clock = [now]
+            sched = Scheduler(
+                [ScheduledTask("snapshot", 86400.0, run_snapshot, next_run=nxt)],
+                clock=lambda: clock[0],
+            )
+            assert sched.run_due() == []
+            assert calls == []
+        finally:
+            db.close()
+
+
+def test_failed_future_due_at_does_not_run_early():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 12, 0, tzinfo=SHANGHAI_TZ).timestamp()
+            due = now + 600
+            mark_failed(
+                repo, "snapshot", due_at=due, now=now,
+                stat_date="2026-11-03", error="boom",
+            )
+            nxt = startup_next_run(
+                repo, "snapshot",
+                wall_due_at=next_daily_run(now, "23:15"),
+                now=now,
+            )
+            assert nxt == due
+            clock = [now]
+            calls = []
+            sched = Scheduler(
+                [
+                    ScheduledTask(
+                        "snapshot", 86400.0, lambda: calls.append(1), next_run=nxt,
+                    )
+                ],
+                clock=lambda: clock[0],
+            )
+            assert sched.run_due() == []
+            assert calls == []
+        finally:
+            db.close()
+
+
+def test_due_at_passed_runs_including_failed_retry():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 12, 10, tzinfo=SHANGHAI_TZ).timestamp()
+            due = now - 30  # 已过重试点
+            mark_failed(
+                repo, "snapshot", due_at=due, now=due - 300,
+                stat_date="2026-11-03", error="boom",
+            )
+            nxt = startup_next_run(
+                repo, "snapshot",
+                wall_due_at=next_daily_run(now, "23:15"),
+                now=now,
+            )
+            # 不得退回今晚墙钟；应立刻可跑
+            assert nxt == due
+            assert nxt <= now
+            wall = next_daily_run(now, "23:15")
+            assert nxt != wall
+            calls = []
+            clock = [now]
+            sched = Scheduler(
+                [
+                    ScheduledTask(
+                        "snapshot", 86400.0, lambda: calls.append(1), next_run=nxt,
+                    )
+                ],
+                clock=lambda: clock[0],
+            )
+            results = sched.run_due()
+            assert len(results) == 1
+            assert calls == [1]
+        finally:
+            db.close()
+
+
+def test_done_writes_next_due_and_blocks_until_then():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 23, 40, tzinfo=SHANGHAI_TZ).timestamp()
+            next_due = next_daily_run(now, "23:15")
+            mark_done(repo, "snapshot", due_at=next_due, now=now, stat_date="2026-11-03")
+            state = load_state(repo, "snapshot")
+            assert state is not None
+            assert state.status == STATUS_DONE
+            nxt = startup_next_run(
+                repo, "snapshot", wall_due_at=next_due, now=now + 60,
+            )
+            assert nxt == next_due
+            assert nxt > now + 60
+        finally:
+            db.close()
+
+
+def test_no_state_falls_back_to_wall_clock():
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 10, 0, tzinfo=SHANGHAI_TZ).timestamp()
+            wall = next_daily_run(now, "23:15")
+            nxt = startup_next_run(repo, "snapshot", wall_due_at=wall, now=now)
+            assert nxt == wall
+        finally:
+            db.close()
+
+
+
+def test_fail_or_defer_preserves_progress():
+    """H7: mark_failed / mark_deferred 不传 progress 时保留已有 progress 与 stat_date。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 12, 0, tzinfo=SHANGHAI_TZ).timestamp()
+            progress = {"done": 42, "total": 100, "cursor": "aid:99"}
+            mark_running(
+                repo, "snapshot", stat_date="2026-11-03", now=now,
+                progress=progress,
+            )
+            before = load_state(repo, "snapshot")
+            assert before is not None
+            assert before.progress == progress
+            assert before.stat_date == "2026-11-03"
+
+            due_fail = now + 300
+            mark_failed(
+                repo, "snapshot", due_at=due_fail, now=now + 1,
+                progress=None, error="boom",
+            )
+            failed = load_state(repo, "snapshot")
+            assert failed is not None
+            assert failed.status == STATUS_FAILED
+            assert failed.due_at == due_fail
+            assert failed.progress == progress
+            assert failed.stat_date == "2026-11-03"
+            assert failed.error == "boom"
+
+            due_defer = now + 600
+            mark_deferred(
+                repo, "snapshot", due_at=due_defer, now=now + 2,
+                progress=None, error="snapshot_quiet_window",
+            )
+            deferred = load_state(repo, "snapshot")
+            assert deferred is not None
+            assert deferred.status == STATUS_DEFERRED
+            assert deferred.due_at == due_defer
+            assert deferred.progress == progress
+            assert deferred.stat_date == "2026-11-03"
+            assert deferred.error == "snapshot_quiet_window"
+        finally:
+            db.close()
+
+
+
+def test_bootstrap_from_last_fail_due_soon():
+    """A: 仅有 task:last 失败、无 state → 回填 failed 且 due_at 可立刻补跑。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 12, 0, tzinfo=SHANGHAI_TZ).timestamp()
+            wall = next_daily_run(now, "23:15")
+            repo.kv_set(
+                "task:last:snapshot",
+                json.dumps({
+                    "name": "snapshot",
+                    "ok": False,
+                    "skipped": False,
+                    "started_at": int(now - 600),
+                    "finished_at": int(now - 500),
+                    "error": "boom",
+                    "summary": {"stat_date": "2026-11-03"},
+                }, ensure_ascii=False),
+            )
+            assert load_state(repo, "snapshot") is None
+            boot = bootstrap_from_last(
+                repo, "snapshot", wall_due_at=wall, now=now, retry_seconds=300.0,
+            )
+            assert boot is not None
+            assert boot.status == STATUS_FAILED
+            assert boot.stat_date == "2026-11-03"
+            assert boot.due_at <= now
+            nxt = startup_next_run(
+                repo, "snapshot", wall_due_at=wall, now=now, slot_hhmm="23:15",
+            )
+            assert nxt <= now
+            assert nxt != wall
+        finally:
+            db.close()
+
+
+def test_ensure_pending_does_not_glue_yesterday_stat_date():
+    """B: 昨天未完成保留；不得把今晚 due 粘到昨天的 stat_date 上。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 0, 10, tzinfo=SHANGHAI_TZ).timestamp()
+            yday_due = now - 60
+            ensure_pending(
+                repo, "snapshot", due_at=yday_due,
+                stat_date="2026-11-02", now=now - 120,
+            )
+            tonight = datetime(2026, 11, 3, 23, 15, tzinfo=SHANGHAI_TZ).timestamp()
+            # seed 想挂今晚：不得冲掉昨天未完成
+            kept = ensure_pending(
+                repo, "snapshot", due_at=tonight,
+                stat_date="2026-11-03", now=now,
+            )
+            assert kept.stat_date == "2026-11-02"
+            assert kept.due_at == yday_due
+            assert resume_stat_date(
+                repo, "snapshot", fallback="2026-11-03",
+            ) == "2026-11-02"
+        finally:
+            db.close()
+
+
+def test_cross_midnight_mid_seal_resumes_yesterday():
+    """规则1 / B: 23:15 崩溃 → 00:10 重启续跑昨天，due 立刻可跑。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            crash = datetime(2026, 11, 2, 23, 40, tzinfo=SHANGHAI_TZ).timestamp()
+            mark_running(
+                repo, "snapshot", stat_date="2026-11-02", now=crash,
+                progress={"done": 3, "total": 10},
+            )
+            repo.kv_set("snapshot:seeded:2026-11-02", "1")
+            restart = datetime(2026, 11, 3, 0, 10, tzinfo=SHANGHAI_TZ).timestamp()
+            wall = next_daily_run(restart, "23:15")
+            nxt = startup_next_run(
+                repo, "snapshot", wall_due_at=wall, now=restart, slot_hhmm="23:15",
+            )
+            assert nxt <= restart
+            assert resume_stat_date(
+                repo, "snapshot", fallback="2026-11-03",
+            ) == "2026-11-02"
+            state = load_state(repo, "snapshot")
+            assert state is not None
+            assert state.status == STATUS_PENDING
+            assert state.progress == {"done": 3, "total": 10}
+        finally:
+            db.close()
+
+
+def test_scheduler_due_reader_picks_up_enqueued_seal():
+    """D: 同进程 seed 写盘 due_at=now 后，下一轮 run_due 须跑 seal。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, repo = _repo(tmp)
+        try:
+            now = datetime(2026, 11, 3, 23, 20, tzinfo=SHANGHAI_TZ).timestamp()
+            # snapshot 原 next_run 挂次日；盘上被 enqueue 成 now
+            tomorrow = next_daily_run(now, "23:15")
+            calls = []
+
+            def run_snap():
+                calls.append("snap")
+                return None
+
+            def due_reader(name: str):
+                st = load_state(repo, name)
+                return float(st.due_at) if st and st.due_at else None
+
+            clock = [now]
+            sched = Scheduler(
+                [ScheduledTask("snapshot", 86400.0, run_snap, next_run=tomorrow)],
+                clock=lambda: clock[0],
+                due_reader=due_reader,
+            )
+            assert sched.run_due() == []
+            assert calls == []
+            ensure_pending(
+                repo, "snapshot", due_at=now, stat_date="2026-11-03", now=now,
+            )
+            results = sched.run_due()
+            assert len(results) == 1
+            assert calls == ["snap"]
+        finally:
+            db.close()
+
+
+def main_test():
+    test_restart_mid_run_resumes_same_stat_date()
+    test_seeded_before_snapshot_at_does_not_force_run()
+    test_failed_future_due_at_does_not_run_early()
+    test_due_at_passed_runs_including_failed_retry()
+    test_done_writes_next_due_and_blocks_until_then()
+    test_no_state_falls_back_to_wall_clock()
+    test_fail_or_defer_preserves_progress()
+    test_bootstrap_from_last_fail_due_soon()
+    test_ensure_pending_does_not_glue_yesterday_stat_date()
+    test_cross_midnight_mid_seal_resumes_yesterday()
+    test_scheduler_due_reader_picks_up_enqueued_seal()
+    print("OK: task_state persistence / resume / due_at gates")
+
+
+if __name__ == "__main__":
+    main_test()

tests/test_timeutil.py

@@ -11,7 +11,10 @@ sys.stdout.reconfigure(encoding="utf-8")
 from reaction_board.timeutil import (  # noqa: E402
     SHANGHAI_TZ,
     day_bounds,
+    in_snapshot_quiet_window,
+    interval_next_run,
     is_date,
+    next_after_snapshot_quiet_window,
     next_daily_run,
     parse_hhmm,
     shanghai_date,
@@ -72,11 +75,40 @@ def test_day_bounds_and_date_validation():
         raise AssertionError("非法日期应失败")
 
 
+
+def test_snapshot_quiet_window():
+    assert in_snapshot_quiet_window(ts("2026-11-02 23:00:00"))
+    assert in_snapshot_quiet_window(ts("2026-11-02 23:59:59"))
+    assert in_snapshot_quiet_window(ts("2026-11-03 00:00:00"))
+    assert in_snapshot_quiet_window(ts("2026-11-03 00:59:59"))
+    assert not in_snapshot_quiet_window(ts("2026-11-03 01:00:00"))
+    assert not in_snapshot_quiet_window(ts("2026-11-02 22:59:59"))
+    assert not in_snapshot_quiet_window(ts("2026-11-02 12:00:00"))
+
+    assert next_after_snapshot_quiet_window(ts("2026-11-02 22:00:00")) == ts(
+        "2026-11-02 22:00:00"
+    )
+    assert next_after_snapshot_quiet_window(ts("2026-11-02 23:15:00")) == ts(
+        "2026-11-03 01:00:00"
+    )
+    assert next_after_snapshot_quiet_window(ts("2026-11-03 00:30:00")) == ts(
+        "2026-11-03 01:00:00"
+    )
+    # 22:50 + 60min = 23:50 → 推到次日 01:00
+    assert interval_next_run(ts("2026-11-02 22:50:00"), 3600) == ts(
+        "2026-11-03 01:00:00"
+    )
+    assert interval_next_run(ts("2026-11-02 12:00:00"), 3600) == ts(
+        "2026-11-02 13:00:00"
+    )
+
+
 def main():
     test_shanghai_date_boundary()
     test_parse_hhmm()
     test_next_daily_run()
     test_day_bounds_and_date_validation()
+    test_snapshot_quiet_window()
     print("OK: 上海时区与每日时间计算测试全部通过")
 
 

tests/test_web_queries.py

@@ -15,7 +15,7 @@ from reaction_board.repository import Repository  # noqa: E402
 from reaction_board.timeutil import day_bounds  # noqa: E402
 from reaction_board.web.cache import DASHBOARD_CACHE  # noqa: E402
 from reaction_board.web.queries import DashboardQueries  # noqa: E402
-from tests.web_fixtures import DATES, build_db, write_config  # noqa: E402
+from tests.web_fixtures import DATES, FOLDER, build_db, resource, write_config  # noqa: E402
 
 FIXED_NOW = 1_793_459_000  # 2026-10-31 前后
 
@@ -46,7 +46,7 @@ def test_overview():
             assert overview["counts"]["candidates_pending"] == 1
             assert overview["totals"]["view"] == 1200 + 5200 + 120
             assert overview["interactions"] > 0
-            assert overview["view_delta"] == (100 + 100 + 0)  # 101/102 的日增量
+            assert overview["view_delta"] == (100 + 100 + 120)  # 101/102 跨日 + 103 当日发布
             assert overview["snapshot_health"]["total"] == 3
             assert overview["folders"][0]["media_count"] == 3
             assert overview["folders"][0]["ratio"] < 1
@@ -484,7 +484,7 @@ def test_about_payload():
             assert about["snapshot_days"] == 3
             assert about["snapshot_first_date"] == DATES[0]
             assert about["fans_start_date"] == DATES[0]
-            assert about["schema_version"] == 7
+            assert about["schema_version"] == 8
             assert about["analytics"] == {
                 "min_duration_seconds": 600,
                 "maturity_days": 2,
@@ -492,6 +492,7 @@ def test_about_payload():
                 "video_shrink_minutes": 60,
                 "shrink_minutes": None,
                 "home_growth_days": "auto",
+                "growth_rate_prior": 1000,
             }
             assert about["heat_weights"]["view"] == 0.30
             assert about["followers"]["videos_total"] == 3
@@ -577,9 +578,13 @@ def test_growth_auto_falls_back_to_one_day():
             assert auto["available"] is True
             assert auto["growth_days"] == 1
             assert auto["days"] == 1
-            assert {item["aid"] for item in auto["items"]} == {101, 102}
+            assert {item["aid"] for item in auto["items"]} == {101, 102, 103}
             first = next(item for item in auto["items"] if item["aid"] == 101)
             assert first["growth"]["delta"] == 100  # 相邻日
+            assert first["growth"]["baseline"] == "snapshot"
+            today = next(item for item in auto["items"] if item["aid"] == 103)
+            assert today["growth"]["baseline"] == "zero"
+            assert today["growth"]["delta"] == 120
 
             forced7 = q.list_videos(sort="growth", growth_days=7)
             assert forced7["available"] is False
@@ -605,12 +610,15 @@ def test_growth_auto_falls_back_to_one_day():
 
 def test_snapshot_recent_next_run_follows_completion():
     import json
+    from datetime import datetime
+    from reaction_board.timeutil import SHANGHAI_TZ, interval_next_run
 
     with tempfile.TemporaryDirectory() as tmp:
         db, q = setup(tmp)
         try:
             repo = Repository(db)
-            finished = FIXED_NOW - 10
+            # 午后完成：间隔后仍在静默窗外，应严格 finished+3600
+            finished = int(datetime(2026, 10, 31, 15, 0, 0, tzinfo=SHANGHAI_TZ).timestamp())
             repo.kv_set(
                 "task:last:snapshot_recent",
                 json.dumps({
@@ -622,6 +630,128 @@ def test_snapshot_recent_next_run_follows_completion():
             DASHBOARD_CACHE.clear()
             system = q.system()
             assert system["next_runs"]["snapshot_recent"] == finished + 3600
+            assert system["next_runs"]["snapshot_recent"] == int(
+                interval_next_run(float(finished), 3600)
+            )
+        finally:
+            db.close()
+
+
+def test_snapshot_recent_next_run_defers_quiet_window_and_retry_on_skip():
+    import json
+    from datetime import datetime
+    from reaction_board.timeutil import SHANGHAI_TZ, next_after_snapshot_quiet_window
+
+    with tempfile.TemporaryDirectory() as tmp:
+        db, q = setup(tmp)
+        try:
+            repo = Repository(db)
+            # 22:50 成功 → 下一档 23:50 落入静默窗 → 推到 01:00
+            finished = int(datetime(2026, 10, 31, 22, 50, 0, tzinfo=SHANGHAI_TZ).timestamp())
+            repo.kv_set(
+                "task:last:snapshot_recent",
+                json.dumps({
+                    "name": "snapshot_recent", "ok": True,
+                    "started_at": finished - 5, "finished_at": finished,
+                    "error": None, "summary": {},
+                }),
+            )
+            DASHBOARD_CACHE.clear()
+            system = q.system()
+            expected = int(datetime(2026, 11, 1, 1, 0, 0, tzinfo=SHANGHAI_TZ).timestamp())
+            assert system["next_runs"]["snapshot_recent"] == expected
+
+            # skip 记录：补跑到窗后 01:00，不得当成功推进
+            quiet = int(datetime(2026, 10, 31, 23, 20, 0, tzinfo=SHANGHAI_TZ).timestamp())
+            repo.kv_set(
+                "task:last:snapshot_recent",
+                json.dumps({
+                    "name": "snapshot_recent", "ok": False, "skipped": True,
+                    "started_at": quiet, "finished_at": quiet,
+                    "error": "snapshot_quiet_window",
+                    "summary": {"skipped": True, "reason": "snapshot_quiet_window"},
+                }),
+            )
+            DASHBOARD_CACHE.clear()
+            system = q.system()
+            assert system["next_runs"]["snapshot_recent"] == int(
+                next_after_snapshot_quiet_window(float(quiet))
+            )
+            assert not any(e["task"] == "snapshot_recent" for e in system["errors"])
+        finally:
+            db.close()
+
+
+def test_snapshot_failed_next_run_retries_soon_not_next_day():
+    import json
+    from datetime import datetime
+    from reaction_board.timeutil import SHANGHAI_TZ, next_daily_run
+
+    with tempfile.TemporaryDirectory() as tmp:
+        db, q = setup(tmp)
+        try:
+            repo = Repository(db)
+            finished = int(datetime(2026, 10, 31, 23, 15, 0, tzinfo=SHANGHAI_TZ).timestamp())
+            repo.kv_set(
+                "task:last:snapshot",
+                json.dumps({
+                    "name": "snapshot", "ok": False,
+                    "started_at": finished - 1, "finished_at": finished,
+                    "error": "boom", "summary": {"completed": False},
+                }),
+            )
+            DASHBOARD_CACHE.clear()
+            system = q.system()
+            assert system["next_runs"]["snapshot"] == finished + 300
+            assert system["next_runs"]["snapshot"] != int(
+                next_daily_run(finished, "23:15")
+            )
+            assert any(e["task"] == "snapshot" for e in system["errors"])
+        finally:
+            db.close()
+
+
+def test_snapshot_seed_skip_next_run_retries_ten_minutes():
+    import json
+    from datetime import datetime
+    from reaction_board.timeutil import SHANGHAI_TZ, next_daily_run
+
+    with tempfile.TemporaryDirectory() as tmp:
+        db, q = setup(tmp)
+        try:
+            repo = Repository(db)
+            finished = int(datetime(2026, 11, 1, 0, 0, 0, tzinfo=SHANGHAI_TZ).timestamp())
+            repo.kv_set(
+                "task:last:snapshot_seed",
+                json.dumps({
+                    "name": "snapshot_seed", "ok": False, "skipped": True,
+                    "started_at": finished - 1, "finished_at": finished,
+                    "error": "snapshot_running",
+                    "summary": {"skipped": True, "reason": "snapshot_running"},
+                }),
+            )
+            DASHBOARD_CACHE.clear()
+            system = q.system()
+            assert system["next_runs"]["snapshot_seed"] == finished + 600
+            assert system["next_runs"]["snapshot_seed"] != int(
+                next_daily_run(finished, "00:00")
+            )
+            # skip 不算 errors
+            assert not any(e["task"] == "snapshot_seed" for e in system["errors"])
+
+            # no_source 同样 10 分钟
+            repo.kv_set(
+                "task:last:snapshot_seed",
+                json.dumps({
+                    "name": "snapshot_seed", "ok": False, "skipped": True,
+                    "started_at": finished - 1, "finished_at": finished,
+                    "error": "no_source",
+                    "summary": {"skipped": True, "reason": "no_source"},
+                }),
+            )
+            DASHBOARD_CACHE.clear()
+            system = q.system()
+            assert system["next_runs"]["snapshot_seed"] == finished + 600
         finally:
             db.close()
 
@@ -667,9 +797,10 @@ def test_overview_adjacent_previous_with_seed_and_dual_fields():
                 + overview["counts"]["unfaved"]
                 + overview["counts"]["removed"]
             )
-            # 当前=seed(=完成日数值)，相邻 previous=完成日本身 → 增量 0
+            # 101/102：当前=seed(=完成日数值)，相邻 previous=完成日本身 → 增量 0
+            # 103：当日发布，近 1 日相对 0，计入当前播放
             assert overview["view_delta_available"] is True
-            assert overview["view_delta"] == 0
+            assert overview["view_delta"] == 120
 
             growth = q.list_videos(sort="growth", growth_days=1)
             assert growth["available"] is True
@@ -678,6 +809,141 @@ def test_overview_adjacent_previous_with_seed_and_dual_fields():
                     assert item["growth"]["available"] is True
                     assert item["growth"]["delta"] == 0
                     assert item["growth"]["gap_days"] == 1
+                    assert item["growth"]["baseline"] == "snapshot"
+        finally:
+            db.close()
+
+
+def _view_metrics(view: int) -> dict[str, int]:
+    return {
+        "view": view, "like": 0, "coin": 0, "favorite": 0,
+        "share": 0, "reply": 0, "danmaku": 0,
+    }
+
+
+def test_same_day_growth_baselines():
+    """近 1 日：当日发布相对 0；更早发布用 view_open；7 日不用这两条退路。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, q = setup(tmp)
+        try:
+            repo = Repository(db)
+            today = DATES[-1]
+            start = int(day_bounds(today)[0])
+            old_pub = int(day_bounds("2026-09-01")[0])
+
+            repo.upsert_snapshot(
+                today, 103, metrics=_view_metrics(200),
+                ok=True, err=None, captured_at=start + 3600,
+            )
+            repo.upsert_from_fav(
+                resource(105, up_mid=1, up_name="UP A", pubdate=old_pub),
+                FOLDER.media_id, was_added=True, now=start + 10,
+            )
+            repo.upsert_snapshot(
+                today, 105, metrics=_view_metrics(500),
+                ok=True, err=None, captured_at=start + 10,
+            )
+            repo.upsert_snapshot(
+                today, 105, metrics=_view_metrics(800),
+                ok=True, err=None, captured_at=start + 20,
+            )
+            repo.upsert_from_fav(
+                resource(106, up_mid=1, up_name="UP A", pubdate=old_pub),
+                FOLDER.media_id, was_added=True, now=start + 30,
+            )
+            repo.upsert_snapshot(
+                today, 106, metrics=_view_metrics(900),
+                ok=True, err=None, captured_at=start + 30,
+            )
+            repo.upsert_from_fav(
+                resource(107, up_mid=1, up_name="UP A", pubdate=old_pub),
+                FOLDER.media_id, was_added=True, now=start + 40,
+            )
+            repo.upsert_snapshot(
+                today, 107, metrics=_view_metrics(50),
+                ok=True, err=None, captured_at=start + 40,
+            )
+            db.write(lambda conn: conn.execute(
+                "UPDATE video SET pubdate = 0 WHERE aid = 107"
+            ))
+            DASHBOARD_CACHE.clear()
+
+            one = q.list_videos(sort="growth", growth_days=1, page_size=20)
+            by_aid = {item["aid"]: item for item in one["items"]}
+            assert 103 in by_aid and 105 in by_aid
+            assert 106 not in by_aid and 107 not in by_aid
+            assert by_aid[103]["growth"]["baseline"] == "zero"
+            assert by_aid[103]["growth"]["delta"] == 200
+            assert by_aid[103]["growth"]["rate"] is None
+            assert by_aid[103]["growth"]["gap_days"] == 0
+            assert by_aid[105]["growth"]["baseline"] == "open"
+            assert by_aid[105]["growth"]["delta"] == 300
+            assert by_aid[101]["growth"]["baseline"] == "snapshot"
+            assert by_aid[101]["growth"]["delta"] == 100
+
+            rate = q.list_videos(sort="growth_rate", growth_days=1, page_size=20)
+            rate_aids = {item["aid"] for item in rate["items"]}
+            assert 103 not in rate_aids and 105 not in rate_aids
+            assert 101 in rate_aids
+
+            seven = q.list_videos(sort="growth", growth_days=7)
+            seven_aids = {item["aid"] for item in seven["items"]}
+            assert 103 not in seven_aids
+            assert 105 not in seven_aids
+
+            detail103 = q.video_detail(103)
+            assert detail103["growth"]["baseline"] == "zero"
+            assert detail103["growth"]["delta"] == 200
+            detail105 = q.video_detail(105)
+            assert detail105["growth"]["baseline"] == "open"
+            assert detail105["growth"]["delta"] == 300
+            assert q.video_detail(106)["growth"]["available"] is False
+            assert q.video_detail(107)["growth"]["available"] is False
+
+            overview = q.overview()
+            assert overview["view_delta"] == 100 + 100 + 200 + 300
+        finally:
+            db.close()
+
+
+def test_growth_rate_sort_uses_prior():
+    """基线不足 1000 不进增速榜；达标后按基线 + prior 排序，不按 raw rate。"""
+    with tempfile.TemporaryDirectory() as tmp:
+        db, q = setup(tmp)
+        try:
+            repo = Repository(db)
+            latest = DATES[-1]
+            previous_day = DATES[-2]
+            start = int(day_bounds(latest)[0])
+            pub = int(day_bounds("2026-09-01")[0])
+            # 301：80→512，真实 +540%，基线低于 1000，只留在增量榜。
+            # 304：1000→2500，真实增速高于 302，平滑后应排在 302 后面。
+            for aid, views in ((301, (80, 512)), (302, (5000, 10000)), (304, (1000, 2500))):
+                repo.upsert_from_fav(
+                    resource(aid, up_mid=1, up_name="UP A", pubdate=pub),
+                    FOLDER.media_id, was_added=True, now=start,
+                )
+                repo.upsert_snapshot(
+                    previous_day, aid, metrics=_view_metrics(views[0]),
+                    ok=True, err=None, captured_at=start - 86400,
+                )
+                repo.upsert_snapshot(
+                    latest, aid, metrics=_view_metrics(views[1]),
+                    ok=True, err=None, captured_at=start + 60,
+                )
+            DASHBOARD_CACHE.clear()
+            ranked = q.list_videos(sort="growth_rate", growth_days=1, page_size=20)
+            order = [item["aid"] for item in ranked["items"]]
+            assert ranked["growth_rate_prior"] == 1000
+            assert 301 not in order
+            assert order.index(302) < order.index(304)
+            modest = next(item for item in ranked["items"] if item["aid"] == 304)
+            assert abs(modest["growth"]["rate"] - 1500 / 1000) < 1e-12
+
+            growth = q.list_videos(sort="growth", growth_days=1, page_size=20)
+            grown = next(item for item in growth["items"] if item["aid"] == 301)
+            assert grown["growth"]["delta"] == 432
+            assert abs(grown["growth"]["rate"] - 432 / 80) < 1e-12
         finally:
             db.close()
 
@@ -741,7 +1007,12 @@ def main_test():
     test_system_tasks_sorted_by_started_at_desc()
     test_growth_auto_falls_back_to_one_day()
     test_snapshot_recent_next_run_follows_completion()
+    test_snapshot_recent_next_run_defers_quiet_window_and_retry_on_skip()
+    test_snapshot_failed_next_run_retries_soon_not_next_day()
+    test_snapshot_seed_skip_next_run_retries_ten_minutes()
     test_overview_adjacent_previous_with_seed_and_dual_fields()
+    test_same_day_growth_baselines()
+    test_growth_rate_sort_uses_prior()
     print("OK: Web 查询层总览/榜单/详情/UP/时长效率/粉丝/候选/系统/访问统计测试全部通过")
 
 
@@ -805,13 +1076,15 @@ def test_system_status_uses_completed_date_and_running_flags():
 
 def test_system_next_runs_follow_last_completion_not_wall_clock():
     import json
+    from datetime import datetime
+    from reaction_board.timeutil import SHANGHAI_TZ
 
     with tempfile.TemporaryDirectory() as tmp:
         db, q = setup(tmp)
         try:
             repo = Repository(db)
-            # 发现任务刚完成不久：下次 = finished + 60min，即使 now 已超过也不跳档
-            finished = FIXED_NOW - 30
+            # 午后完成：下次 = finished + 60min，不跳档、不误入静默窗推算
+            finished = int(datetime(2026, 10, 31, 15, 0, 0, tzinfo=SHANGHAI_TZ).timestamp())
             repo.kv_set(
                 "task:last:discovery",
                 json.dumps({
@@ -823,7 +1096,6 @@ def test_system_next_runs_follow_last_completion_not_wall_clock():
             DASHBOARD_CACHE.clear()
             system = q.system()
             assert system["next_runs"]["discover"] == finished + 3600
-            # 仍小于/等于 now 也可以（表示本轮后的下一档已到期或未到），但绝不能跳到 +7200
             assert system["next_runs"]["discover"] != finished + 7200
         finally:
             db.close()

tests/test_web_server.py

@@ -75,7 +75,7 @@ def test_readonly_routes_and_static():
             assert status == 200 and payload["ok"] is True
             status, _, payload = running.json("/api/about")
             assert status == 200
-            assert payload["data"]["schema_version"] == 7
+            assert payload["data"]["schema_version"] == 8
             assert payload["data"]["analytics"]["min_duration_seconds"] == 600
 
             status, _, overview = running.json("/api/overview")

tests/web_fixtures.py

@@ -138,7 +138,7 @@ def build_db(tmp: str | Path) -> Path:
                 day, 37754047, follower=1000 + index * 10, following=5,
                 ok=True, captured_at=ts(day) + 60,
             )
-        # 103 只有最后一天快照，用于验证样本不足；104 另加一条失败快照
+        # 103 只有最后一天快照（发布日=快照日，近 1 日相对 0）；104 另加一条失败快照
         repo.upsert_snapshot(
             DATES[-1], 103,
             metrics={"view": 120, "like": 5, "coin": 1, "favorite": 2, "share": 0, "reply": 1, "danmaku": 1},

tools/seed_preview_db.py

@@ -237,8 +237,8 @@ def _insert_snapshot(
         """
         INSERT OR REPLACE INTO snapshot(
           stat_date, aid, "view", "like", favorite, coin, share, reply,
-          danmaku, captured_at, ok, err, src
-        ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 'preview')
+          danmaku, captured_at, ok, err, src, view_open, open_captured_at
+        ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, 'preview', ?, ?)
         """,
         (
             stat_date,
@@ -253,6 +253,8 @@ def _insert_snapshot(
             int(captured_at),
             1 if ok else 0,
             err,
+            values.get("view") if ok else None,
+            int(captured_at) if ok else None,
         ),
     )