fix: 将未达起送归为比价成功

## 需求背景
将比价、短信与登录、广告、引导与账号、风控免告警等限制统一配置，并支持按手机号或设备设置有有效期的临时白名单。

## 主要改动
- 新增统一限制策略注册表、全局 JSON 配置与白名单覆盖表
- 新增白名单管理、设备检索、批量追加与主体统一编辑接口
- 接入比价、短信登录、广告奖励、引导视频、账号换绑及风险告警调用链
- 保留旧配置接口兼容，并同步统一策略全局值
- 增加单主体唯一有效期、恢复全局、审计日志和风险事件自动处理
- 增加数据库迁移及完整回归测试

## 验证
- 白名单、权限、配置及风控测试 50 项通过
- 短信、登录、比价、广告关联测试 98 项通过
- Ruff 与 Python 编译检查通过
- Alembic 保持单一 head
- 已同步最新 main

---------

Co-authored-by: unknown <798648091@qq.com>
Reviewed-on: #207
Co-authored-by: linkeyu <linkeyu@wonderable.ai>
Co-committed-by: linkeyu <linkeyu@wonderable.ai>

.env.example

@@ -45,6 +45,9 @@ SMS_SEND_INTERVAL_SEC=60
 # chuanglan:创蓝云智(253)模板短信,本服务生成码、创蓝只下发、本地校验(Mode B,与极光同);缺凭证 503。
 #            用 YZM 前缀验证码账号;服务器出网 IP 需在创蓝控制台加白名单(否则 117)。见 docs/integrations/chuanglan/tpl-send.md。
 SMS_PROVIDER=jiguang
+# 极光→创蓝 容灾:主 provider(SMS_PROVIDER)供应商不可用(503)时转投的备。
+# 空=关闭 fallback(默认);生产设 chuanglan 开启极光主、创蓝备;置空即秒回退。
+SMS_FALLBACK_PROVIDER=
 ALIYUN_SMS_ACCESS_KEY_ID=
 ALIYUN_SMS_ACCESS_KEY_SECRET=
 ALIYUN_SMS_SIGN_NAME=
@@ -69,8 +72,10 @@ CHUANGLAN_SMS_TIMEOUT_SEC=10
 # 配一个固定测试手机号,专供无 SIM 卡 / 不走一键登录时打通全流程:该号登录【免短信验证码】
 # (real 模式下也跳过校验)、每次登录【都重走新手引导】,并有【每日登录上限】防被人猜到号后脚本刷。
 # 逻辑见 app/core/test_account.py,与其他业务解耦。
-# ⚠️ 留空 = 关闭整功能(生产默认);要启用才填号(如 11111111111)。改完重启生效,随时可清空停用。
+# ⚠️ 两项都留空 = 关闭整功能(生产默认)。多账号用英文逗号分隔,改完重启生效。
 TEST_ACCOUNT_PHONE=
+# 推荐新配置；例如联调环境填 11111111111,22222222222。
+TEST_ACCOUNT_PHONES=
 # 该测试号每日最多登录次数,当日超过即拒绝(429),次日归零。
 TEST_ACCOUNT_DAILY_LIMIT=500
 
@@ -175,3 +180,16 @@ PANGLE_REPORT_SITE_ID_TEST=5832303
 # APPLOG_MAX_BATCH=500                      # 单批最大条数(超 → 422;导入期常量,改需重启)
 # APPLOG_MAX_BODY_BYTES=2097152            # 请求体上限 2MB(超 → 413;运行期可调)
 # APPLOG_MAX_MSG_BYTES=8192                # 单条 msg 超此字节数截断
+
+# ===== 荣耀 Push =====
+HONOR_PUSH_APP_ID=
+HONOR_PUSH_CLIENT_ID=
+HONOR_PUSH_CLIENT_SECRET=
+# 0=正式消息(默认)；1=测试消息(仅开发联调，生产必须保持 0)
+HONOR_PUSH_TARGET_USER_TYPE=0
+
+# ===== 华为 Push =====
+HUAWEI_PUSH_APP_ID=
+HUAWEI_PUSH_APP_SECRET=
+# 0=正式消息(默认)；1=测试消息(仅开发联调，生产必须保持 0)
+HUAWEI_PUSH_TARGET_USER_TYPE=0

alembic/versions/6d2309208549_merge_comparison_platforms_fail_reason_.py

@@ -0,0 +1,26 @@
+"""merge comparison platforms + fail_reason heads
+
+Revision ID: 6d2309208549
+Revises: comparison_platforms_col, comparison_record_fail_reason
+Create Date: 2026-07-29 01:48:41.868083
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '6d2309208549'
+down_revision: Union[str, Sequence[str], None] = ('comparison_platforms_col', 'comparison_record_fail_reason')
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    pass
+
+
+def downgrade() -> None:
+    pass

alembic/versions/comparison_below_minimum_as_success.py

@@ -0,0 +1,56 @@
+"""count completed below-minimum comparisons as success
+
+Revision ID: comparison_below_min_success
+Revises: limit_policy_global_bundle
+"""
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+from alembic import op
+
+revision: str = "comparison_below_min_success"
+down_revision: str | Sequence[str] | None = "limit_policy_global_bundle"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_JSON = sa.JSON().with_variant(postgresql.JSONB(), "postgresql")
+
+
+def _comparison_record() -> sa.TableClause:
+    return sa.table(
+        "comparison_record",
+        sa.column("status", sa.String(16)),
+        sa.column("fail_reason", sa.String(256)),
+        sa.column("raw_payload", _JSON),
+    )
+
+
+def upgrade() -> None:
+    comparison_record = _comparison_record()
+    op.execute(
+        comparison_record.update()
+        .where(comparison_record.c.status == "below_minimum")
+        .values(status="success", fail_reason=None)
+    )
+
+
+def downgrade() -> None:
+    comparison_record = _comparison_record()
+    # The write paths deliberately preserve the granular outcome. Restore only
+    # rows that this change normalized, without touching ordinary successes.
+    raw_outcome = sa.func.coalesce(
+        comparison_record.c.raw_payload["record_status"].as_string(),
+        comparison_record.c.raw_payload["status"].as_string(),
+    )
+    op.execute(
+        comparison_record.update()
+        .where(
+            comparison_record.c.status == "success",
+            raw_outcome == "below_minimum",
+        )
+        .values(status="below_minimum")
+    )

alembic/versions/comparison_platforms_col.py

@@ -0,0 +1,47 @@
+"""add platforms unified array column to comparison_record
+
+展示模型统一数组(pricebot done.params.platforms 原样存): 每平台一行、自带
+status/is_best/display, 记录页据此直接渲染, 不再靠 comparison_results + 客户端合并 + 前端派生。
+纯新增列, 老记录为空 → 前端回退老 comparison_results。
+
+Revision ID: comparison_platforms_col
+Revises: user_manual_risk_fields
+"""
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+from alembic import op
+
+revision: str = "comparison_platforms_col"
+down_revision: str | None = "user_manual_risk_fields"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_JSON = sa.JSON().with_variant(postgresql.JSONB(), "postgresql")
+
+
+def upgrade() -> None:
+    # 幂等: 线上为了提前给历史数据补 platforms(2026-07-29), 已手动
+    # `ALTER TABLE comparison_record ADD COLUMN IF NOT EXISTS platforms jsonb
+    #  NOT NULL DEFAULT '[]'::jsonb`(与本 migration 定义一致)。列已存在时跳过,
+    # 否则上线 alembic upgrade head 会撞 DuplicateColumn 直接部署失败。
+    bind = op.get_bind()
+    cols = {c["name"] for c in sa.inspect(bind).get_columns("comparison_record")}
+    if "platforms" in cols:
+        return
+    with op.batch_alter_table("comparison_record") as batch_op:
+        batch_op.add_column(
+            sa.Column(
+                "platforms", _JSON, nullable=False,
+                server_default=sa.text("'[]'"),
+            )
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("comparison_record") as batch_op:
+        batch_op.drop_column("platforms")

alembic/versions/guide_video_scene_unique.py

@@ -0,0 +1,31 @@
+"""guide video play count is independent for coupon and comparison
+
+Revision ID: guide_video_scene_unique
+Revises: 6d2309208549
+"""
+from alembic import op
+
+revision = "guide_video_scene_unique"
+down_revision = "6d2309208549"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.drop_index("uq_guide_video_play_user_seq", table_name="guide_video_play")
+    op.create_index(
+        "uq_guide_video_play_user_scene_seq",
+        "guide_video_play",
+        ["user_id", "scene", "seq"],
+        unique=True,
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("uq_guide_video_play_user_scene_seq", table_name="guide_video_play")
+    op.create_index(
+        "uq_guide_video_play_user_seq",
+        "guide_video_play",
+        ["user_id", "seq"],
+        unique=True,
+    )

alembic/versions/guide_video_ten_circle_v2.py

@@ -0,0 +1,164 @@
+"""guide video prepare/start/ten-circle reward state machine
+
+Revision ID: guide_video_ten_circle_v2
+Revises: guide_video_scene_unique
+"""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "guide_video_ten_circle_v2"
+down_revision = "guide_video_scene_unique"
+branch_labels = None
+depends_on = None
+
+
+def _backfill_configs(connection) -> None:
+    from app.core import media
+    from app.core.config import settings
+
+    for key in ("coupon_guide_video", "comparison_guide_video"):
+        row = connection.execute(
+            sa.text("SELECT value FROM app_config WHERE key = :key"), {"key": key}
+        ).first()
+        if not row:
+            continue
+        raw = row[0]
+        value = json.loads(raw) if isinstance(raw, str) else dict(raw or {})
+        if value.get("guide_video_v2_migrated") is True:
+            continue
+        maximum = int(value.get("max_plays", 3) or 0)
+        reward = int(value.get("reward_coin", 100) or 0)
+        video_url = str(value.get("video_url") or "")
+        error = None
+        analysis = None
+        if not 1 <= maximum <= 50:
+            error = "旧播放次数不符合 1～50 约束"
+        elif not 10 <= reward <= 10_000 or reward % 10:
+            error = "旧金币总价不符合 10～10000 且为 10 的倍数约束"
+        elif not video_url:
+            error = "尚未上传视频"
+        else:
+            prefix = f"{settings.MEDIA_URL_PREFIX}/guide_video/"
+            if not video_url.startswith(prefix):
+                error = "旧视频不是服务端托管文件，请重新上传"
+            else:
+                path = Path(settings.MEDIA_ROOT) / "guide_video" / video_url[len(prefix):]
+                if not path.is_file():
+                    error = "旧视频文件缺失，请重新上传"
+                else:
+                    try:
+                        analysis = media._probe_guide_video(path)
+                    except media.MediaError as exc:
+                        error = str(exc)
+        value["config_version"] = int(value.get("config_version", 0) or 0) + 1
+        value["guide_video_v2_migrated"] = True
+        if analysis:
+            value.update(analysis)
+        else:
+            value.update(
+                enabled=False,
+                duration_ms=None,
+                video_codec=None,
+                audio_codec=None,
+                analysis_status="invalid" if video_url else "missing",
+                analysis_error=error,
+            )
+        statement = sa.text(
+            "UPDATE app_config SET value = :value WHERE key = :key"
+        ).bindparams(sa.bindparam("value", type_=sa.JSON()))
+        connection.execute(statement, {"key": key, "value": value})
+
+
+def _seal_legacy_plays(connection) -> None:
+    """旧整笔发奖 token 永久封口；旧未发奖会话终止但仍保留计次行。"""
+    connection.execute(
+        sa.text(
+            """
+            UPDATE guide_video_play
+            SET status = CASE WHEN status = 'granted' THEN 'legacy_completed' ELSE 'legacy_closed' END,
+                settled_circles = CASE WHEN status = 'granted' THEN 10 ELSE 0 END,
+                prepared_at = started_at,
+                expires_at = started_at
+            WHERE status IN ('granted', 'playing')
+            """
+        )
+    )
+
+
+def _prepare_downgrade(connection) -> None:
+    """清除未起播计划，并把所有保留 token 封成旧版不可再领奖的 granted。"""
+    connection.execute(
+        sa.text("DELETE FROM guide_video_play WHERE status = 'prepared'")
+    )
+    connection.execute(
+        sa.text(
+            """
+            UPDATE guide_video_play
+            SET status = 'granted',
+                completed = CASE WHEN settled_circles >= 10 THEN 1 ELSE completed END,
+                granted_at = CASE
+                    WHEN settled_circles >= 10 AND granted_at IS NULL THEN started_at
+                    ELSE granted_at
+                END
+            WHERE status IN (
+                'started', 'completed', 'legacy_completed', 'legacy_closed'
+            )
+            """
+        )
+    )
+    # 防御未知/中间状态：宁可删除，也绝不让旧应用把它当 playing 整笔发奖。
+    connection.execute(
+        sa.text(
+            """
+            DELETE FROM guide_video_play
+            WHERE seq IS NULL OR started_at IS NULL OR status != 'granted'
+            """
+        )
+    )
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("guide_video_play") as batch:
+        batch.alter_column("seq", existing_type=sa.Integer(), nullable=True)
+        batch.alter_column("started_at", existing_type=sa.DateTime(timezone=True), nullable=True)
+        batch.add_column(sa.Column("duration_ms", sa.Integer(), nullable=False, server_default="0"))
+        batch.add_column(sa.Column("config_version", sa.Integer(), nullable=False, server_default="0"))
+        batch.add_column(sa.Column("settled_circles", sa.Integer(), nullable=False, server_default="0"))
+        batch.add_column(
+            sa.Column("prepared_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now())
+        )
+        batch.add_column(sa.Column("expires_at", sa.DateTime(timezone=True), nullable=True))
+        batch.alter_column("status", existing_type=sa.String(length=16), type_=sa.String(length=24))
+    op.create_index("ix_guide_video_play_expires_at", "guide_video_play", ["expires_at"])
+    op.create_index(
+        "ux_coin_transaction_guide_video_ref",
+        "coin_transaction",
+        ["user_id", "biz_type", "ref_id"],
+        unique=True,
+        sqlite_where=sa.text("biz_type = 'guide_video' AND ref_id IS NOT NULL"),
+        postgresql_where=sa.text("biz_type = 'guide_video' AND ref_id IS NOT NULL"),
+    )
+    connection = op.get_bind()
+    _seal_legacy_plays(connection)
+    _backfill_configs(connection)
+
+
+def downgrade() -> None:
+    _prepare_downgrade(op.get_bind())
+    op.drop_index("ux_coin_transaction_guide_video_ref", table_name="coin_transaction")
+    op.drop_index("ix_guide_video_play_expires_at", table_name="guide_video_play")
+    with op.batch_alter_table("guide_video_play") as batch:
+        batch.alter_column("status", existing_type=sa.String(length=24), type_=sa.String(length=16))
+        batch.drop_column("expires_at")
+        batch.drop_column("prepared_at")
+        batch.drop_column("settled_circles")
+        batch.drop_column("config_version")
+        batch.drop_column("duration_ms")
+        batch.alter_column("started_at", existing_type=sa.DateTime(timezone=True), nullable=False)
+        batch.alter_column("seq", existing_type=sa.Integer(), nullable=False)

alembic/versions/limit_policy_global_bundle.py

@@ -0,0 +1,193 @@
+"""store all global limit values in one complete JSON document
+
+Revision ID: limit_policy_global_bundle
+Revises: limit_policy_whitelist
+"""
+from __future__ import annotations
+
+from collections.abc import Sequence
+from typing import Any
+
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+from alembic import op
+
+revision: str = "limit_policy_global_bundle"
+down_revision: str | Sequence[str] | None = "limit_policy_whitelist"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_JSON = sa.JSON().with_variant(postgresql.JSONB(), "postgresql")
+_BUNDLE_KEY = "limit_policy_global"
+
+# rule_code, old sparse key, default, legacy structured key, legacy JSON field
+_RULES: tuple[tuple[str, str, int, str | None, str | None], ...] = (
+    ("compare.start.daily", "compare_daily_limit", 100, None, None),
+    ("sms.send.hourly", "sms_send_hourly_limit", 5, None, None),
+    ("sms.send.daily", "sms_send_daily_limit", 20, None, None),
+    ("sms.phone.cooldown", "sms_phone_cooldown_seconds", 60, None, None),
+    ("sms.code.failed_attempts", "sms_code_max_failed_attempts", 5, None, None),
+    ("sms.login.hourly", "sms_login_hourly_limit", 5, None, None),
+    ("wechat.bind.hourly", "wechat_bind_sms_hourly_limit", 5, None, None),
+    ("wechat.conflict.hourly", "wechat_conflict_hourly_limit", 5, None, None),
+    (
+        "ad.reward_video.daily",
+        "ad_reward_video_daily_limit",
+        500,
+        "ad_daily_limit",
+        None,
+    ),
+    (
+        "ad.feed.daily",
+        "ad_feed_daily_limit",
+        500,
+        "ad_daily_limit",
+        None,
+    ),
+    ("ad.reward_video.cooldown", "ad_cooldown_sec", 3, None, None),
+    (
+        "guide.video.lifetime",
+        "guide_video_max_plays",
+        3,
+        "coupon_guide_video",
+        "max_plays",
+    ),
+    ("phone.rebind.days", "phone_rebind_days", 30, None, None),
+    ("risk.sms.hourly", "risk_sms_hourly_threshold", 5, None, None),
+    (
+        "risk.oneclick.daily",
+        "risk_oneclick_daily_threshold",
+        20,
+        None,
+        None,
+    ),
+    (
+        "risk.compare.daily",
+        "risk_compare_daily_threshold",
+        100,
+        None,
+        None,
+    ),
+)
+
+
+def _table() -> sa.TableClause:
+    return sa.table(
+        "app_config",
+        sa.column("key", sa.String(64)),
+        sa.column("value", _JSON),
+        sa.column("updated_by_admin_id", sa.Integer),
+        sa.column("updated_at", sa.DateTime(timezone=True)),
+    )
+
+
+def _row(conn, table, key: str):
+    return conn.execute(
+        sa.select(
+            table.c.value,
+            table.c.updated_by_admin_id,
+        ).where(table.c.key == key)
+    ).mappings().first()
+
+
+def _int_or_none(value: Any) -> int | None:
+    if isinstance(value, bool):
+        return None
+    try:
+        return int(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    table = _table()
+    values = {rule_code: default for rule_code, _, default, _, _ in _RULES}
+
+    # Old shared/structured values are the lowest-precedence compatibility
+    # source.  Dedicated sparse keys override them.
+    for rule_code, _, _, legacy_key, legacy_field in _RULES:
+        if legacy_key is None:
+            continue
+        legacy = _row(conn, table, legacy_key)
+        if legacy is None:
+            continue
+        raw = legacy["value"]
+        if legacy_field is not None:
+            raw = raw.get(legacy_field) if isinstance(raw, dict) else None
+        parsed = _int_or_none(raw)
+        if parsed is not None:
+            values[rule_code] = parsed
+
+    for rule_code, sparse_key, _, _, _ in _RULES:
+        sparse = _row(conn, table, sparse_key)
+        parsed = _int_or_none(sparse["value"]) if sparse is not None else None
+        if parsed is not None:
+            values[rule_code] = parsed
+
+    # If a deployment already wrote the new key, preserve it over old keys.
+    bundle = _row(conn, table, _BUNDLE_KEY)
+    if bundle is not None and isinstance(bundle["value"], dict):
+        for rule_code, raw in bundle["value"].items():
+            if rule_code not in values:
+                continue
+            parsed = _int_or_none(raw)
+            if parsed is not None:
+                values[rule_code] = parsed
+
+    if bundle is None:
+        conn.execute(
+            table.insert().values(
+                key=_BUNDLE_KEY,
+                value=values,
+                updated_by_admin_id=None,
+            )
+        )
+    else:
+        conn.execute(
+            table.update()
+            .where(table.c.key == _BUNDLE_KEY)
+            .values(value=values, updated_at=sa.func.now())
+        )
+
+    sparse_keys = [sparse_key for _, sparse_key, _, _, _ in _RULES]
+    conn.execute(table.delete().where(table.c.key.in_(sparse_keys)))
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    table = _table()
+    bundle = _row(conn, table, _BUNDLE_KEY)
+    values = (
+        bundle["value"]
+        if bundle is not None and isinstance(bundle["value"], dict)
+        else {}
+    )
+    admin_id = bundle["updated_by_admin_id"] if bundle is not None else None
+
+    for rule_code, sparse_key, default, _, _ in _RULES:
+        value = _int_or_none(values.get(rule_code))
+        if value is None:
+            value = default
+        existing = _row(conn, table, sparse_key)
+        if existing is None:
+            conn.execute(
+                table.insert().values(
+                    key=sparse_key,
+                    value=value,
+                    updated_by_admin_id=admin_id,
+                )
+            )
+        else:
+            conn.execute(
+                table.update()
+                .where(table.c.key == sparse_key)
+                .values(
+                    value=value,
+                    updated_by_admin_id=admin_id,
+                    updated_at=sa.func.now(),
+                )
+            )
+
+    conn.execute(table.delete().where(table.c.key == _BUNDLE_KEY))

alembic/versions/limit_policy_whitelist.py

@@ -0,0 +1,128 @@
+"""add per-subject limit policy whitelist
+
+Revision ID: limit_policy_whitelist
+Revises: push_binding_isolation
+"""
+from __future__ import annotations
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+from alembic import op
+
+revision: str = "limit_policy_whitelist"
+down_revision: str | Sequence[str] | None = "push_binding_isolation"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+_JSON = sa.JSON().with_variant(postgresql.JSONB(), "postgresql")
+_PAGE = "limit-whitelist"
+_DEFAULT_ROLES = ("operator", "tech")
+
+
+def _role_table() -> sa.TableClause:
+    return sa.table(
+        "admin_role",
+        sa.column("name", sa.String),
+        sa.column("pages", _JSON),
+    )
+
+
+def _add_default_role_permissions() -> None:
+    """Grant the page without replacing existing role customisations."""
+    role = _role_table()
+    conn = op.get_bind()
+    rows = conn.execute(
+        sa.select(role.c.name, role.c.pages).where(
+            role.c.name.in_(_DEFAULT_ROLES)
+        )
+    ).all()
+    for name, pages in rows:
+        current_pages = list(pages or [])
+        if _PAGE not in current_pages:
+            conn.execute(
+                role.update()
+                .where(role.c.name == name)
+                .values(pages=[*current_pages, _PAGE])
+            )
+
+
+def _remove_default_role_permissions() -> None:
+    role = _role_table()
+    conn = op.get_bind()
+    rows = conn.execute(
+        sa.select(role.c.name, role.c.pages).where(
+            role.c.name.in_(_DEFAULT_ROLES)
+        )
+    ).all()
+    for name, pages in rows:
+        current_pages = list(pages or [])
+        if _PAGE in current_pages:
+            conn.execute(
+                role.update()
+                .where(role.c.name == name)
+                .values(
+                    pages=[page for page in current_pages if page != _PAGE]
+                )
+            )
+
+
+def upgrade() -> None:
+    op.create_table(
+        "limit_policy_override",
+        sa.Column("id", sa.Integer(), autoincrement=True, nullable=False),
+        sa.Column("subject_type", sa.String(length=16), nullable=False),
+        sa.Column("subject_value", sa.String(length=128), nullable=False),
+        sa.Column("rule_code", sa.String(length=64), nullable=False),
+        sa.Column("mode", sa.String(length=24), nullable=False),
+        sa.Column("limit_value", sa.Integer(), nullable=True),
+        sa.Column(
+            "enabled", sa.Boolean(), server_default=sa.true(), nullable=False
+        ),
+        sa.Column("starts_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("expires_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("reset_at", sa.DateTime(timezone=True), nullable=True),
+        sa.Column("reason", sa.String(length=256), nullable=True),
+        sa.Column("created_by_admin_id", sa.Integer(), nullable=True),
+        sa.Column(
+            "created_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.func.now(),
+            nullable=False,
+        ),
+        sa.Column(
+            "updated_at",
+            sa.DateTime(timezone=True),
+            server_default=sa.func.now(),
+            nullable=False,
+        ),
+        sa.PrimaryKeyConstraint("id"),
+        sa.UniqueConstraint(
+            "subject_type",
+            "subject_value",
+            "rule_code",
+            name="uq_limit_policy_subject_rule",
+        ),
+    )
+    op.create_index(
+        "ix_limit_policy_lookup",
+        "limit_policy_override",
+        ["subject_type", "subject_value", "rule_code", "enabled"],
+        unique=False,
+    )
+    op.create_index(
+        "ix_limit_policy_expires",
+        "limit_policy_override",
+        ["expires_at"],
+        unique=False,
+    )
+    _add_default_role_permissions()
+
+
+def downgrade() -> None:
+    _remove_default_role_permissions()
+    op.drop_index("ix_limit_policy_expires", table_name="limit_policy_override")
+    op.drop_index("ix_limit_policy_lookup", table_name="limit_policy_override")
+    op.drop_table("limit_policy_override")

alembic/versions/push_binding_isolation.py

@@ -0,0 +1,89 @@
+"""isolate vendor push binding across user accounts
+
+Revision ID: push_binding_isolation
+Revises: guide_video_ten_circle_v2
+"""
+from __future__ import annotations
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "push_binding_isolation"
+down_revision = "guide_video_ten_circle_v2"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("device_liveness") as batch_op:
+        batch_op.add_column(sa.Column("push_binding_id", sa.String(length=128), nullable=True))
+        batch_op.add_column(
+            sa.Column(
+                "push_binding_revoked",
+                sa.Boolean(),
+                nullable=False,
+                server_default=sa.false(),
+            )
+        )
+        batch_op.create_index(
+            "ix_device_liveness_push_binding_id",
+            ["push_binding_id"],
+            unique=False,
+        )
+
+    connection = op.get_bind()
+    device = sa.table(
+        "device_liveness",
+        sa.column("id", sa.Integer),
+        sa.column("push_vendor", sa.String),
+        sa.column("push_token", sa.String),
+        sa.column("updated_at", sa.DateTime),
+    )
+    connection.execute(
+        device.update()
+        .where(
+            sa.or_(
+                device.c.push_vendor == "",
+                device.c.push_token == "",
+            )
+        )
+        .values(push_vendor=None, push_token=None)
+    )
+
+    rows = connection.execute(
+        sa.select(device.c.id, device.c.push_vendor, device.c.push_token)
+        .where(
+            device.c.push_vendor.is_not(None),
+            device.c.push_token.is_not(None),
+        )
+        .order_by(device.c.updated_at.desc(), device.c.id.desc())
+    )
+    seen: set[tuple[str, str]] = set()
+    duplicate_ids: list[int] = []
+    for row in rows:
+        key = (row.push_vendor, row.push_token)
+        if key in seen:
+            duplicate_ids.append(row.id)
+        else:
+            seen.add(key)
+    if duplicate_ids:
+        connection.execute(
+            device.update()
+            .where(device.c.id.in_(duplicate_ids))
+            .values(push_vendor=None, push_token=None)
+        )
+
+    with op.batch_alter_table("device_liveness") as batch_op:
+        batch_op.create_unique_constraint(
+            "uq_device_liveness_vendor_token",
+            ["push_vendor", "push_token"],
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("device_liveness") as batch_op:
+        batch_op.drop_constraint("uq_device_liveness_vendor_token", type_="unique")
+        batch_op.drop_index("ix_device_liveness_push_binding_id")
+        batch_op.drop_column("push_binding_revoked")
+        batch_op.drop_column("push_binding_id")

app/admin/main.py

@@ -1,6 +1,6 @@
 """Admin 后台 FastAPI app(独立进程)。
 
-启动:uvicorn app.admin.main:admin_app --host 127.0.0.1 --port 8771
+启动:uvicorn app.admin.main:admin_app --host 127.0.0.1 --port 8773
 复用 App 的 DB/models/repositories/integrations;鉴权独立(admin JWT,见 app/admin/security.py)。
 现有 app.main:app 不 import 本模块,两进程互不影响。
 """
@@ -31,6 +31,7 @@ from app.admin.routers.feedback import router as feedback_router
 from app.admin.routers.feedback_qr import router as feedback_qr_router
 from app.admin.routers.guide_video import router as guide_video_router
 from app.admin.routers.huawei_review import router as huawei_review_router
+from app.admin.routers.limit_whitelist import router as limit_whitelist_router
 from app.admin.routers.onboarding import router as onboarding_router
 from app.admin.routers.ops_marquee_seed import router as ops_marquee_seed_router
 from app.admin.routers.ops_stat_config import router as ops_stat_config_router
@@ -103,6 +104,7 @@ admin_app.include_router(wallet_router)
 admin_app.include_router(withdraw_router)
 admin_app.include_router(price_report_router)
 admin_app.include_router(risk_monitor_router)
+admin_app.include_router(limit_whitelist_router)
 admin_app.include_router(feedback_router)
 admin_app.include_router(event_logs_router)
 admin_app.include_router(analytics_health_router)

app/admin/permissions.py

@@ -40,6 +40,7 @@ PERMISSION_CATALOG: list[dict] = [
         {"key": "analytics-health", "label": "埋点成功率"},
         {"key": "event-logs", "label": "埋点日志"},
         {"key": "audit-logs", "label": "审计日志"},
+        {"key": "limit-whitelist", "label": "白名单"},
     ]},
     {"group": "其他", "pages": [
         {"key": "admins", "label": "权限管理"},
@@ -58,13 +59,14 @@ BUILTIN_ROLES: list[dict] = [
     {"name": "operator", "label": "运营", "pages": [
         "dashboard", "coupon-data", "ad-revenue-report", "comparison-records",
         "cps", "risk-monitor", "device-liveness", "price-reports", "feedbacks", "huawei-review",
+        "limit-whitelist",
     ]},
     {"name": "finance", "label": "财务", "pages": [
         "dashboard", "ad-revenue-report", "cps", "invite-withdraws", "withdraws",
     ]},
     {"name": "tech", "label": "技术", "pages": [
         "dashboard", "risk-monitor", "device-liveness", "analytics-health", "config", "ad-revenue", "huawei-review",
-        "event-logs", "audit-logs",
+        "event-logs", "audit-logs", "limit-whitelist",
     ]},
 ]
 

app/admin/repositories/ad_revenue.py

@@ -43,7 +43,7 @@ _KNOWN_PROD_BUSINESS_CODE_IDS = frozenset({"104098712", "104099389"})
 _TEST_BUSINESS_CODE_IDS = frozenset({"104127529", "104127626", "104137445"})
 
 
-def _business_code_ids(db: Session, app_env: str | None) -> set[str]:
+def business_code_ids(db: Session, app_env: str | None) -> set[str]:
     """返回指定应用环境下可用于业务收益对账的 GroMore 聚合代码位。"""
     prod_config = app_config.get_ad_config(db)
     prod_ids = set(_KNOWN_PROD_BUSINESS_CODE_IDS) | {
@@ -320,10 +320,10 @@ def ad_revenue_report(
 
     # 业务口径仅保留正式配置/测试业务链路实际使用的代码位。穿山甲“全量”还包含广告测试
     # demo、插屏等没有客户端收益上报的曝光，两边直接比较会天然产生假差额。
-    business_code_ids: set[str] | None = None
+    business_ids: set[str] | None = None
     if revenue_scope == "business":
-        business_code_ids = _business_code_ids(db, app_env)
-        events = [e for e in events if e.get("our_code_id") in business_code_ids]
+        business_ids = business_code_ids(db, app_env)
+        events = [e for e in events if e.get("our_code_id") in business_ids]
 
     # 排序:time=按时间倒序(新→旧);ecpm=按 eCPM 数值倒序(eCPM 原值是字符串「分」,转数值排;
     # 纯发奖行用其发奖采用的 eCPM,缺失/非法计 0 排末尾)。
@@ -381,7 +381,7 @@ def ad_revenue_report(
             date_from=date_from,
             date_to=date_to,
             app_env=app_env,
-            our_code_ids=business_code_ids,
+            our_code_ids=business_ids,
         )
         if pangle_aggs:
             by_date = {a["date"]: a for a in pangle_aggs}

app/admin/repositories/limit_whitelist.py

@@ -0,0 +1,583 @@
+"""CRUD and presentation helpers for limit policy overrides."""
+
+from __future__ import annotations
+
+from datetime import UTC, datetime
+from hashlib import blake2b
+
+from sqlalchemy import String, cast, func, or_, select, text
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
+
+from app.core import limit_policy
+from app.models.app_config import AppConfig
+from app.models.comparison import ComparisonRecord
+from app.models.limit_policy import LimitPolicyOverride
+from app.models.risk import BehaviorEvent
+from app.models.user import User
+from app.repositories import risk as risk_repo
+
+
+class DuplicateOverrideError(Exception):
+    pass
+
+
+def lock_subject(db: Session, *, subject_type: str, subject_value: str) -> None:
+    """Serialize writes for one logical whitelist subject on PostgreSQL.
+
+    Row locks cannot protect a brand-new subject because there is no row to
+    lock yet.  A transaction-scoped advisory lock closes that gap and also
+    serializes concurrent appends that touch different rule rows.
+    """
+
+    bind = db.get_bind()
+    if bind.dialect.name != "postgresql":
+        return
+    identity = f"limit-whitelist:{subject_type}:{subject_value}".encode()
+    lock_id = int.from_bytes(
+        blake2b(identity, digest_size=8).digest(),
+        byteorder="big",
+        signed=True,
+    )
+    db.execute(
+        text("SELECT pg_advisory_xact_lock(:lock_id)"),
+        {"lock_id": lock_id},
+    )
+
+
+_EVENT_SOURCE_LABELS = {
+    risk_repo.EVENT_SMS_SEND: "短信验证码",
+    risk_repo.EVENT_SMS_LOGIN: "短信登录",
+    risk_repo.EVENT_ONECLICK_LOGIN: "一键登录",
+}
+
+
+def _matching_users(keyword: str):
+    pattern = f"%{keyword}%"
+    return select(User.id).where(
+        or_(
+            cast(User.id, String).ilike(pattern),
+            User.username.ilike(pattern),
+            User.phone.ilike(pattern),
+            User.nickname.ilike(pattern),
+        )
+    )
+
+
+def _user_maps(
+    db: Session, *, user_ids: set[int], phones: set[str]
+) -> tuple[dict[int, User], dict[str, User]]:
+    conditions = []
+    if user_ids:
+        conditions.append(User.id.in_(user_ids))
+    if phones:
+        conditions.append(User.phone.in_(phones))
+    if not conditions:
+        return {}, {}
+    users = list(db.scalars(select(User).where(or_(*conditions))).all())
+    return {user.id: user for user in users}, {user.phone: user for user in users}
+
+
+def _behavior_device_candidates(db: Session, *, keyword: str | None, limit: int) -> list[dict]:
+    conditions = [
+        BehaviorEvent.subject_type == limit_policy.SUBJECT_DEVICE,
+        BehaviorEvent.subject_id != "",
+        BehaviorEvent.subject_id.not_like(f"{limit_policy.LEGACY_IP_DEVICE_PREFIX}%"),
+    ]
+    if keyword:
+        value = keyword.strip()
+        pattern = f"%{value}%"
+        matched_users = _matching_users(value)
+        conditions.append(
+            or_(
+                BehaviorEvent.subject_id.ilike(pattern),
+                BehaviorEvent.device_id.ilike(pattern),
+                BehaviorEvent.device_model.ilike(pattern),
+                BehaviorEvent.phone.ilike(pattern),
+                cast(BehaviorEvent.user_id, String).ilike(pattern),
+                BehaviorEvent.user_id.in_(matched_users),
+                BehaviorEvent.phone.in_(
+                    select(User.phone).where(User.id.in_(_matching_users(value)))
+                ),
+            )
+        )
+
+    ranked = (
+        select(
+            BehaviorEvent.subject_id.label("device_id"),
+            BehaviorEvent.event_type,
+            BehaviorEvent.user_id,
+            BehaviorEvent.phone,
+            BehaviorEvent.device_model,
+            BehaviorEvent.occurred_at.label("last_active_at"),
+            func.row_number()
+            .over(
+                partition_by=BehaviorEvent.subject_id,
+                order_by=(
+                    BehaviorEvent.occurred_at.desc(),
+                    BehaviorEvent.id.desc(),
+                ),
+            )
+            .label("rank"),
+        )
+        .where(*conditions)
+        .subquery()
+    )
+    rows = db.execute(
+        select(ranked)
+        .where(ranked.c.rank == 1)
+        .order_by(ranked.c.last_active_at.desc(), ranked.c.device_id.asc())
+        .limit(limit)
+    ).all()
+    by_id, by_phone = _user_maps(
+        db,
+        user_ids={row.user_id for row in rows if row.user_id is not None},
+        phones={row.phone for row in rows if row.phone},
+    )
+    result = []
+    for row in rows:
+        user = by_id.get(row.user_id) if row.user_id is not None else None
+        user = user or (by_phone.get(row.phone) if row.phone else None)
+        result.append(
+            {
+                "device_id": row.device_id,
+                "source": row.event_type,
+                "source_label": _EVENT_SOURCE_LABELS.get(row.event_type, "登录/鉴权流水"),
+                "user_id": user.id if user else row.user_id,
+                "username": user.username if user else None,
+                "phone": user.phone if user else row.phone,
+                "nickname": user.nickname if user else None,
+                "device_model": row.device_model,
+                "last_active_at": row.last_active_at,
+            }
+        )
+    return result
+
+
+def _comparison_device_candidates(db: Session, *, keyword: str | None, limit: int) -> list[dict]:
+    conditions = [
+        ComparisonRecord.device_id.is_not(None),
+        ComparisonRecord.device_id != "",
+    ]
+    if keyword:
+        value = keyword.strip()
+        pattern = f"%{value}%"
+        conditions.append(
+            or_(
+                ComparisonRecord.device_id.ilike(pattern),
+                ComparisonRecord.device_model.ilike(pattern),
+                cast(ComparisonRecord.user_id, String).ilike(pattern),
+                ComparisonRecord.user_id.in_(_matching_users(value)),
+            )
+        )
+    ranked = (
+        select(
+            ComparisonRecord.device_id,
+            ComparisonRecord.user_id,
+            ComparisonRecord.device_model,
+            ComparisonRecord.created_at.label("last_active_at"),
+            func.row_number()
+            .over(
+                partition_by=ComparisonRecord.device_id,
+                order_by=(
+                    ComparisonRecord.created_at.desc(),
+                    ComparisonRecord.id.desc(),
+                ),
+            )
+            .label("rank"),
+        )
+        .where(*conditions)
+        .subquery()
+    )
+    rows = db.execute(
+        select(ranked)
+        .where(ranked.c.rank == 1)
+        .order_by(ranked.c.last_active_at.desc(), ranked.c.device_id.asc())
+        .limit(limit)
+    ).all()
+    by_id, _ = _user_maps(
+        db,
+        user_ids={row.user_id for row in rows if row.user_id is not None},
+        phones=set(),
+    )
+    return [
+        {
+            "device_id": row.device_id,
+            "source": "comparison_record",
+            "source_label": "比价记录",
+            "user_id": user.id if user else row.user_id,
+            "username": user.username if user else None,
+            "phone": user.phone if user else None,
+            "nickname": user.nickname if user else None,
+            "device_model": row.device_model,
+            "last_active_at": row.last_active_at,
+        }
+        for row in rows
+        for user in [by_id.get(row.user_id) if row.user_id is not None else None]
+    ]
+
+
+def list_device_candidates(
+    db: Session,
+    *,
+    rule_code: str,
+    keyword: str | None = None,
+    limit: int = 30,
+) -> list[dict]:
+    rule = limit_policy.get_rule(rule_code)
+    if limit_policy.SUBJECT_DEVICE not in rule.subject_types:
+        raise ValueError("当前限制项不支持设备白名单")
+    if limit_policy.device_source_scope(rule_code) == "comparison":
+        return _comparison_device_candidates(db, keyword=keyword, limit=limit)
+    return _behavior_device_candidates(db, keyword=keyword, limit=limit)
+
+
+def _aware(value: datetime | None) -> datetime | None:
+    if value is None:
+        return None
+    return value.replace(tzinfo=UTC) if value.tzinfo is None else value
+
+
+def status_of(row: LimitPolicyOverride, *, now: datetime | None = None) -> str:
+    now = _aware(now) or datetime.now(UTC)
+    if not row.enabled:
+        return "disabled"
+    if _aware(row.starts_at) and _aware(row.starts_at) > now:
+        return "scheduled"
+    if _aware(row.expires_at) and _aware(row.expires_at) <= now:
+        return "expired"
+    return "active"
+
+
+def to_dict(db: Session, row: LimitPolicyOverride) -> dict:
+    rule = limit_policy.get_rule(row.rule_code)
+    effective = limit_policy.resolve(
+        db,
+        row.rule_code,
+        phone=row.subject_value if row.subject_type == "phone" else None,
+        device=row.subject_value if row.subject_type == "device" else None,
+    )
+    return {
+        "id": row.id,
+        "subject_type": row.subject_type,
+        "subject_value": row.subject_value,
+        "rule_code": row.rule_code,
+        "rule_label": rule.label,
+        "rule_group": rule.group,
+        "mode": row.mode,
+        "limit_value": row.limit_value,
+        "global_limit": effective.global_limit,
+        "effective_limit": effective.limit,
+        "enabled": row.enabled,
+        "starts_at": row.starts_at,
+        "expires_at": row.expires_at,
+        "reset_at": row.reset_at,
+        "reason": row.reason,
+        "status": status_of(row),
+        "created_by_admin_id": row.created_by_admin_id,
+        "created_at": row.created_at,
+        "updated_at": row.updated_at,
+    }
+
+
+def list_rows(
+    db: Session,
+    *,
+    subject_type: str | None = None,
+    keyword: str | None = None,
+    rule_code: str | None = None,
+    offset: int = 0,
+    limit: int = 100,
+) -> tuple[list[LimitPolicyOverride], int]:
+    stmt = select(LimitPolicyOverride)
+    count_stmt = select(func.count(LimitPolicyOverride.id))
+    conditions = []
+    if subject_type:
+        conditions.append(LimitPolicyOverride.subject_type == subject_type)
+    if keyword:
+        conditions.append(LimitPolicyOverride.subject_value.ilike(f"%{keyword.strip()}%"))
+    if rule_code:
+        conditions.append(LimitPolicyOverride.rule_code == rule_code)
+    if conditions:
+        stmt = stmt.where(*conditions)
+        count_stmt = count_stmt.where(*conditions)
+    total = int(db.scalar(count_stmt) or 0)
+    rows = list(
+        db.execute(
+            stmt.order_by(
+                LimitPolicyOverride.created_at.desc(),
+                LimitPolicyOverride.id.desc(),
+            )
+            .offset(offset)
+            .limit(limit)
+        ).scalars()
+    )
+    return rows, total
+
+
+def list_subject_rows(
+    db: Session,
+    *,
+    subject_type: str | None = None,
+    keyword: str | None = None,
+    rule_code: str | None = None,
+    offset: int = 0,
+    limit: int = 10,
+) -> tuple[list[tuple[str, str, list[LimitPolicyOverride], datetime]], int]:
+    """List whitelist entries grouped and paginated by subject.
+
+    The filter decides which subjects match.  Once a subject matches, all of
+    its rules are returned so category counts and the edit form stay complete.
+    Ordering uses the original creation time, therefore editing a subject does
+    not unexpectedly move it to the first page.
+    """
+
+    conditions = [
+        LimitPolicyOverride.mode.in_(
+            [limit_policy.MODE_UNLIMITED, limit_policy.MODE_SUPPRESS_ALERT]
+        )
+    ]
+    if subject_type:
+        conditions.append(LimitPolicyOverride.subject_type == subject_type)
+    if keyword:
+        conditions.append(LimitPolicyOverride.subject_value.ilike(f"%{keyword.strip()}%"))
+    if rule_code:
+        conditions.append(LimitPolicyOverride.rule_code == rule_code)
+
+    matched_subjects = (
+        select(
+            LimitPolicyOverride.subject_type.label("subject_type"),
+            LimitPolicyOverride.subject_value.label("subject_value"),
+            func.min(LimitPolicyOverride.created_at).label("subject_created_at"),
+        )
+        .where(*conditions)
+        .group_by(
+            LimitPolicyOverride.subject_type,
+            LimitPolicyOverride.subject_value,
+        )
+        .subquery()
+    )
+    total = int(db.scalar(select(func.count()).select_from(matched_subjects)) or 0)
+    subjects = list(
+        db.execute(
+            select(
+                matched_subjects.c.subject_type,
+                matched_subjects.c.subject_value,
+                matched_subjects.c.subject_created_at,
+            )
+            .order_by(
+                matched_subjects.c.subject_created_at.desc(),
+                matched_subjects.c.subject_type,
+                matched_subjects.c.subject_value,
+            )
+            .offset(offset)
+            .limit(limit)
+        ).all()
+    )
+    if not subjects:
+        return [], total
+
+    pair_conditions = [
+        (
+            (LimitPolicyOverride.subject_type == item.subject_type)
+            & (LimitPolicyOverride.subject_value == item.subject_value)
+        )
+        for item in subjects
+    ]
+    rows = list(
+        db.scalars(
+            select(LimitPolicyOverride)
+            .where(
+                or_(*pair_conditions),
+                LimitPolicyOverride.mode.in_(
+                    [
+                        limit_policy.MODE_UNLIMITED,
+                        limit_policy.MODE_SUPPRESS_ALERT,
+                    ]
+                ),
+            )
+            .order_by(
+                LimitPolicyOverride.created_at.desc(),
+                LimitPolicyOverride.id.desc(),
+            )
+        ).all()
+    )
+    rows_by_subject: dict[tuple[str, str], list[LimitPolicyOverride]] = {}
+    for row in rows:
+        rows_by_subject.setdefault((row.subject_type, row.subject_value), []).append(row)
+
+    return [
+        (
+            item.subject_type,
+            item.subject_value,
+            rows_by_subject.get((item.subject_type, item.subject_value), []),
+            item.subject_created_at,
+        )
+        for item in subjects
+    ], total
+
+
+def rows_for_subject(
+    db: Session, *, subject_type: str, subject_value: str
+) -> list[LimitPolicyOverride]:
+    return list(
+        db.scalars(
+            select(LimitPolicyOverride)
+            .where(
+                LimitPolicyOverride.subject_type == subject_type,
+                LimitPolicyOverride.subject_value == subject_value,
+            )
+            .order_by(
+                LimitPolicyOverride.created_at.desc(),
+                LimitPolicyOverride.id.desc(),
+            )
+        ).all()
+    )
+
+
+def create(
+    db: Session,
+    *,
+    subject_type: str,
+    subject_value: str,
+    rule_code: str,
+    mode: str,
+    limit_value: int | None,
+    enabled: bool,
+    starts_at: datetime | None,
+    expires_at: datetime | None,
+    reason: str,
+    admin_id: int,
+    reactivate_inactive: bool = False,
+) -> LimitPolicyOverride:
+    rule = limit_policy.get_rule(rule_code)
+    subject_value = limit_policy.validate_whitelist_subject(subject_type, subject_value)
+    limit_policy.validate_override(
+        rule,
+        subject_type=subject_type,
+        mode=mode,
+        limit_value=limit_value,
+        starts_at=starts_at,
+        expires_at=expires_at,
+    )
+    existing = db.execute(
+        select(LimitPolicyOverride).where(
+            LimitPolicyOverride.subject_type == subject_type,
+            LimitPolicyOverride.subject_value == subject_value,
+            LimitPolicyOverride.rule_code == rule_code,
+        )
+    ).scalar_one_or_none()
+    if existing is not None:
+        if not reactivate_inactive or status_of(existing) not in {"disabled", "expired"}:
+            raise DuplicateOverrideError
+        # “恢复全局”与自然过期都会保留原行供审计。再次加入白名单时
+        # 复用该行，避免唯一键让批量新增永久失败；完整变更仍写 admin 审计日志。
+        existing.mode = mode
+        existing.limit_value = limit_value if mode == limit_policy.MODE_OVERRIDE else None
+        existing.enabled = enabled
+        existing.starts_at = starts_at
+        existing.expires_at = expires_at
+        existing.reset_at = None
+        existing.reason = reason.strip() or None
+        db.flush()
+        return existing
+
+    row = LimitPolicyOverride(
+        subject_type=subject_type,
+        subject_value=subject_value,
+        rule_code=rule_code,
+        mode=mode,
+        limit_value=limit_value if mode == limit_policy.MODE_OVERRIDE else None,
+        enabled=enabled,
+        starts_at=starts_at,
+        expires_at=expires_at,
+        reason=reason.strip() or None,
+        created_by_admin_id=admin_id,
+    )
+    db.add(row)
+    try:
+        db.flush()
+    except IntegrityError as exc:
+        db.rollback()
+        raise DuplicateOverrideError from exc
+    return row
+
+
+def update(
+    db: Session,
+    row: LimitPolicyOverride,
+    *,
+    enabled: bool | None,
+    starts_at: datetime | None,
+    expires_at: datetime | None,
+    reason: str | None,
+    fields_set: set[str],
+) -> LimitPolicyOverride:
+    new_starts = starts_at if "starts_at" in fields_set else row.starts_at
+    new_expires = expires_at if "expires_at" in fields_set else row.expires_at
+    new_enabled = enabled if "enabled" in fields_set and enabled is not None else row.enabled
+    rule = limit_policy.get_rule(row.rule_code)
+    if row.mode == limit_policy.MODE_OVERRIDE and new_enabled:
+        raise ValueError("历史覆盖配置只能停用、恢复全局或删除")
+    if new_enabled:
+        limit_policy.validate_override(
+            rule,
+            subject_type=row.subject_type,
+            mode=row.mode,
+            limit_value=None,
+            starts_at=new_starts,
+            expires_at=new_expires,
+        )
+    if "enabled" in fields_set and enabled is not None:
+        row.enabled = enabled
+    if "starts_at" in fields_set:
+        row.starts_at = starts_at
+    if "expires_at" in fields_set:
+        row.expires_at = expires_at
+    if "reason" in fields_set:
+        row.reason = reason.strip() if reason else None
+    db.flush()
+    return row
+
+
+def update_global_limit(
+    db: Session, rule_code: str, value: int, *, admin_id: int
+) -> tuple[int, int]:
+    rule = limit_policy.get_rule(rule_code)
+    if not rule.min_value <= value <= rule.max_value:
+        raise ValueError(f"限制值必须在 {rule.min_value} 到 {rule.max_value} 之间")
+    before = limit_policy.resolve(db, rule_code).global_limit
+    limit_policy.set_global_limits(
+        db,
+        {rule.code: value},
+        admin_id=admin_id,
+        commit=False,
+    )
+    # 引导视频仍有一个旧的专用配置页。同步其结构化配置，保证两个入口读取
+    # 同一数值；业务判定仍统一走 limit_policy。
+    if rule.legacy_config_key and rule.legacy_json_field:
+        legacy = db.get(AppConfig, rule.legacy_config_key)
+        legacy_value = (
+            dict(legacy.value) if legacy is not None and isinstance(legacy.value, dict) else {}
+        )
+        legacy_value[rule.legacy_json_field] = value
+        if legacy is None:
+            legacy = AppConfig(
+                key=rule.legacy_config_key,
+                value=legacy_value,
+                updated_by_admin_id=admin_id,
+            )
+            db.add(legacy)
+        else:
+            legacy.value = legacy_value
+            legacy.updated_by_admin_id = admin_id
+        db.flush()
+    return before, value
+
+
+def restore_global(row: LimitPolicyOverride) -> LimitPolicyOverride:
+    """Stop applying the exception while retaining its history for audit."""
+
+    row.enabled = False
+    row.reset_at = None
+    return row

app/admin/repositories/queries.py

@@ -14,6 +14,7 @@ from sqlalchemy.orm import Session
 
 from app.core import rewards
 from app.core.config import settings
+from app.models.ad_ecpm import AdEcpmRecord
 from app.models.ad_feed_reward import AdFeedRewardRecord
 from app.models.ad_reward import AdRewardRecord
 from app.models.admin import AdminAuditLog
@@ -1257,11 +1258,15 @@ def user_reward_stats(
     date_from: datetime | None = None,
     date_to: datetime | None = None,
     withdraw_source: str | None = None,
+    app_env: str | None = None,
+    revenue_scope: str = "all",
+    feed_scene: str | None = None,
 ) -> dict:
     """提现详情「用户统计区」10 项。窗口作用于除「现金余额」外的所有项(余额是当前快照)。
 
-    口径:激励视频/信息流只统计 granted;数量——视频按条数、信息流按份数(unit_count 累加);
-    平均 eCPM 用原始分值(分/千次)按记录取算术平均;各「提现」= 该来源累计金币折现。
+    口径:激励视频/信息流奖励数量只统计 granted;数量——视频按条数、信息流按份数(unit_count 累加)。
+    平均 Draw eCPM 与广告收益报表一致:基于 ad_ecpm_record 的全部 draw/feed 展示记录计算，
+    不以是否发奖为筛选条件。各「提现」= 该来源累计金币折现。
     传统任务 = 窗口内正向金币中,排除广告(reward_video/feed_ad_reward)与人工调整后的折现。
     """
     withdraw_source_conds = (
@@ -1296,30 +1301,68 @@ def user_reward_stats(
 
     # 只投影本统计实际使用的列。避免滚动发布或旧本地库尚未补齐无关新列时，
     # SQLAlchemy 因 select(ORM) 自动展开整表字段而让提现详情整体 500。
+    business_ids: set[str] | None = None
+    if revenue_scope == "business":
+        # 与广告收益报表共用正式/测试业务代码位集合，避免两个页面随配置切换后再次漂移。
+        from app.admin.repositories.ad_revenue import business_code_ids
+
+        business_ids = business_code_ids(db, app_env)
+
+    rv_conds = [
+        AdRewardRecord.user_id == user_id,
+        AdRewardRecord.reward_scene == "reward_video",
+        AdRewardRecord.status == "granted",
+        *_window_conds(AdRewardRecord.created_at, date_from, date_to),
+    ]
+    if app_env is not None:
+        rv_conds.append(AdRewardRecord.app_env == app_env)
+    if business_ids is not None:
+        rv_conds.append(AdRewardRecord.our_code_id.in_(business_ids))
     rv = db.execute(
         select(AdRewardRecord.ecpm_raw, AdRewardRecord.coin).where(
-            AdRewardRecord.user_id == user_id,
-            AdRewardRecord.reward_scene == "reward_video",
-            AdRewardRecord.status == "granted",
-            *_window_conds(AdRewardRecord.created_at, date_from, date_to),
+            *rv_conds,
         )
     ).all()
     rv_ecpms = [rewards.parse_ecpm_fen(r.ecpm_raw) for r in rv if r.ecpm_raw]
     rv_coins = sum(r.coin for r in rv)
 
-    feed = db.execute(
+    feed_reward_conds = [
+        AdFeedRewardRecord.user_id == user_id,
+        AdFeedRewardRecord.status == "granted",
+        *_window_conds(AdFeedRewardRecord.created_at, date_from, date_to),
+    ]
+    if app_env is not None:
+        feed_reward_conds.append(AdFeedRewardRecord.app_env == app_env)
+    if feed_scene is not None:
+        feed_reward_conds.append(AdFeedRewardRecord.feed_scene == feed_scene)
+    if business_ids is not None:
+        feed_reward_conds.append(AdFeedRewardRecord.our_code_id.in_(business_ids))
+    feed_rewards = db.execute(
         select(
             AdFeedRewardRecord.unit_count,
-            AdFeedRewardRecord.ecpm_raw,
             AdFeedRewardRecord.coin,
         ).where(
-            AdFeedRewardRecord.user_id == user_id,
-            AdFeedRewardRecord.status == "granted",
-            *_window_conds(AdFeedRewardRecord.created_at, date_from, date_to),
+            *feed_reward_conds,
         )
     ).all()
-    feed_ecpms = [rewards.parse_ecpm_fen(f.ecpm_raw) for f in feed if f.ecpm_raw]
-    feed_coins = sum(f.coin for f in feed)
+    feed_coins = sum(f.coin for f in feed_rewards)
+
+    feed_impression_conds = [
+        AdEcpmRecord.user_id == user_id,
+        AdEcpmRecord.ad_type.in_(("draw", "feed")),
+        *_window_conds(AdEcpmRecord.created_at, date_from, date_to),
+    ]
+    if app_env is not None:
+        feed_impression_conds.append(AdEcpmRecord.app_env == app_env)
+    if feed_scene is not None:
+        feed_impression_conds.append(AdEcpmRecord.feed_scene == feed_scene)
+    if business_ids is not None:
+        feed_impression_conds.append(AdEcpmRecord.our_code_id.in_(business_ids))
+    feed_impressions = db.execute(
+        select(AdEcpmRecord.ecpm_raw).where(*feed_impression_conds)
+    ).all()
+    # 与 ad_revenue.category_stats 相同:每次展示权重相同，非法原值按 parse_ecpm_fen 记 0。
+    feed_ecpms = [rewards.parse_ecpm_fen(row.ecpm_raw) for row in feed_impressions]
 
     trad_coins = db.execute(
         select(func.coalesce(func.sum(CoinTransaction.amount), 0)).where(
@@ -1338,7 +1381,7 @@ def user_reward_stats(
         "reward_video_count": len(rv),
         "reward_video_avg_ecpm": round(sum(rv_ecpms) / len(rv_ecpms), 2) if rv_ecpms else 0.0,
         "reward_video_cash_cents": _coins_to_cents(rv_coins),
-        "feed_count": int(sum(f.unit_count for f in feed)),
+        "feed_count": int(sum(f.unit_count for f in feed_rewards)),
         "feed_avg_ecpm": round(sum(feed_ecpms) / len(feed_ecpms), 2) if feed_ecpms else 0.0,
         "feed_cash_cents": _coins_to_cents(feed_coins),
     }

app/admin/routers/config.py

@@ -12,9 +12,11 @@ from fastapi import APIRouter, Depends, HTTPException, Request
 from app.admin.audit import write_audit
 from app.admin.deps import AdminDb, get_client_ip, get_current_admin, require_role
 from app.admin.schemas.config import ConfigItemOut, ConfigUpdateRequest
+from app.core import limit_policy
 from app.core.config_schema import CONFIG_DEFS
 from app.core.rewards import SIGNIN_CYCLE_LEN
 from app.models.admin import AdminUser
+from app.models.app_config import AppConfig
 from app.repositories import app_config
 
 router = APIRouter(
@@ -54,7 +56,35 @@ def _validate(key: str, value: Any) -> None:
             raise ValueError("需为布尔值")
 
 
+def _limit_item(
+    key: str,
+    values: dict[str, int],
+    bundle: AppConfig | None,
+) -> ConfigItemOut:
+    definition = CONFIG_DEFS[key]
+    rule_code = limit_policy.RULE_CODE_BY_CONFIG_KEY[key]
+    return ConfigItemOut(
+        key=key,
+        value=values[rule_code],
+        label=definition["label"],
+        group=definition["group"],
+        type=definition["type"],
+        help=definition.get("help"),
+        default=definition["default"],
+        overridden=bundle is not None,
+        updated_at=(
+            bundle.updated_at.isoformat() if bundle is not None else None
+        ),
+    )
+
+
 def _item(db, key: str) -> ConfigItemOut:
+    if key in limit_policy.RULE_CODE_BY_CONFIG_KEY:
+        return _limit_item(
+            key,
+            limit_policy.get_global_limits(db),
+            db.get(AppConfig, limit_policy.LIMIT_POLICY_GLOBAL_KEY),
+        )
     for item in app_config.list_all(db):
         if item["key"] == key:
             return ConfigItemOut(**item)
@@ -64,11 +94,20 @@ def _item(db, key: str) -> ConfigItemOut:
 @router.get("", response_model=list[ConfigItemOut], summary="所有可配项 + 当前值(不含 hidden)")
 def list_config(db: AdminDb) -> list[ConfigItemOut]:
     # hidden 项(已下线/由专用页管理,如福利页任务·里程碑·看广告调参、首页轮播数据源)不在本页渲染。
-    return [
-        ConfigItemOut(**item)
-        for item in app_config.list_all(db)
-        if not CONFIG_DEFS[item["key"]].get("hidden")
-    ]
+    legacy_items = {
+        item["key"]: item for item in app_config.list_all(db)
+    }
+    values = limit_policy.get_global_limits(db)
+    bundle = db.get(AppConfig, limit_policy.LIMIT_POLICY_GLOBAL_KEY)
+    out: list[ConfigItemOut] = []
+    for key, definition in CONFIG_DEFS.items():
+        if definition.get("hidden"):
+            continue
+        if key in limit_policy.RULE_CODE_BY_CONFIG_KEY:
+            out.append(_limit_item(key, values, bundle))
+        else:
+            out.append(ConfigItemOut(**legacy_items[key]))
+    return out
 
 
 @router.patch("/{key}", response_model=ConfigItemOut, summary="改某项配置(带审计)")
@@ -83,11 +122,39 @@ def update_config(
         raise HTTPException(status_code=404, detail="未知配置项")
     try:
         _validate(key, body.value)
+        rule_code = limit_policy.RULE_CODE_BY_CONFIG_KEY.get(key)
+        if rule_code is not None:
+            before = limit_policy.get_global_limits(db)[rule_code]
+            limit_policy.set_global_limits(
+                db,
+                {rule_code: body.value},
+                admin_id=admin.id,
+                commit=False,
+            )
+        else:
+            before = app_config.get_value(db, key)
+            app_config.set_value(
+                db,
+                key,
+                body.value,
+                admin_id=admin.id,
+                commit=False,
+            )
+        if key == "ad_daily_limit":
+            # 旧系统配置接口过去只有一个广告日上限。仍有人直接调用时，同时同步
+            # 新的激励视频/Draw 两项，避免旧入口写入后业务实际值不变。
+            limit_policy.set_global_limits(
+                db,
+                {
+                    "ad.reward_video.daily": body.value,
+                    "ad.feed.daily": body.value,
+                },
+                admin_id=admin.id,
+                commit=False,
+            )
     except ValueError as e:
+        db.rollback()
         raise HTTPException(status_code=400, detail=str(e)) from e
-
-    before = app_config.get_value(db, key)
-    app_config.set_value(db, key, body.value, admin_id=admin.id, commit=False)
     write_audit(
         db, admin, action="config.set", target_type="config", target_id=key,
         detail={"before": before, "after": body.value}, ip=get_client_ip(request), commit=False,

app/admin/routers/feedback.py

@@ -1,6 +1,7 @@
 """admin 反馈工单:列表筛选 + 审核采纳/拒绝(带金币发放与审计)。"""
 from __future__ import annotations
 
+import logging
 from datetime import datetime
 from typing import Annotated
 
@@ -25,6 +26,8 @@ from app.models.feedback import Feedback
 from app.repositories import wallet as wallet_repo
 from app.services import notification_events
 
+logger = logging.getLogger("shagua.admin.feedback")
+
 router = APIRouter(
     prefix="/admin/api/feedbacks",
     tags=["admin-feedback"],
@@ -46,10 +49,24 @@ def _approve_feedback(
     *,
     bulk: bool = False,
 ) -> FeedbackOut:
+    logger.info(
+        "feedback approve started feedback_id=%s admin_id=%s bulk=%s",
+        feedback_id, admin.id, bulk,
+    )
     fb = db.get(Feedback, feedback_id, with_for_update=True)
     if fb is None:
+        logger.warning(
+            "feedback approve rejected not found feedback_id=%s admin_id=%s bulk=%s",
+            feedback_id, admin.id, bulk,
+        )
         raise HTTPException(status_code=404, detail="反馈不存在")
-    _ensure_pending(fb)
+    if fb.status not in {"pending", "new"}:
+        logger.warning(
+            "feedback approve rejected invalid status feedback_id=%s user_id=%s "
+            "admin_id=%s status=%s bulk=%s",
+            feedback_id, fb.user_id, admin.id, fb.status, bulk,
+        )
+        _ensure_pending(fb)
 
     before = fb.status
     mutations.review_feedback(
@@ -91,8 +108,18 @@ def _approve_feedback(
     )
     db.commit()
     db.refresh(fb)
+    logger.info(
+        "feedback approve committed feedback_id=%s user_id=%s admin_id=%s "
+        "before=%s after=%s reward_coins=%s bulk=%s",
+        feedback_id, fb.user_id, admin.id, before, fb.status, payload.reward_coins, bulk,
+    )
     out = FeedbackOut.model_validate(fb)
     notification_events.notify_feedback_reward(db, fb)
+    logger.info(
+        "feedback approve notification dispatch returned feedback_id=%s user_id=%s "
+        "admin_id=%s notification_type=feedback_reward bulk=%s",
+        feedback_id, fb.user_id, admin.id, bulk,
+    )
     return out
 
 
@@ -105,10 +132,24 @@ def _reject_feedback(
     *,
     bulk: bool = False,
 ) -> FeedbackOut:
+    logger.info(
+        "feedback reject started feedback_id=%s admin_id=%s bulk=%s",
+        feedback_id, admin.id, bulk,
+    )
     fb = db.get(Feedback, feedback_id, with_for_update=True)
     if fb is None:
+        logger.warning(
+            "feedback reject rejected not found feedback_id=%s admin_id=%s bulk=%s",
+            feedback_id, admin.id, bulk,
+        )
         raise HTTPException(status_code=404, detail="反馈不存在")
-    _ensure_pending(fb)
+    if fb.status not in {"pending", "new"}:
+        logger.warning(
+            "feedback reject rejected invalid status feedback_id=%s user_id=%s "
+            "admin_id=%s status=%s bulk=%s",
+            feedback_id, fb.user_id, admin.id, fb.status, bulk,
+        )
+        _ensure_pending(fb)
 
     before = fb.status
     mutations.review_feedback(
@@ -142,8 +183,18 @@ def _reject_feedback(
     )
     db.commit()
     db.refresh(fb)
+    logger.info(
+        "feedback reject committed feedback_id=%s user_id=%s admin_id=%s "
+        "before=%s after=%s bulk=%s",
+        feedback_id, fb.user_id, admin.id, before, fb.status, bulk,
+    )
     out = FeedbackOut.model_validate(fb)
     notification_events.notify_feedback_reply(db, fb)
+    logger.info(
+        "feedback reject notification dispatch returned feedback_id=%s user_id=%s "
+        "admin_id=%s notification_type=feedback_reply bulk=%s",
+        feedback_id, fb.user_id, admin.id, bulk,
+    )
     return out
 
 
@@ -201,6 +252,10 @@ def bulk_approve_feedbacks(
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
 ) -> FeedbackBulkResult:
+    logger.info(
+        "feedback bulk approve started admin_id=%s item_count=%s",
+        admin.id, len(body.ids),
+    )
     results: list[FeedbackBulkItemResult] = []
     ip = get_client_ip(request)
     for feedback_id in body.ids:
@@ -209,11 +264,24 @@ def bulk_approve_feedbacks(
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=True, status=out.status))
         except HTTPException as exc:
             db.rollback()
+            logger.warning(
+                "feedback bulk approve item failed feedback_id=%s admin_id=%s error=%s",
+                feedback_id, admin.id, exc.detail,
+            )
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=False, error=str(exc.detail)))
         except Exception:  # noqa: BLE001 - 单笔失败不打断整批
             db.rollback()
+            logger.exception(
+                "feedback bulk approve item failed feedback_id=%s admin_id=%s",
+                feedback_id, admin.id,
+            )
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=False, error="系统异常"))
-    return _bulk_result(results)
+    result = _bulk_result(results)
+    logger.info(
+        "feedback bulk approve completed admin_id=%s total=%s success=%s failed=%s",
+        admin.id, result.total, result.success, result.failed,
+    )
+    return result
 
 
 @router.post("/bulk/reject", response_model=FeedbackBulkResult, summary="批量拒绝采纳反馈")
@@ -223,6 +291,10 @@ def bulk_reject_feedbacks(
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
 ) -> FeedbackBulkResult:
+    logger.info(
+        "feedback bulk reject started admin_id=%s item_count=%s",
+        admin.id, len(body.ids),
+    )
     results: list[FeedbackBulkItemResult] = []
     ip = get_client_ip(request)
     for feedback_id in body.ids:
@@ -231,11 +303,24 @@ def bulk_reject_feedbacks(
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=True, status=out.status))
         except HTTPException as exc:
             db.rollback()
+            logger.warning(
+                "feedback bulk reject item failed feedback_id=%s admin_id=%s error=%s",
+                feedback_id, admin.id, exc.detail,
+            )
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=False, error=str(exc.detail)))
         except Exception:  # noqa: BLE001 - 单笔失败不打断整批
             db.rollback()
+            logger.exception(
+                "feedback bulk reject item failed feedback_id=%s admin_id=%s",
+                feedback_id, admin.id,
+            )
             results.append(FeedbackBulkItemResult(id=feedback_id, ok=False, error="系统异常"))
-    return _bulk_result(results)
+    result = _bulk_result(results)
+    logger.info(
+        "feedback bulk reject completed admin_id=%s total=%s success=%s failed=%s",
+        admin.id, result.total, result.success, result.failed,
+    )
+    return result
 
 
 @router.post("/{feedback_id}/handle", response_model=OkResponse, summary="标记反馈已处理")
@@ -258,7 +343,16 @@ def approve_feedback(
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
 ) -> FeedbackOut:
-    return _approve_feedback(db, admin, feedback_id, payload, get_client_ip(request))
+    try:
+        return _approve_feedback(db, admin, feedback_id, payload, get_client_ip(request))
+    except HTTPException:
+        raise
+    except Exception:
+        logger.exception(
+            "feedback approve failed feedback_id=%s admin_id=%s",
+            feedback_id, admin.id,
+        )
+        raise
 
 
 @router.post("/{feedback_id}/reject", response_model=FeedbackOut, summary="拒绝采纳反馈")
@@ -269,4 +363,13 @@ def reject_feedback(
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
 ) -> FeedbackOut:
-    return _reject_feedback(db, admin, feedback_id, payload, get_client_ip(request))
+    try:
+        return _reject_feedback(db, admin, feedback_id, payload, get_client_ip(request))
+    except HTTPException:
+        raise
+    except Exception:
+        logger.exception(
+            "feedback reject failed feedback_id=%s admin_id=%s",
+            feedback_id, admin.id,
+        )
+        raise

app/admin/routers/guide_video.py

@@ -9,7 +9,7 @@ client_max_body_size,见 shaguabijia-admin-web/deploy/nginx/admin.shaguabijia.co
 """
 from __future__ import annotations
 
-from typing import Annotated
+from typing import Annotated, Literal
 
 from fastapi import APIRouter, Depends, File, HTTPException, Request, UploadFile
 
@@ -27,14 +27,16 @@ router = APIRouter(
 )
 
 
-def _out(db: AdminDb) -> GuideVideoConfigOut:
-    """配置 + 播放统计合成响应(四个写接口都以最新状态返回,前端一次同步到位)。"""
-    return GuideVideoConfigOut(**guide_video.get_config(db), **guide_video.play_stats(db))
+GuideScene = Literal["coupon", "comparison"]
+
+
+def _out(db: AdminDb, scene: GuideScene) -> GuideVideoConfigOut:
+    return GuideVideoConfigOut(scene=scene, **guide_video.get_config(db, scene))
 
 
 @router.get("", response_model=GuideVideoConfigOut, summary="新手引导视频配置(领券浮层)")
-def get_config(db: AdminDb) -> GuideVideoConfigOut:
-    return _out(db)
+def get_config(db: AdminDb, scene: GuideScene = "coupon") -> GuideVideoConfigOut:
+    return _out(db, scene)
 
 
 @router.patch("", response_model=GuideVideoConfigOut, summary="改开关/次数/金币(带审计)")
@@ -43,21 +45,26 @@ def update_config(
     request: Request,
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
+    scene: GuideScene = "coupon",
 ) -> GuideVideoConfigOut:
-    before, after = guide_video.update_config(
-        db,
-        enabled=body.enabled,
-        max_plays=body.max_plays,
-        reward_coin=body.reward_coin,
-        admin_id=admin.id,
-        commit=False,
-    )
+    try:
+        before, after = guide_video.update_config(
+            db,
+            enabled=body.enabled,
+            max_plays=body.max_plays,
+            reward_coin=body.reward_coin,
+            scene=scene,
+            admin_id=admin.id,
+            commit=False,
+        )
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
     write_audit(
         db, admin, action="guide_video.update", target_type="guide_video", target_id=None,
-        detail={"before": before, "after": after}, ip=get_client_ip(request), commit=False,
+        detail={"scene": scene, "before": before, "after": after}, ip=get_client_ip(request), commit=False,
     )
     db.commit()
-    return _out(db)
+    return _out(db, scene)
 
 
 @router.post("/video", response_model=GuideVideoConfigOut, summary="上传新手引导视频(MP4,带审计)")
@@ -65,23 +72,40 @@ async def upload_video(
     request: Request,
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
-    file: UploadFile = File(...),
+    file: Annotated[UploadFile, File()],
+    scene: GuideScene = "coupon",
 ) -> GuideVideoConfigOut:
     data = await file.read()
+    url: str | None = None
     try:
-        url = media.save_guide_video(data)
+        url, analysis = media.save_guide_video(data)
     except media.MediaError as e:
         raise HTTPException(status_code=400, detail=str(e)) from e
-    before, after = guide_video.set_video(db, url, admin_id=admin.id, commit=False)
-    write_audit(
-        db, admin, action="guide_video.set_video", target_type="guide_video", target_id=None,
-        detail={"before": before.get("video_url"), "after": url, "bytes": len(data)},
-        ip=get_client_ip(request), commit=False,
-    )
-    db.commit()
-    # 提交成功后再删旧片,避免新片没落库就把旧片丢了
-    media.delete_guide_video(before.get("video_url"))
-    return _out(db)
+    try:
+        before, after = guide_video.set_video(
+            db, url, analysis=analysis, scene=scene, admin_id=admin.id, commit=False
+        )
+        write_audit(
+            db, admin, action="guide_video.set_video", target_type="guide_video", target_id=None,
+            detail={
+                "scene": scene,
+                "before": before.get("video_url"),
+                "after": url,
+                "bytes": len(data),
+                "duration_ms": analysis["duration_ms"],
+                "video_codec": analysis["video_codec"],
+                "audio_codec": analysis["audio_codec"],
+            },
+            ip=get_client_ip(request), commit=False,
+        )
+        db.commit()
+    except Exception:
+        db.rollback()
+        media.delete_guide_video(url)
+        raise
+    # 旧文件仍可能被 prepare/已起播快照引用，不能在这里立即删除。
+    guide_video.cleanup_old_videos(db)
+    return _out(db, scene)
 
 
 @router.delete("/video", response_model=GuideVideoConfigOut, summary="移除新手引导视频(带审计)")
@@ -89,13 +113,17 @@ def delete_video(
     request: Request,
     admin: Annotated[AdminUser, Depends(require_role("operator"))],
     db: AdminDb,
+    scene: GuideScene = "coupon",
 ) -> GuideVideoConfigOut:
     """移除后 /guide-video/start 一律返回 should_play=false,领券浮层回到「只放广告」。"""
-    before, after = guide_video.set_video(db, None, admin_id=admin.id, commit=False)
+    before, after = guide_video.set_video(
+        db, None, scene=scene, admin_id=admin.id, commit=False
+    )
     write_audit(
         db, admin, action="guide_video.delete_video", target_type="guide_video", target_id=None,
-        detail={"before": before.get("video_url")}, ip=get_client_ip(request), commit=False,
+        detail={"scene": scene, "before": before.get("video_url")}, ip=get_client_ip(request), commit=False,
     )
     db.commit()
-    media.delete_guide_video(before.get("video_url"))
-    return _out(db)
+    # 旧文件由保护期清理任务处理，避免破坏有效计划与已起播快照。
+    guide_video.cleanup_old_videos(db)
+    return _out(db, scene)

app/admin/routers/limit_whitelist.py

@@ -0,0 +1,838 @@
+"""Unified limit rules and per-phone/device whitelist overrides."""
+
+from __future__ import annotations
+
+from fastapi import APIRouter, Depends, HTTPException, Query, Request, status
+
+from app.admin.audit import write_audit
+from app.admin.deps import AdminDb, CurrentAdmin, get_client_ip, require_page
+from app.admin.repositories import limit_whitelist as repo
+from app.admin.schemas.limit_whitelist import (
+    DeviceCandidateOut,
+    GlobalLimitUpdate,
+    LimitOverrideBulkWrite,
+    LimitOverrideList,
+    LimitOverrideOut,
+    LimitOverridePatch,
+    LimitOverrideWrite,
+    LimitRuleOut,
+    LimitSubjectEnabledPatch,
+    LimitSubjectList,
+    LimitSubjectOut,
+)
+from app.core import limit_policy
+from app.models.limit_policy import LimitPolicyOverride
+from app.repositories import risk as risk_repo
+
+router = APIRouter(
+    prefix="/admin/api/limit-whitelist",
+    tags=["admin-limit-whitelist"],
+    dependencies=[Depends(require_page("limit-whitelist"))],
+)
+
+_WHITELIST_MODES = {
+    limit_policy.MODE_UNLIMITED,
+    limit_policy.MODE_SUPPRESS_ALERT,
+}
+
+
+def _subject_whitelist_rows(
+    db,
+    *,
+    subject_type: str,
+    subject_value: str,
+) -> list[LimitPolicyOverride]:
+    return [
+        row
+        for row in repo.rows_for_subject(
+            db,
+            subject_type=subject_type,
+            subject_value=subject_value,
+        )
+        if row.mode in _WHITELIST_MODES
+    ]
+
+
+def _reconcile_risk_rule(db, rule_code: str) -> None:
+    now = risk_repo.utcnow()
+    if rule_code == "risk.sms.hourly":
+        risk_repo.reconcile_behavior_rule(
+            db,
+            rule_code=risk_repo.RULE_SMS_HOURLY,
+            at=now,
+            commit=False,
+        )
+    elif rule_code == "risk.oneclick.daily":
+        risk_repo.reconcile_behavior_rule(
+            db,
+            rule_code=risk_repo.RULE_ONECLICK_DAILY,
+            at=now,
+            commit=False,
+        )
+    elif rule_code == "risk.compare.daily":
+        risk_repo.reconcile_compare_rule(db, at=now, commit=False)
+
+
+def _row_or_404(db, override_id: int) -> LimitPolicyOverride:
+    row = db.get(LimitPolicyOverride, override_id, populate_existing=True)
+    if row is None:
+        raise HTTPException(status_code=404, detail="白名单配置不存在")
+    return row
+
+
+def _out(db, row: LimitPolicyOverride) -> LimitOverrideOut:
+    return LimitOverrideOut(**repo.to_dict(db, row))
+
+
+def _audit_payload(value: LimitOverrideOut) -> dict:
+    return value.model_dump(mode="json")
+
+
+def _subject_out(
+    db,
+    *,
+    subject_type: str,
+    subject_value: str,
+    rows: list[LimitPolicyOverride],
+    created_at,
+) -> LimitSubjectOut:
+    items = [_out(db, row) for row in rows]
+    group_counts: dict[str, int] = {}
+    for item in items:
+        group_counts[item.rule_group] = group_counts.get(item.rule_group, 0) + 1
+    return LimitSubjectOut(
+        subject_type=subject_type,
+        subject_value=subject_value,
+        group_counts=group_counts,
+        total_rules=len(items),
+        items=items,
+        created_at=created_at,
+        updated_at=max(
+            (item.updated_at for item in items),
+            default=created_at,
+        ),
+    )
+
+
+@router.get("/rules", response_model=list[LimitRuleOut], summary="读取所有限制规则")
+def list_rules(db: AdminDb) -> list[LimitRuleOut]:
+    return [LimitRuleOut(**item) for item in limit_policy.rule_catalog(db)]
+
+
+@router.patch(
+    "/rules/{rule_code}",
+    response_model=LimitRuleOut,
+    summary="修改规则的全局限制值",
+)
+def update_rule(
+    rule_code: str,
+    body: GlobalLimitUpdate,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitRuleOut:
+    try:
+        before, after = repo.update_global_limit(db, rule_code, body.value, admin_id=admin.id)
+    except KeyError as exc:
+        raise HTTPException(status_code=404, detail="未知限制规则") from exc
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+    _reconcile_risk_rule(db, rule_code)
+    write_audit(
+        db,
+        admin,
+        action="limit.rule.update",
+        target_type="limit_rule",
+        target_id=rule_code,
+        detail={"before": before, "after": after},
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    item = next(item for item in limit_policy.rule_catalog(db) if item["code"] == rule_code)
+    return LimitRuleOut(**item)
+
+
+@router.get(
+    "/device-candidates",
+    response_model=list[DeviceCandidateOut],
+    summary="按限制项搜索可加入白名单的设备",
+)
+def list_device_candidates(
+    db: AdminDb,
+    rule_code: str = Query(..., min_length=1, max_length=64),
+    keyword: str | None = Query(None, max_length=128),
+    limit: int = Query(30, ge=1, le=50),
+) -> list[DeviceCandidateOut]:
+    try:
+        rows = repo.list_device_candidates(
+            db,
+            rule_code=rule_code,
+            keyword=keyword,
+            limit=limit,
+        )
+    except KeyError as exc:
+        raise HTTPException(status_code=404, detail="未知限制规则") from exc
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+    return [DeviceCandidateOut(**row) for row in rows]
+
+
+@router.get("", response_model=LimitOverrideList, summary="读取白名单配置")
+def list_overrides(
+    db: AdminDb,
+    subject_type: str | None = Query(None, pattern="^(phone|device)$"),
+    keyword: str | None = Query(None, max_length=128),
+    rule_code: str | None = Query(None, max_length=64),
+    offset: int = Query(0, ge=0),
+    limit: int = Query(100, ge=1, le=500),
+) -> LimitOverrideList:
+    rows, total = repo.list_rows(
+        db,
+        subject_type=subject_type,
+        keyword=keyword,
+        rule_code=rule_code,
+        offset=offset,
+        limit=limit,
+    )
+    return LimitOverrideList(items=[_out(db, row) for row in rows], total=total)
+
+
+@router.get(
+    "/subjects",
+    response_model=LimitSubjectList,
+    summary="按手机号或设备聚合读取白名单配置",
+)
+def list_override_subjects(
+    db: AdminDb,
+    subject_type: str | None = Query(None, pattern="^(phone|device)$"),
+    keyword: str | None = Query(None, max_length=128),
+    rule_code: str | None = Query(None, max_length=64),
+    offset: int = Query(0, ge=0),
+    limit: int = Query(10, ge=1, le=100),
+) -> LimitSubjectList:
+    subjects, total = repo.list_subject_rows(
+        db,
+        subject_type=subject_type,
+        keyword=keyword,
+        rule_code=rule_code,
+        offset=offset,
+        limit=limit,
+    )
+    return LimitSubjectList(
+        items=[
+            _subject_out(
+                db,
+                subject_type=item_subject_type,
+                subject_value=subject_value,
+                rows=rows,
+                created_at=created_at,
+            )
+            for item_subject_type, subject_value, rows, created_at in subjects
+        ],
+        total=total,
+    )
+
+
+@router.patch(
+    "/subjects/enabled",
+    response_model=LimitSubjectOut,
+    summary="整体启用或停用一个手机号或设备的白名单",
+)
+def set_override_subject_enabled(
+    body: LimitSubjectEnabledPatch,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitSubjectOut:
+    try:
+        subject_value = limit_policy.validate_whitelist_subject(
+            body.subject_type, body.subject_value
+        )
+        repo.lock_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        rows = _subject_whitelist_rows(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        if not rows:
+            raise HTTPException(status_code=404, detail="白名单主体不存在")
+        before = [_audit_payload(_out(db, row)) for row in rows]
+        for row in rows:
+            repo.update(
+                db,
+                row,
+                enabled=body.enabled,
+                starts_at=None,
+                expires_at=None,
+                reason=None,
+                fields_set={"enabled"},
+            )
+        for rule_code in {row.rule_code for row in rows}:
+            _reconcile_risk_rule(db, rule_code)
+        after = [_audit_payload(_out(db, row)) for row in rows]
+        write_audit(
+            db,
+            admin,
+            action="limit.override.subject_enabled",
+            target_type="limit_override_subject",
+            target_id=f"{body.subject_type}:{subject_value}",
+            detail={"before": before, "after": after},
+            ip=get_client_ip(request),
+            commit=False,
+        )
+        db.commit()
+    except HTTPException:
+        raise
+    except (KeyError, ValueError) as exc:
+        db.rollback()
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    rows = [
+        row
+        for row in repo.rows_for_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        if row.mode
+        in {
+            limit_policy.MODE_UNLIMITED,
+            limit_policy.MODE_SUPPRESS_ALERT,
+        }
+    ]
+    return _subject_out(
+        db,
+        subject_type=body.subject_type,
+        subject_value=subject_value,
+        rows=rows,
+        created_at=min(row.created_at for row in rows),
+    )
+
+
+@router.put(
+    "/subjects",
+    response_model=LimitSubjectOut,
+    summary="整体更新一个手机号或设备的白名单限制项",
+)
+def replace_override_subject(
+    body: LimitOverrideBulkWrite,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitSubjectOut:
+    current_rule_code = ""
+    try:
+        subject_value = limit_policy.validate_whitelist_subject(
+            body.subject_type, body.subject_value
+        )
+        repo.lock_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        existing_rows = repo.rows_for_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        subject_created_at = min(
+            (row.created_at for row in existing_rows),
+            default=None,
+        )
+        before = [_audit_payload(_out(db, row)) for row in existing_rows]
+        existing_by_rule = {row.rule_code: row for row in existing_rows}
+        selected_codes = set(body.rule_codes)
+        touched_rule_codes = set(selected_codes)
+
+        for current_rule_code in body.rule_codes:
+            rule = limit_policy.get_rule(current_rule_code)
+            mode = (
+                limit_policy.MODE_SUPPRESS_ALERT if rule.alert_only else limit_policy.MODE_UNLIMITED
+            )
+            row = existing_by_rule.get(current_rule_code)
+            if row is None:
+                row = repo.create(
+                    db,
+                    subject_type=body.subject_type,
+                    subject_value=subject_value,
+                    rule_code=current_rule_code,
+                    mode=mode,
+                    limit_value=None,
+                    enabled=body.enabled,
+                    starts_at=body.starts_at,
+                    expires_at=body.expires_at,
+                    reason=body.reason,
+                    admin_id=admin.id,
+                )
+                # The table is ordered by the subject's original creation
+                # time.  If an edit replaces every rule, carry that timestamp
+                # to the new rows so the subject does not jump to the top.
+                if subject_created_at is not None:
+                    row.created_at = subject_created_at
+                continue
+
+            # Historical custom-value rows are converted to the only supported
+            # product modes when the administrator selects that rule again.
+            row.mode = mode
+            row.limit_value = None
+            row.reset_at = None
+            repo.update(
+                db,
+                row,
+                enabled=body.enabled,
+                starts_at=body.starts_at,
+                expires_at=body.expires_at,
+                reason=body.reason,
+                fields_set={"enabled", "starts_at", "expires_at", "reason"},
+            )
+
+        for row in existing_rows:
+            if row.rule_code not in selected_codes and row.mode in {
+                limit_policy.MODE_UNLIMITED,
+                limit_policy.MODE_SUPPRESS_ALERT,
+            }:
+                touched_rule_codes.add(row.rule_code)
+                db.delete(row)
+        db.flush()
+    except KeyError as exc:
+        raise HTTPException(status_code=404, detail="未知限制规则") from exc
+    except repo.DuplicateOverrideError as exc:
+        rule = limit_policy.get_rule(current_rule_code)
+        raise HTTPException(
+            status_code=409,
+            detail=f"“{rule.label}”已有白名单配置，请刷新后重试",
+        ) from exc
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    for rule_code in touched_rule_codes:
+        _reconcile_risk_rule(db, rule_code)
+    rows = [
+        row
+        for row in repo.rows_for_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        if row.mode
+        in {
+            limit_policy.MODE_UNLIMITED,
+            limit_policy.MODE_SUPPRESS_ALERT,
+        }
+    ]
+    after = [_audit_payload(_out(db, row)) for row in rows]
+    write_audit(
+        db,
+        admin,
+        action="limit.override.subject_replace",
+        target_type="limit_override_subject",
+        target_id=f"{body.subject_type}:{subject_value}",
+        detail={"before": before, "after": after},
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    rows = [
+        row
+        for row in repo.rows_for_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        if row.mode
+        in {
+            limit_policy.MODE_UNLIMITED,
+            limit_policy.MODE_SUPPRESS_ALERT,
+        }
+    ]
+    return _subject_out(
+        db,
+        subject_type=body.subject_type,
+        subject_value=subject_value,
+        rows=rows,
+        created_at=min(row.created_at for row in rows),
+    )
+
+
+@router.post(
+    "",
+    response_model=LimitOverrideOut,
+    status_code=status.HTTP_201_CREATED,
+    summary="新增白名单配置",
+)
+def create_override(
+    body: LimitOverrideWrite,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitOverrideOut:
+    before: list[dict] = []
+    touched_rows: list[LimitPolicyOverride] = []
+    try:
+        subject_value = limit_policy.validate_whitelist_subject(
+            body.subject_type, body.subject_value
+        )
+        repo.lock_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        existing_rows = _subject_whitelist_rows(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        before = [_audit_payload(_out(db, item)) for item in existing_rows]
+        row = repo.create(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+            rule_code=body.rule_code,
+            mode=body.mode,
+            enabled=body.enabled,
+            starts_at=body.starts_at,
+            expires_at=body.expires_at,
+            reason=body.reason,
+            limit_value=None,
+            admin_id=admin.id,
+        )
+        touched_rows = [*existing_rows, row]
+        if row.mode in _WHITELIST_MODES:
+            # Keep the legacy single-rule endpoint compatible without letting
+            # it create a second validity period for the same logical subject.
+            for existing in existing_rows:
+                repo.update(
+                    db,
+                    existing,
+                    enabled=body.enabled,
+                    starts_at=body.starts_at,
+                    expires_at=body.expires_at,
+                    reason=None,
+                    fields_set={"enabled", "starts_at", "expires_at"},
+                )
+    except KeyError as exc:
+        raise HTTPException(status_code=404, detail="未知限制规则") from exc
+    except repo.DuplicateOverrideError as exc:
+        raise HTTPException(
+            status_code=409,
+            detail="该手机号或设备已配置此规则，请编辑现有配置",
+        ) from exc
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+    for rule_code in {item.rule_code for item in touched_rows}:
+        _reconcile_risk_rule(db, rule_code)
+    after_rows = _subject_whitelist_rows(
+        db,
+        subject_type=row.subject_type,
+        subject_value=row.subject_value,
+    )
+    write_audit(
+        db,
+        admin,
+        action="limit.override.create",
+        target_type="limit_override_subject",
+        target_id=f"{row.subject_type}:{row.subject_value}",
+        detail={
+            "before": before,
+            "after": [_audit_payload(_out(db, item)) for item in after_rows],
+        },
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    return _out(db, row)
+
+
+@router.post(
+    "/bulk",
+    response_model=list[LimitOverrideOut],
+    status_code=status.HTTP_201_CREATED,
+    summary="批量新增临时不限或免告警白名单",
+)
+def create_overrides_bulk(
+    body: LimitOverrideBulkWrite,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> list[LimitOverrideOut]:
+    rows: list[LimitPolicyOverride] = []
+    before: list[dict] = []
+    current_rule_code = ""
+    try:
+        subject_value = limit_policy.validate_whitelist_subject(
+            body.subject_type, body.subject_value
+        )
+        repo.lock_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        existing_rows = repo.rows_for_subject(
+            db,
+            subject_type=body.subject_type,
+            subject_value=subject_value,
+        )
+        whitelist_modes = {
+            limit_policy.MODE_UNLIMITED,
+            limit_policy.MODE_SUPPRESS_ALERT,
+        }
+        existing_whitelist_rows = [row for row in existing_rows if row.mode in whitelist_modes]
+        before = [_audit_payload(_out(db, row)) for row in existing_whitelist_rows]
+        existing_by_rule = {row.rule_code: row for row in existing_rows}
+        selected_codes = set(body.rule_codes)
+        for current_rule_code in body.rule_codes:
+            rule = limit_policy.get_rule(current_rule_code)
+            mode = (
+                limit_policy.MODE_SUPPRESS_ALERT if rule.alert_only else limit_policy.MODE_UNLIMITED
+            )
+            row = existing_by_rule.get(current_rule_code)
+            if row is None:
+                row = repo.create(
+                    db,
+                    subject_type=body.subject_type,
+                    subject_value=subject_value,
+                    rule_code=current_rule_code,
+                    mode=mode,
+                    limit_value=None,
+                    enabled=body.enabled,
+                    starts_at=body.starts_at,
+                    expires_at=body.expires_at,
+                    reason=body.reason,
+                    admin_id=admin.id,
+                )
+            else:
+                row.mode = mode
+                row.limit_value = None
+                row.reset_at = None
+                repo.update(
+                    db,
+                    row,
+                    enabled=body.enabled,
+                    starts_at=body.starts_at,
+                    expires_at=body.expires_at,
+                    reason=body.reason,
+                    fields_set={
+                        "enabled",
+                        "starts_at",
+                        "expires_at",
+                        "reason",
+                    },
+                )
+
+        # 同一手机号或设备在产品上是一条白名单。追加限制项时保留原规则，
+        # 但统一使用最后一次配置的启用状态与有效期，避免一个主体出现多套时间。
+        rows = [
+            row
+            for row in repo.rows_for_subject(
+                db,
+                subject_type=body.subject_type,
+                subject_value=subject_value,
+            )
+            if row.mode in whitelist_modes
+        ]
+        for row in rows:
+            if row.rule_code in selected_codes:
+                continue
+            repo.update(
+                db,
+                row,
+                enabled=body.enabled,
+                starts_at=body.starts_at,
+                expires_at=body.expires_at,
+                reason=None,
+                fields_set={"enabled", "starts_at", "expires_at"},
+            )
+    except KeyError as exc:
+        raise HTTPException(status_code=404, detail="未知限制规则") from exc
+    except repo.DuplicateOverrideError as exc:
+        rule = limit_policy.get_rule(current_rule_code)
+        raise HTTPException(
+            status_code=409,
+            detail=f"“{rule.label}”白名单配置发生并发更新，请刷新后重试",
+        ) from exc
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    for row in rows:
+        _reconcile_risk_rule(db, row.rule_code)
+    results = [_out(db, row) for row in rows]
+    write_audit(
+        db,
+        admin,
+        action="limit.override.bulk_create",
+        target_type="limit_override",
+        target_id=",".join(str(row.id) for row in rows),
+        detail={
+            "before": before,
+            "after": [_audit_payload(item) for item in results],
+        },
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    return [_out(db, row) for row in rows]
+
+
+@router.patch(
+    "/{override_id}",
+    response_model=LimitOverrideOut,
+    summary="编辑白名单配置",
+)
+def update_override(
+    override_id: int,
+    body: LimitOverridePatch,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitOverrideOut:
+    row = _row_or_404(db, override_id)
+    repo.lock_subject(
+        db,
+        subject_type=row.subject_type,
+        subject_value=row.subject_value,
+    )
+    row = _row_or_404(db, override_id)
+    subject_rows = (
+        _subject_whitelist_rows(
+            db,
+            subject_type=row.subject_type,
+            subject_value=row.subject_value,
+        )
+        if row.mode in _WHITELIST_MODES
+        else [row]
+    )
+    before = [_audit_payload(_out(db, item)) for item in subject_rows]
+    fields_set = set(body.model_fields_set)
+    try:
+        period_fields = {"enabled", "starts_at", "expires_at"}
+        if row.mode in _WHITELIST_MODES and fields_set & period_fields:
+            new_enabled = body.enabled if "enabled" in fields_set else row.enabled
+            new_starts = body.starts_at if "starts_at" in fields_set else row.starts_at
+            new_expires = body.expires_at if "expires_at" in fields_set else row.expires_at
+            for item in subject_rows:
+                item_fields = set(period_fields)
+                if item.id == row.id and "reason" in fields_set:
+                    item_fields.add("reason")
+                repo.update(
+                    db,
+                    item,
+                    enabled=new_enabled,
+                    starts_at=new_starts,
+                    expires_at=new_expires,
+                    reason=body.reason if item.id == row.id else None,
+                    fields_set=item_fields,
+                )
+        else:
+            repo.update(
+                db,
+                row,
+                **body.model_dump(),
+                fields_set=fields_set,
+            )
+    except (KeyError, ValueError) as exc:
+        db.rollback()
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+    for rule_code in {item.rule_code for item in subject_rows}:
+        _reconcile_risk_rule(db, rule_code)
+    after_rows = [_audit_payload(_out(db, item)) for item in subject_rows]
+    write_audit(
+        db,
+        admin,
+        action="limit.override.update",
+        target_type="limit_override_subject",
+        target_id=f"{row.subject_type}:{row.subject_value}",
+        detail={"before": before, "after": after_rows},
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    return _out(db, row)
+
+
+@router.post(
+    "/{override_id}/reset",
+    response_model=LimitOverrideOut,
+    summary="停用例外配置并恢复全局策略",
+)
+def restore_global_policy(
+    override_id: int,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> LimitOverrideOut:
+    row = _row_or_404(db, override_id)
+    repo.lock_subject(
+        db,
+        subject_type=row.subject_type,
+        subject_value=row.subject_value,
+    )
+    row = _row_or_404(db, override_id)
+    subject_rows = (
+        _subject_whitelist_rows(
+            db,
+            subject_type=row.subject_type,
+            subject_value=row.subject_value,
+        )
+        if row.mode in _WHITELIST_MODES
+        else [row]
+    )
+    before = [_audit_payload(_out(db, item)) for item in subject_rows]
+    for item in subject_rows:
+        repo.restore_global(item)
+    for rule_code in {item.rule_code for item in subject_rows}:
+        _reconcile_risk_rule(db, rule_code)
+    after = [_audit_payload(_out(db, item)) for item in subject_rows]
+    write_audit(
+        db,
+        admin,
+        action="limit.override.restore_global",
+        target_type="limit_override_subject",
+        target_id=f"{row.subject_type}:{row.subject_value}",
+        detail={"before": before, "after": after},
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()
+    return _out(db, row)
+
+
+@router.delete(
+    "/{override_id}",
+    status_code=status.HTTP_204_NO_CONTENT,
+    summary="删除白名单配置",
+)
+def delete_override(
+    override_id: int,
+    request: Request,
+    admin: CurrentAdmin,
+    db: AdminDb,
+) -> None:
+    row = _row_or_404(db, override_id)
+    repo.lock_subject(
+        db,
+        subject_type=row.subject_type,
+        subject_value=row.subject_value,
+    )
+    row = _row_or_404(db, override_id)
+    before = _audit_payload(_out(db, row))
+    target_id = str(row.id)
+    rule_code = row.rule_code
+    db.delete(row)
+    db.flush()
+    _reconcile_risk_rule(db, rule_code)
+    write_audit(
+        db,
+        admin,
+        action="limit.override.delete",
+        target_type="limit_override",
+        target_id=target_id,
+        detail={"before": before},
+        ip=get_client_ip(request),
+        commit=False,
+    )
+    db.commit()

app/admin/routers/risk_monitor.py

@@ -23,13 +23,8 @@ from app.admin.schemas.risk_monitor import (
     RiskResetResponse,
     RiskRuleConfig,
 )
-from app.core.config_schema import (
-    RISK_COMPARE_DAILY_THRESHOLD_KEY,
-    RISK_ONECLICK_DAILY_THRESHOLD_KEY,
-    RISK_SMS_HOURLY_THRESHOLD_KEY,
-)
+from app.core import limit_policy
 from app.models.risk import RiskIncident, SubjectRestriction
-from app.repositories import app_config
 from app.repositories import risk as risk_repo
 
 router = APIRouter(
@@ -83,15 +78,16 @@ def update_rules(
 ) -> RiskRuleConfig:
     before = _rule_config(db).model_dump()
     after = body.model_dump()
-    values = (
-        (RISK_SMS_HOURLY_THRESHOLD_KEY, body.sms_hourly_threshold),
-        (RISK_ONECLICK_DAILY_THRESHOLD_KEY, body.oneclick_daily_threshold),
-        (RISK_COMPARE_DAILY_THRESHOLD_KEY, body.compare_daily_threshold),
+    limit_policy.set_global_limits(
+        db,
+        {
+            "risk.sms.hourly": body.sms_hourly_threshold,
+            "risk.oneclick.daily": body.oneclick_daily_threshold,
+            "risk.compare.daily": body.compare_daily_threshold,
+        },
+        admin_id=admin.id,
+        commit=False,
     )
-    for key, value in values:
-        app_config.set_value(
-            db, key, value, admin_id=admin.id, commit=False
-        )
 
     now = risk_repo.utcnow()
     risk_repo.reconcile_behavior_rule(

app/admin/routers/users.py

@@ -88,6 +88,11 @@ def get_user_reward_stats(
     withdraw_source: Annotated[
         str | None, Query(pattern="^(coin_cash|invite_cash)$")
     ] = None,
+    app_env: Annotated[str | None, Query(pattern="^(prod|test)$")] = None,
+    revenue_scope: Annotated[str, Query(pattern="^(business|all)$")] = "all",
+    feed_scene: Annotated[
+        str | None, Query(pattern="^(comparison|coupon|welfare)$")
+    ] = None,
 ) -> UserRewardStats:
     """提现详情抽屉「用户统计区」。date_from/date_to 都不传 = 注册至今(全量)。"""
     if not user_repo.user_exists(db, user_id):
@@ -99,6 +104,9 @@ def get_user_reward_stats(
             date_from=date_from,
             date_to=date_to,
             withdraw_source=withdraw_source,
+            app_env=app_env,
+            revenue_scope=revenue_scope,
+            feed_scene=feed_scene,
         )
     )
 

app/admin/schemas/guide_video.py

@@ -1,25 +1,44 @@
-"""admin 新手引导视频配置 schemas(开关 / 视频地址 / 前几次 / 每次金币)。"""
+"""后台引导视频配置协议。"""
 from __future__ import annotations
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, field_validator
 
-from app.repositories.guide_video import MAX_PLAYS_LIMIT, REWARD_COIN_LIMIT
+from app.repositories.guide_video import (
+    MAX_PLAYS_LIMIT,
+    MIN_PLAYS,
+    MIN_REWARD_COIN,
+    REWARD_COIN_LIMIT,
+)
 
 
 class GuideVideoConfigOut(BaseModel):
+    scene: str
     enabled: bool
-    video_url: str | None = None   # 相对地址 /media/guide_video/xxx.mp4;未配片 = None
+    video_url: str | None = None
     max_plays: int
     reward_coin: int
+    duration_ms: int | None = None
+    circle_count: int = 10
+    circle_duration_ms: float | None = None
+    reward_per_circle: int
+    video_codec: str | None = None
+    audio_codec: str | None = None
+    analysis_status: str
+    analysis_error: str | None = None
+    config_version: int
     updated_at: str | None = None
-    # 只读统计,后台展示用:已有多少次播放、其中已发币多少次。
-    total_plays: int = 0
-    granted_plays: int = 0
 
 
 class GuideVideoConfigUpdate(BaseModel):
-    """部分更新:只改传入(非 None)字段。视频文件走 /video 上传接口。"""
-
     enabled: bool | None = None
-    max_plays: int | None = Field(default=None, ge=0, le=MAX_PLAYS_LIMIT)
-    reward_coin: int | None = Field(default=None, ge=0, le=REWARD_COIN_LIMIT)
+    max_plays: int | None = Field(default=None, ge=MIN_PLAYS, le=MAX_PLAYS_LIMIT)
+    reward_coin: int | None = Field(
+        default=None, ge=MIN_REWARD_COIN, le=REWARD_COIN_LIMIT
+    )
+
+    @field_validator("reward_coin")
+    @classmethod
+    def reward_must_be_multiple_of_ten(cls, value: int | None) -> int | None:
+        if value is not None and value % 10:
+            raise ValueError("金币总价必须是 10 的倍数")
+        return value

app/admin/schemas/limit_whitelist.py

@@ -0,0 +1,150 @@
+"""Admin contracts for global limit rules and per-subject overrides."""
+from __future__ import annotations
+
+from datetime import UTC, datetime
+from typing import Literal
+
+from pydantic import BaseModel, ConfigDict, Field, model_validator
+
+SubjectType = Literal["phone", "device"]
+PolicyMode = Literal["unlimited", "suppress_alert"]
+
+
+class LimitRuleOut(BaseModel):
+    code: str
+    label: str
+    group: str
+    global_limit: int
+    default_limit: int
+    window_label: str
+    subject_types: list[str]
+    allowed_modes: list[str]
+    min_value: int
+    max_value: int
+    supports_reset: bool
+    alert_only: bool
+
+
+class GlobalLimitUpdate(BaseModel):
+    value: int = Field(ge=0, le=1_000_000)
+
+
+class DeviceCandidateOut(BaseModel):
+    device_id: str
+    source: str
+    source_label: str
+    user_id: int | None = None
+    username: str | None = None
+    phone: str | None = None
+    nickname: str | None = None
+    device_model: str | None = None
+    last_active_at: datetime
+
+
+class LimitOverrideWrite(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    subject_type: SubjectType
+    subject_value: str = Field(min_length=1, max_length=128)
+    rule_code: str = Field(min_length=1, max_length=64)
+    mode: PolicyMode
+    enabled: bool = True
+    starts_at: datetime | None = None
+    expires_at: datetime | None = None
+    reason: str = Field("", max_length=256)
+
+    @model_validator(mode="after")
+    def validate_time_range(self):
+        if self.starts_at and self.expires_at and self.expires_at <= self.starts_at:
+            raise ValueError("失效时间必须晚于生效时间")
+        return self
+
+
+class LimitOverrideBulkWrite(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    subject_type: SubjectType
+    subject_value: str = Field(min_length=1, max_length=128)
+    rule_codes: list[str] = Field(min_length=1, max_length=32)
+    enabled: bool = True
+    starts_at: datetime | None = None
+    expires_at: datetime
+    reason: str = Field("", max_length=256)
+
+    @model_validator(mode="after")
+    def validate_bulk_request(self):
+        self.rule_codes = list(dict.fromkeys(self.rule_codes))
+        starts_at = (
+            self.starts_at.replace(tzinfo=UTC)
+            if self.starts_at and self.starts_at.tzinfo is None
+            else self.starts_at
+        )
+        expires_at = (
+            self.expires_at.replace(tzinfo=UTC)
+            if self.expires_at.tzinfo is None
+            else self.expires_at
+        )
+        if starts_at and expires_at <= starts_at:
+            raise ValueError("失效时间必须晚于生效时间")
+        if expires_at <= datetime.now(UTC):
+            raise ValueError("失效时间必须晚于当前时间")
+        return self
+
+
+class LimitOverridePatch(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    enabled: bool | None = None
+    starts_at: datetime | None = None
+    expires_at: datetime | None = None
+    reason: str | None = Field(None, max_length=256)
+
+
+class LimitSubjectEnabledPatch(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    subject_type: SubjectType
+    subject_value: str = Field(min_length=1, max_length=128)
+    enabled: bool
+
+
+class LimitOverrideOut(BaseModel):
+    id: int
+    subject_type: str
+    subject_value: str
+    rule_code: str
+    rule_label: str
+    rule_group: str
+    mode: str
+    limit_value: int | None
+    global_limit: int
+    effective_limit: int | None
+    enabled: bool
+    starts_at: datetime | None
+    expires_at: datetime | None
+    reset_at: datetime | None
+    reason: str | None
+    status: str
+    created_by_admin_id: int | None
+    created_at: datetime
+    updated_at: datetime
+
+
+class LimitOverrideList(BaseModel):
+    items: list[LimitOverrideOut]
+    total: int
+
+
+class LimitSubjectOut(BaseModel):
+    subject_type: str
+    subject_value: str
+    group_counts: dict[str, int]
+    total_rules: int
+    items: list[LimitOverrideOut]
+    created_at: datetime
+    updated_at: datetime
+
+
+class LimitSubjectList(BaseModel):
+    items: list[LimitSubjectOut]
+    total: int

app/admin/schemas/risk_monitor.py

@@ -21,9 +21,9 @@ class RiskMonitorSummary(BaseModel):
 
 
 class RiskRuleConfig(BaseModel):
-    sms_hourly_threshold: int = Field(ge=1, le=5)
+    sms_hourly_threshold: int = Field(ge=1, le=100_000)
     oneclick_daily_threshold: int = Field(ge=1, le=100_000)
-    compare_daily_threshold: int = Field(ge=1, le=100)
+    compare_daily_threshold: int = Field(ge=1, le=100_000)
 
 
 class RiskIncidentItem(BaseModel):

app/admin/schemas/user.py

@@ -60,7 +60,7 @@ class UserRewardStats(BaseModel):
     reward_video_avg_ecpm: float  # 平均激励视频 eCPM(分/千次)
     reward_video_cash_cents: int  # 激励视频提现(金币折现)
     feed_count: int  # 累计信息流广告数(granted 份数,unit_count 累加)
-    feed_avg_ecpm: float  # 平均信息流广告 eCPM(分/千次)
+    feed_avg_ecpm: float  # 全部 Draw/feed 实际展示的平均 eCPM(分/千次,含未发奖展示)
     feed_cash_cents: int  # 信息流广告提现(金币折现)
 
 

app/api/v1/ad.py

@@ -18,7 +18,7 @@ import uuid
 from fastapi import APIRouter, Depends, HTTPException, Path, Request, status
 
 from app.api.deps import CurrentUser, DbSession
-from app.core import rewards
+from app.core import limit_policy, rewards
 from app.core.config import settings
 from app.core.ratelimit import rate_limit
 from app.integrations import pangle
@@ -413,12 +413,23 @@ def feed_reward(payload: FeedRewardIn, user: CurrentUser, db: DbSession) -> Feed
         "feed ad reward user_id=%d event=%s status=%s units=%d coin=%d",
         user.id, rec.client_event_id, rec.status, rec.unit_count, rec.coin,
     )
+    feed_policy = limit_policy.resolve_for_user(db, "ad.feed.daily", user.id)
+    feed_limit = (
+        rewards.get_ad_daily_limit(db)
+        if feed_policy.override_id is None
+        and feed_policy.bucket_version == "default"
+        else feed_policy.limit
+    )
     return FeedRewardOut(
         granted=(rec.status == "granted"),
         status=rec.status,
         coin=rec.coin,
         unit_count=rec.unit_count,
-        daily_limit=rewards.get_ad_daily_limit(db),
+        daily_limit=(
+            feed_limit
+            if feed_limit is not None
+            else limit_policy.get_rule("ad.feed.daily").max_value
+        ),
     )
 
 

app/api/v1/auth.py

@@ -16,7 +16,7 @@ from fastapi import APIRouter, HTTPException, Request
 from sqlalchemy.exc import IntegrityError
 
 from app.api.deps import CurrentUser, DbSession
-from app.core import test_account
+from app.core import limit_policy, test_account
 from app.core.ratelimit import (
     RateLimitRule,
     check_rate_limits,
@@ -69,6 +69,7 @@ SMS_LOGIN_MAX_PER_HOUR = 5
 # 堵「换手机号绕开单号 60s 冷却」的洞 —— 冷却是单号维度,一机换号能绕开。
 SMS_SEND_MAX_PER_HOUR_PER_DEVICE = 5     # 每小时上限
 SMS_SEND_MAX_PER_DAY_PER_DEVICE = 20     # 每天上限(再叠一层日封顶,挡低频长时间轰炸)
+UNLIMITED_VERIFY_ATTEMPTS = 2_147_483_647
 
 
 def _client_ip(request: Request) -> str:
@@ -198,16 +199,54 @@ def sms_send(req: SmsSendRequest, request: Request, db: DbSession) -> SmsSendRes
     # 补「换手机号绕开单号 60s 冷却」的洞(冷却是单号维度,一机换号能绕);设备维度按机器封顶,挡短信轰炸/烧钱。
     # 关键:被单号 60s 冷却挡下的重发是「没真发、没烧钱」→ 不该占额度。故 check(先判)放在真发之前
     # (超限直接 429、不真发),record(计数)只在 send_code 成功后调 —— 冷却/供应商失败抛 429 时直接返回、不计数。
+    hourly_policy = limit_policy.resolve(
+        db,
+        "sms.send.hourly",
+        phone=req.phone,
+        device=subject_id,
+    )
+    daily_policy = limit_policy.resolve(
+        db,
+        "sms.send.daily",
+        phone=req.phone,
+        device=subject_id,
+    )
+    cooldown_policy = limit_policy.resolve(
+        db,
+        "sms.phone.cooldown",
+        phone=req.phone,
+    )
+    hourly_limit = (
+        SMS_SEND_MAX_PER_HOUR_PER_DEVICE
+        if hourly_policy.override_id is None
+        and hourly_policy.bucket_version == "default"
+        else hourly_policy.limit
+    )
+    daily_limit = (
+        SMS_SEND_MAX_PER_DAY_PER_DEVICE
+        if daily_policy.override_id is None
+        and daily_policy.bucket_version == "default"
+        else daily_policy.limit
+    )
     send_rules = [
-        RateLimitRule("sms-send-device", SMS_SEND_MAX_PER_HOUR_PER_DEVICE, 3600,
-                      "操作过于频繁,请稍后再试"),
-        RateLimitRule("sms-send-device-daily", SMS_SEND_MAX_PER_DAY_PER_DEVICE, 86400,
-                      "今日验证码发送次数过多,请明天再试"),
+        RateLimitRule("sms-send-device", hourly_limit, 3600,
+                      "操作过于频繁,请稍后再试", hourly_policy.bucket_version),
+        RateLimitRule("sms-send-device-daily", daily_limit, 86400,
+                      "今日验证码发送次数过多,请明天再试", daily_policy.bucket_version),
     ]
-    check_rate_limits(request, subject=req.device_id, rules=send_rules)
+    check_rate_limits(request, subject=subject_id, rules=send_rules)
 
     try:
-        cooldown = send_code(req.phone)
+        send_result = send_code(
+            req.phone,
+            cooldown_sec=(
+                None
+                if cooldown_policy.override_id is None
+                and cooldown_policy.bucket_version == "default"
+                else (0 if cooldown_policy.limit is None else cooldown_policy.limit)
+            ),
+        )
+        cooldown = send_result.cooldown_sec
     except SmsError as e:
         risk_repo.record_behavior_event(
             db,
@@ -224,7 +263,7 @@ def sms_send(req: SmsSendRequest, request: Request, db: DbSession) -> SmsSendRes
         raise HTTPException(status_code=e.status_code, detail=str(e)) from e
 
     # 发码成功 → 两道闸各 +1(被单号冷却挡下的重发走不到这里,故不占额度)
-    record_rate_limits(request, subject=req.device_id, rules=send_rules)
+    record_rate_limits(request, subject=subject_id, rules=send_rules)
 
     from app.core.config import settings  # 局部 import 避免循环
 
@@ -238,7 +277,9 @@ def sms_send(req: SmsSendRequest, request: Request, db: DbSession) -> SmsSendRes
         phone=req.phone,
         client_ip=_client_ip(request),
         outcome="success",
-        details={"mock": settings.SMS_MOCK},
+        details={"mock": settings.SMS_MOCK,
+                 "provider": send_result.provider,
+                 "fallback": send_result.fallback},
         evaluate_rule=risk_repo.RULE_SMS_HOURLY,
     )
 
@@ -256,7 +297,7 @@ def sms_login(req: SmsLoginRequest, request: Request, db: DbSession) -> TokenWit
     # 放在最前面:命中即不校验验证码;先扣当日额度,超限直接拒,挡住有人猜到号后脚本刷。
     # 测试账号走自己的每日额度、不受下面 (设备+IP) 每小时限流约束(QA 需在一小时内反复登录联调)。
     if test_account.is_test_account(req.phone):
-        if not test_account.try_consume_quota():
+        if not test_account.try_consume_quota(req.phone):
             raise HTTPException(status_code=429, detail="测试账号今日使用次数已达上限,请明天再试")
         user = user_repo.upsert_user_for_login(db, phone=req.phone, register_channel="sms")
         risk_repo.record_behavior_event(
@@ -283,17 +324,48 @@ def sms_login(req: SmsLoginRequest, request: Request, db: DbSession) -> TokenWit
     # **之前** → 输错验证码的失败尝试也计数,才挡得住撞库/爆破(另有单码失败 SMS_MAX_VERIFY_ATTEMPTS 次即作废兜底)。
     # ⚠️ 按设备而非手机号 → 一台机器换不同手机号刷登录也受限(防一机狂登多号);device_id 空(老客户端)时
     #    退化为该 IP 下所有空设备聚一桶,仍受限。
+    login_policy = limit_policy.resolve(
+        db,
+        "sms.login.hourly",
+        phone=req.phone,
+        device=subject_id,
+    )
+    verify_policy = limit_policy.resolve(
+        db,
+        "sms.code.failed_attempts",
+        phone=req.phone,
+    )
+    login_limit = (
+        SMS_LOGIN_MAX_PER_HOUR
+        if login_policy.override_id is None
+        and login_policy.bucket_version == "default"
+        else login_policy.limit
+    )
     enforce_rate_limit(
         request,
         scope="sms-login-device",
-        subject=req.device_id,
-        limit=SMS_LOGIN_MAX_PER_HOUR,
+        subject=subject_id,
+        limit=login_limit,
         window_sec=3600,
         detail="登录尝试过于频繁,请稍后再试",
+        bucket_suffix=login_policy.bucket_version,
     )
 
     try:
-        ok = verify_code(req.phone, req.code)
+        ok = verify_code(
+            req.phone,
+            req.code,
+            max_failed_attempts=(
+                None
+                if verify_policy.override_id is None
+                and verify_policy.bucket_version == "default"
+                else (
+                    UNLIMITED_VERIFY_ATTEMPTS
+                    if verify_policy.limit is None
+                    else verify_policy.limit
+                )
+            ),
+        )
     except SmsError as e:  # provider 校验降级(如阿里云接口异常)→ 原样透出其状态码(503),别误报「验证码错误」
         raise HTTPException(status_code=e.status_code, detail=str(e)) from e
     if not ok:
@@ -394,15 +466,25 @@ def _finish_wechat_bind(
     未占用 → 新建微信账号(channel=wechat,昵称头像取微信)→ 签 token 登入。"""
     existing = user_repo.get_user_by_phone(db, phone)
     if existing is not None:
-        from app.core.config import settings  # 局部 import,避免循环
-
         ticket = create_conflict_ticket(
             openid=openid,
             wechat_nickname=wechat_nickname,
             wechat_avatar_url=wechat_avatar_url,
             phone=phone,
         )
-        blocked = rebind_repo.rebound_within_days(db, phone, settings.PHONE_REBIND_LIMIT_DAYS)
+        rebind_policy = limit_policy.resolve(
+            db,
+            "phone.rebind.days",
+            phone=phone,
+            device=device_id,
+        )
+        rebind_days = rebind_policy.limit or 0
+        blocked = rebind_repo.rebound_within_days(
+            db,
+            phone,
+            rebind_days,
+            reset_at=rebind_policy.reset_at,
+        )
         logger.info(
             "wechat bind phone occupied phone=%s by user_id=%d has_wechat=%s",
             mask_phone(phone), existing.id, bool(existing.wechat_openid),
@@ -418,7 +500,12 @@ def _finish_wechat_bind(
             conflict_ticket=ticket,
             rebind_available=not blocked,
             rebind_blocked_days=(
-                rebind_repo.remaining_block_days(db, phone, settings.PHONE_REBIND_LIMIT_DAYS)
+                rebind_repo.remaining_block_days(
+                    db,
+                    phone,
+                    rebind_days,
+                    reset_at=rebind_policy.reset_at,
+                )
                 if blocked else 0
             ),
         )
@@ -451,18 +538,50 @@ def wechat_bind_phone_sms(
     except TokenError as e:
         raise HTTPException(status_code=401, detail="授权已过期,请重新用微信登录") from e
 
+    subject_id = _device_subject(req.device_id, request)
     # 防刷:同 sms/login,按 设备+IP 每小时限流(放在验证码校验之前,失败也计数)
+    bind_policy = limit_policy.resolve(
+        db,
+        "wechat.bind.hourly",
+        phone=req.phone,
+        device=subject_id,
+    )
+    verify_policy = limit_policy.resolve(
+        db,
+        "sms.code.failed_attempts",
+        phone=req.phone,
+    )
+    bind_limit = (
+        SMS_LOGIN_MAX_PER_HOUR
+        if bind_policy.override_id is None
+        and bind_policy.bucket_version == "default"
+        else bind_policy.limit
+    )
     enforce_rate_limit(
         request,
         scope="wechat-bind-sms-device",
-        subject=req.device_id,
-        limit=SMS_LOGIN_MAX_PER_HOUR,
+        subject=subject_id,
+        limit=bind_limit,
         window_sec=3600,
         detail="登录尝试过于频繁,请稍后再试",
+        bucket_suffix=bind_policy.bucket_version,
     )
 
     try:
-        ok = verify_code(req.phone, req.code)
+        ok = verify_code(
+            req.phone,
+            req.code,
+            max_failed_attempts=(
+                None
+                if verify_policy.override_id is None
+                and verify_policy.bucket_version == "default"
+                else (
+                    UNLIMITED_VERIFY_ATTEMPTS
+                    if verify_policy.limit is None
+                    else verify_policy.limit
+                )
+            ),
+        )
     except SmsError as e:  # provider 校验降级(如阿里云接口异常)→ 原样透出其状态码(503),别误报「验证码错误」
         raise HTTPException(status_code=e.status_code, detail=str(e)) from e
     if not ok:
@@ -522,9 +641,23 @@ def wechat_conflict_continue(
     except TokenError as e:
         raise HTTPException(status_code=401, detail="操作超时,请重新用微信登录") from e
 
+    subject_id = _device_subject(req.device_id, request)
+    conflict_policy = limit_policy.resolve(
+        db,
+        "wechat.conflict.hourly",
+        phone=claims["phone"],
+        device=subject_id,
+    )
+    conflict_limit = (
+        SMS_LOGIN_MAX_PER_HOUR
+        if conflict_policy.override_id is None
+        and conflict_policy.bucket_version == "default"
+        else conflict_policy.limit
+    )
     enforce_rate_limit(
-        request, scope="wechat-conflict-device", subject=req.device_id,
-        limit=SMS_LOGIN_MAX_PER_HOUR, window_sec=3600, detail="操作过于频繁,请稍后再试",
+        request, scope="wechat-conflict-device", subject=subject_id,
+        limit=conflict_limit, window_sec=3600, detail="操作过于频繁,请稍后再试",
+        bucket_suffix=conflict_policy.bucket_version,
     )
 
     user = user_repo.get_user_by_phone(db, claims["phone"])
@@ -562,21 +695,50 @@ def wechat_conflict_continue(
 def wechat_conflict_rebind(
     req: WechatConflictRebindRequest, request: Request, db: DbSession
 ) -> WechatBindResultResponse:
-    from app.core.config import settings  # 局部 import,避免循环
-
     try:
         claims = decode_conflict_ticket(req.conflict_ticket)
     except TokenError as e:
         raise HTTPException(status_code=401, detail="操作超时,请重新用微信登录") from e
 
+    subject_id = _device_subject(req.device_id, request)
+    conflict_policy = limit_policy.resolve(
+        db,
+        "wechat.conflict.hourly",
+        phone=claims["phone"],
+        device=subject_id,
+    )
+    conflict_limit = (
+        SMS_LOGIN_MAX_PER_HOUR
+        if conflict_policy.override_id is None
+        and conflict_policy.bucket_version == "default"
+        else conflict_policy.limit
+    )
     enforce_rate_limit(
-        request, scope="wechat-conflict-device", subject=req.device_id,
-        limit=SMS_LOGIN_MAX_PER_HOUR, window_sec=3600, detail="操作过于频繁,请稍后再试",
+        request, scope="wechat-conflict-device", subject=subject_id,
+        limit=conflict_limit, window_sec=3600, detail="操作过于频繁,请稍后再试",
+        bucket_suffix=conflict_policy.bucket_version,
     )
 
     phone = claims["phone"]
-    if rebind_repo.rebound_within_days(db, phone, settings.PHONE_REBIND_LIMIT_DAYS):
-        days = rebind_repo.remaining_block_days(db, phone, settings.PHONE_REBIND_LIMIT_DAYS)
+    rebind_policy = limit_policy.resolve(
+        db,
+        "phone.rebind.days",
+        phone=phone,
+        device=req.device_id,
+    )
+    rebind_days = rebind_policy.limit or 0
+    if rebind_repo.rebound_within_days(
+        db,
+        phone,
+        rebind_days,
+        reset_at=rebind_policy.reset_at,
+    ):
+        days = rebind_repo.remaining_block_days(
+            db,
+            phone,
+            rebind_days,
+            reset_at=rebind_policy.reset_at,
+        )
         raise HTTPException(status_code=409, detail=f"该手机号 {days} 天内已换绑过,暂不能再次换绑")
 
     user = user_repo.rebind_account(

app/api/v1/compare_record.py

@@ -16,6 +16,7 @@ import logging
 from fastapi import APIRouter, BackgroundTasks, HTTPException, Query, status
 
 from app.api.deps import CurrentUser, DbSession
+from app.core import limit_policy
 from app.repositories import comparison as crud_compare
 from app.repositories import risk as risk_repo
 from app.schemas.compare_record import (
@@ -53,17 +54,29 @@ def reserve_compare_start(
     ):
         raise HTTPException(status_code=403, detail="账号存在异常，该功能暂不可用")
     try:
+        policy = limit_policy.resolve(
+            db,
+            "compare.start.daily",
+            phone=user.phone,
+            device=payload.device_id,
+        )
         rec, used = crud_compare.reserve_daily_start(
             db,
             user_id=user.id,
             trace_id=payload.trace_id,
             business_type=payload.business_type,
             device_id=payload.device_id,
+            limit=policy.limit,
+            reset_at=policy.reset_at,
         )
     except crud_compare.DailyCompareStartLimitExceeded:
         raise HTTPException(
             status_code=status.HTTP_429_TOO_MANY_REQUESTS,
-            detail="今日已比价超过100次，请明天再试",
+            detail=(
+                f"今日已比价超过{policy.limit}次，请明天再试"
+                if policy.limit is not None
+                else "今日比价次数已达上限，请明天再试"
+            ),
         ) from None
     except crud_compare.ComparisonTraceOwnershipError:
         raise HTTPException(
@@ -71,11 +84,16 @@ def reserve_compare_start(
             detail="比价任务标识冲突，请重新发起",
         ) from None
     # 风控阈值由后台动态配置，不能再只在固定的 100 次业务上限处同步。
-    risk_repo.sync_compare_incident(db, user_id=user.id, at=rec.created_at)
+    risk_repo.sync_compare_incident(
+        db,
+        user_id=user.id,
+        at=rec.created_at,
+        device_id=payload.device_id,
+    )
     return CompareStartReserveOut(
-        limit=crud_compare.DAILY_COMPARE_START_LIMIT,
+        limit=policy.limit,
         used=used,
-        remaining=max(crud_compare.DAILY_COMPARE_START_LIMIT - used, 0),
+        remaining=max(policy.limit - used, 0) if policy.limit is not None else None,
     )
 
 

app/api/v1/device.py

@@ -19,8 +19,10 @@ from app.api.deps import CurrentUser, DbSession
 from app.integrations import vendor_push
 from app.repositories import device as device_repo
 from app.schemas.device import (
+    DeferredDeviceUnregisterRequest,
     DeviceOut,
     DeviceRegisterRequest,
+    DeviceUnregisterRequest,
     HeartbeatRequest,
     LivenessAckRequest,
     LivenessOut,
@@ -78,6 +80,7 @@ def register_device(
         registration_id=req.registration_id,
         push_vendor=req.push_vendor,
         push_token=req.push_token,
+        push_binding_id=req.push_binding_id,
         platform=req.platform,
         app_version=req.app_version,
     )
@@ -106,6 +109,36 @@ def report_heartbeat(
         registration_id=req.registration_id,
         push_vendor=req.push_vendor,
         push_token=req.push_token,
+        push_binding_id=req.push_binding_id,
+    )
+    return OkResponse()
+
+
+@router.post("/unregister", response_model=OkResponse, summary="解绑当前用户的本机推送目标")
+def unregister_device(
+    req: DeviceUnregisterRequest,
+    user: CurrentUser,
+    db: DbSession,
+) -> OkResponse:
+    device_repo.unregister_push_binding(db, user_id=user.id, device_id=req.device_id)
+    logger.info("device unregister user_id=%d device_id=%s", user.id, req.device_id)
+    return OkResponse()
+
+
+@router.post(
+    "/unregister/deferred",
+    response_model=OkResponse,
+    summary="离线退出后按登录会话凭据幂等解绑",
+)
+def unregister_device_deferred(
+    req: DeferredDeviceUnregisterRequest,
+    db: DbSession,
+) -> OkResponse:
+    # 不返回是否命中，避免公开接口泄露设备绑定状态。
+    device_repo.unregister_push_binding_deferred(
+        db,
+        device_id=req.device_id,
+        push_binding_id=req.push_binding_id,
     )
     return OkResponse()
 

app/api/v1/guide_video.py

@@ -1,22 +1,16 @@
-"""新手引导视频(领券等候浮层前 N 次替代广告)。
-
-路由前缀 `/api/v1/guide-video`(均需 Bearer):
-  POST /start    这次浮层放引导视频还是放广告?命中则**当场计次**并下发 play_token
-  POST /reward   播完 / 中途关闭都调,按 play_token 幂等发固定金币
-
-发币额度以**服务端配置**为准(运营后台可改),客户端只报"播完/关闭",报不了金额,
-所以被破解也刷不到超额金币;次数上限由 guide_video_play 行数(按账号)硬卡。
-"""
+"""引导视频 prepare/start/reward 客户端 API。"""
 from __future__ import annotations
 
 import logging
 
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException
 
 from app.api.deps import CurrentUser, DbSession
 from app.core.ratelimit import rate_limit
 from app.repositories import guide_video as crud_guide
 from app.schemas.guide_video import (
+    GuideVideoPrepareIn,
+    GuideVideoPrepareOut,
     GuideVideoRewardIn,
     GuideVideoRewardOut,
     GuideVideoStartIn,
@@ -24,25 +18,41 @@ from app.schemas.guide_video import (
 )
 
 logger = logging.getLogger("shagua.guide_video")
-
 router = APIRouter(prefix="/api/v1/guide-video", tags=["guide-video"])
 
 
+@router.post(
+    "/prepare",
+    response_model=GuideVideoPrepareOut,
+    dependencies=[Depends(rate_limit(60, 60, "guide-video-prepare"))],
+)
+def prepare(
+    payload: GuideVideoPrepareIn, user: CurrentUser, db: DbSession
+) -> GuideVideoPrepareOut:
+    result = crud_guide.prepare_play(db, user.id, scene=payload.scene)
+    logger.info(
+        "guide video prepare user_id=%d scene=%s should_play=%s reason=%s",
+        user.id, payload.scene, result["should_play"], result["reason"],
+    )
+    return GuideVideoPrepareOut(**result)
+
+
 @router.post(
     "/start",
     response_model=GuideVideoStartOut,
-    summary="领券浮层是否放新手引导视频(命中即计次)",
     dependencies=[Depends(rate_limit(60, 60, "guide-video-start"))],
 )
-def start(payload: GuideVideoStartIn, user: CurrentUser, db: DbSession) -> GuideVideoStartOut:
-    """开播即计数:返回 should_play=True 时服务端已写下这一次,客户端必须真的播。
-
-    没配视频 / 开关关 / 次数用完 → should_play=False,客户端照旧走广告链路(行为不变)。
-    """
-    result = crud_guide.start_play(db, user.id, scene=payload.scene or "coupon")
+def start(
+    payload: GuideVideoStartIn, user: CurrentUser, db: DbSession
+) -> GuideVideoStartOut:
+    try:
+        result = crud_guide.start_play(db, user.id, play_token=payload.play_token)
+    except crud_guide.PlayStateError as exc:
+        status_code = 404 if exc.code == "play_not_found" else 409
+        raise HTTPException(status_code=status_code, detail=exc.detail()) from exc
     logger.info(
-        "guide video start user_id=%d scene=%s should_play=%s seq=%d remaining=%d",
-        user.id, payload.scene, result["should_play"], result["seq"], result["remaining"],
+        "guide video start user_id=%d token=%s status=%s seq=%d",
+        user.id, payload.play_token[:12], result["status"], result["seq"],
     )
     return GuideVideoStartOut(**result)
 
@@ -50,15 +60,20 @@ def start(payload: GuideVideoStartIn, user: CurrentUser, db: DbSession) -> Guide
 @router.post(
     "/reward",
     response_model=GuideVideoRewardOut,
-    summary="引导视频发金币(播完/中途关闭都发,play_token 幂等)",
-    dependencies=[Depends(rate_limit(60, 60, "guide-video-reward"))],
+    dependencies=[Depends(rate_limit(120, 60, "guide-video-reward"))],
 )
-def reward(payload: GuideVideoRewardIn, user: CurrentUser, db: DbSession) -> GuideVideoRewardOut:
-    result = crud_guide.grant_play(
-        db, user.id, play_token=payload.play_token, completed=payload.completed
+def reward(
+    payload: GuideVideoRewardIn, user: CurrentUser, db: DbSession
+) -> GuideVideoRewardOut:
+    result = crud_guide.grant_circle(
+        db,
+        user.id,
+        play_token=payload.play_token,
+        circle=payload.circle,
     )
     logger.info(
-        "guide video reward user_id=%d token=%s completed=%s granted=%s coin=%d",
-        user.id, payload.play_token[:12], payload.completed, result["granted"], result["coin"],
+        "guide video reward user_id=%d token=%s circle=%d status=%s granted=%s",
+        user.id, payload.play_token[:12], payload.circle,
+        result["status"], result["granted"],
     )
     return GuideVideoRewardOut(**result)

app/core/config.py

@@ -79,6 +79,8 @@ class Settings(BaseSettings):
     HONOR_PUSH_APP_ID: str = ""
     HONOR_PUSH_CLIENT_ID: str = ""
     HONOR_PUSH_CLIENT_SECRET: str = ""
+    # 0=正式消息(默认)；1=测试消息(仅开发联调，勿用于生产)。
+    HONOR_PUSH_TARGET_USER_TYPE: int = Field(default=0, ge=0, le=1)
     HONOR_PUSH_TOKEN_ENDPOINT: str = "https://iam.developer.honor.com/auth/token"
     HONOR_PUSH_SEND_ENDPOINT_TEMPLATE: str = (
         "https://push-api.cloud.honor.com/api/v1/{app_id}/sendMessage"
@@ -88,6 +90,8 @@ class Settings(BaseSettings):
     # (OAuth 换 token 时 client_id 即 AppId)。发送走 v1 messages:send,成功码 80000000。
     HUAWEI_PUSH_APP_ID: str = ""
     HUAWEI_PUSH_APP_SECRET: str = ""
+    # 0=正式消息(默认，受正式消息频控)；1=测试消息(仅开发联调，勿用于生产)。
+    HUAWEI_PUSH_TARGET_USER_TYPE: int = Field(default=0, ge=0, le=1)
     HUAWEI_PUSH_TOKEN_ENDPOINT: str = "https://oauth-login.cloud.huawei.com/oauth2/v3/token"
     HUAWEI_PUSH_SEND_ENDPOINT_TEMPLATE: str = (
         "https://push-api.cloud.huawei.com/v1/{app_id}/messages:send"
@@ -146,6 +150,9 @@ class Settings(BaseSettings):
     # aliyun:阿里云 dypns 号码认证,阿里云生成+下发+校验(Mode A);缺凭证时 /sms/* 返 503(优雅降级)。
     # chuanglan:创蓝云智(253)模板短信,本服务生成码、创蓝只下发、本地校验(Mode B,与极光同);缺凭证 503。
     SMS_PROVIDER: Literal["jiguang", "aliyun", "chuanglan"] = "jiguang"
+    # 主 provider 供应商不可用(SmsError 503)时转投的备 provider；空=关闭 fallback(默认，
+    # 保持现状零风险)。仅 Mode B(jiguang/chuanglan)互为主备有意义；生产建议设 "chuanglan"。
+    SMS_FALLBACK_PROVIDER: Literal["jiguang", "chuanglan", ""] = ""
     ALIYUN_SMS_ACCESS_KEY_ID: str = ""
     ALIYUN_SMS_ACCESS_KEY_SECRET: str = ""
     ALIYUN_SMS_SIGN_NAME: str = ""            # 系统赠送签名(自定义签名下发易失败)
@@ -189,14 +196,27 @@ class Settings(BaseSettings):
     # (real 模式下也跳过校验)、每次登录【强制重走新手引导】,并设【每日使用次数上限】防被人
     # 猜到号后脚本滥用。两个值都能随时改 .env。逻辑全在 app/core/test_account.py,与其他业务解耦。
     # ⚠️ TEST_ACCOUNT_PHONE 留空 = 整个功能关闭(生产默认安全;要启用才显式填号)。
-    TEST_ACCOUNT_PHONE: str = ""           # 测试手机号(11 位,如 11111111111);空=关闭整功能
+    TEST_ACCOUNT_PHONE: str = ""           # 兼容旧配置:单个测试手机号
+    TEST_ACCOUNT_PHONES: str = ""          # 多个测试手机号,英文逗号分隔
     TEST_ACCOUNT_DAILY_LIMIT: int = 500    # 该测试号每日最多登录次数,当日超过即拒绝登录
 
     @property
     def test_account_phone(self) -> str:
-        """规整后的测试手机号(去空白);空串=功能关闭。"""
+        """兼容旧调用:规整后的单个测试手机号。"""
         return self.TEST_ACCOUNT_PHONE.strip()
 
+    @property
+    def test_account_phones(self) -> frozenset[str]:
+        """所有测试手机号；新旧配置取并集，便于线上平滑迁移。"""
+        phones = {
+            phone.strip()
+            for phone in self.TEST_ACCOUNT_PHONES.split(",")
+            if phone.strip()
+        }
+        if self.test_account_phone:
+            phones.add(self.test_account_phone)
+        return frozenset(phones)
+
     # ===== 美团联盟 CPS =====
     # 未配置时所有 /api/v1/meituan/* 接口 200 返空(优雅降级),不影响登录/领券等其他业务。
     MT_CPS_APP_KEY: str = ""
@@ -436,6 +456,7 @@ class Settings(BaseSettings):
     # 运营后台上传的新手引导视频上限。视频比图片大一个量级,单独一档;
     # ⚠️ 改大时同步放宽网关 client_max_body_size(实测 QA 4MiB / prod 32MiB),否则 nginx 先挡下。
     GUIDE_VIDEO_MAX_BYTES: int = 100 * 1024 * 1024  # 引导视频最大 100MB
+    FFPROBE_BINARY: str = "ffprobe"
 
     # ===== 邀请好友 =====
     # 分享落地页(二维码 / 分享链接指向这里;扫码 → 落地页 → 引导浏览器下载 APK)。
@@ -488,7 +509,7 @@ class Settings(BaseSettings):
         return self.APP_ENV == "prod"
 
     @model_validator(mode="after")
-    def _enforce_prod_secrets(self) -> "Settings":
+    def _enforce_prod_secrets(self) -> Settings:
         """prod 下强校验 JWT secret,弱/默认/空即启动报错(fail-fast,挡住 token 被伪造)。
 
         只校验两个签发凭证:App 用户的 JWT_SECRET_KEY、后台的 ADMIN_JWT_SECRET——它们沿用默认值

app/core/config_schema.py

@@ -14,12 +14,138 @@ from app.core import rewards as r
 RISK_SMS_HOURLY_THRESHOLD_KEY = "risk_sms_hourly_threshold"
 RISK_ONECLICK_DAILY_THRESHOLD_KEY = "risk_oneclick_daily_threshold"
 RISK_COMPARE_DAILY_THRESHOLD_KEY = "risk_compare_daily_threshold"
+COMPARE_DAILY_LIMIT_KEY = "compare_daily_limit"
+SMS_SEND_HOURLY_LIMIT_KEY = "sms_send_hourly_limit"
+SMS_SEND_DAILY_LIMIT_KEY = "sms_send_daily_limit"
+SMS_LOGIN_HOURLY_LIMIT_KEY = "sms_login_hourly_limit"
+WECHAT_BIND_SMS_HOURLY_LIMIT_KEY = "wechat_bind_sms_hourly_limit"
+WECHAT_CONFLICT_HOURLY_LIMIT_KEY = "wechat_conflict_hourly_limit"
+SMS_PHONE_COOLDOWN_SECONDS_KEY = "sms_phone_cooldown_seconds"
+SMS_CODE_MAX_FAILED_ATTEMPTS_KEY = "sms_code_max_failed_attempts"
+AD_REWARD_VIDEO_DAILY_LIMIT_KEY = "ad_reward_video_daily_limit"
+AD_FEED_DAILY_LIMIT_KEY = "ad_feed_daily_limit"
+PHONE_REBIND_DAYS_KEY = "phone_rebind_days"
+GUIDE_VIDEO_MAX_PLAYS_KEY = "guide_video_max_plays"
+LIMIT_POLICY_GLOBAL_KEY = "limit_policy_global"
 
 # type 约定(给前端渲染编辑控件用):int / int_list / dict_str_int / bool / enum
 # hidden=True:仍是合法可配项(业务照常 get_value / admin 可经专用端点读写),但**不在通用
 #   「系统配置」页渲染**(admin/routers/config.py:list_config 按此过滤)。用于把已下线/已改由
 #   专用页管理的项从福利页 Tab 收起,同时保留后端默认值与写入能力。
 CONFIG_DEFS: dict[str, dict[str, Any]] = {
+    COMPARE_DAILY_LIMIT_KEY: {
+        "default": 100,
+        "label": "账号每日比价次数上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 100_000,
+        "hidden": True,
+        "help": "北京时间自然日内每个账号可发起的比价次数。",
+    },
+    SMS_SEND_HOURLY_LIMIT_KEY: {
+        "default": 5,
+        "label": "短信每小时发送上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 10_000,
+        "hidden": True,
+        "help": "同一设备和 IP 在固定 1 小时窗口内成功发送短信的次数。",
+    },
+    SMS_SEND_DAILY_LIMIT_KEY: {
+        "default": 20,
+        "label": "短信 24 小时发送上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 100_000,
+        "hidden": True,
+        "help": "同一设备和 IP 在固定 24 小时窗口内成功发送短信的次数。",
+    },
+    SMS_LOGIN_HOURLY_LIMIT_KEY: {
+        "default": 5,
+        "label": "短信登录每小时尝试上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 10_000,
+        "hidden": True,
+        "help": "成功与失败均计入。",
+    },
+    WECHAT_BIND_SMS_HOURLY_LIMIT_KEY: {
+        "default": 5,
+        "label": "微信短信绑定每小时尝试上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 10_000,
+        "hidden": True,
+    },
+    WECHAT_CONFLICT_HOURLY_LIMIT_KEY: {
+        "default": 5,
+        "label": "微信冲突处理每小时尝试上限",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 10_000,
+        "hidden": True,
+        "help": "继续登录与重新绑定共享此额度。",
+    },
+    SMS_PHONE_COOLDOWN_SECONDS_KEY: {
+        "default": 60,
+        "label": "同手机号短信发送冷却秒数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 0,
+        "max": 86_400,
+        "hidden": True,
+    },
+    SMS_CODE_MAX_FAILED_ATTEMPTS_KEY: {
+        "default": 5,
+        "label": "单验证码最大失败次数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 100,
+        "hidden": True,
+    },
+    AD_REWARD_VIDEO_DAILY_LIMIT_KEY: {
+        "default": r.DAILY_AD_REWARD_LIMIT,
+        "label": "激励视频每日发奖次数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 100_000,
+        "hidden": True,
+    },
+    AD_FEED_DAILY_LIMIT_KEY: {
+        "default": r.DAILY_AD_REWARD_LIMIT,
+        "label": "Draw 信息流每日发奖次数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 1,
+        "max": 100_000,
+        "hidden": True,
+    },
+    PHONE_REBIND_DAYS_KEY: {
+        "default": 30,
+        "label": "手机/微信换绑冷却天数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 0,
+        "max": 3650,
+        "hidden": True,
+    },
+    GUIDE_VIDEO_MAX_PLAYS_KEY: {
+        "default": 3,
+        "label": "领券引导视频最大播放次数",
+        "group": "限制策略",
+        "type": "int",
+        "min": 0,
+        "max": 50,
+        "hidden": True,
+    },
     "signin_rewards": {
         "default": list(r.SIGNIN_REWARDS), "label": "签到 7 天金币档位",
         "group": "签到", "type": "int_list",
@@ -56,7 +182,8 @@ CONFIG_DEFS: dict[str, dict[str, Any]] = {
     },
     "ad_daily_limit": {
         "default": r.DAILY_AD_REWARD_LIMIT, "label": "看广告每日上限(次)",
-        "group": "看广告", "type": "int", "help": "福利页激励视频每日可发奖次数上限,默认 500。",
+        "group": "看广告", "type": "int", "min": 1, "max": 100_000, "hidden": True,
+        "help": "历史共享上限；新配置由白名单页分别管理激励视频与 Draw 信息流。",
     },
     "ad_max_coin": {
         "default": r.MAX_AD_REWARD_COIN, "label": "看广告单次金币上限",
@@ -68,7 +195,8 @@ CONFIG_DEFS: dict[str, dict[str, Any]] = {
     },
     "ad_cooldown_sec": {
         "default": r.VIDEO_ROUND_COOLDOWN_SECONDS, "label": "广告关闭后冷却(秒)",
-        "group": "看广告", "type": "int", "help": "点击退出广告后,下次点击观看前的冷却时间,默认 3 秒。",
+        "group": "看广告", "type": "int", "min": 0, "max": 86_400,
+        "help": "点击退出广告后,下次点击观看前的冷却时间,默认 3 秒。",
     },
     "comparing_ad_enabled": {
         "default": True, "label": "比价/领券期信息流广告",
@@ -118,9 +246,9 @@ CONFIG_DEFS: dict[str, dict[str, Any]] = {
         "group": "风控",
         "type": "int",
         "min": 1,
-        "max": 5,
+        "max": 100_000,
         "hidden": True,
-        "help": "同一设备在北京时间同一自然小时内成功下发短信达到该次数时告警；不得高于现有每小时 5 次的发送上限。",
+        "help": "同一设备在北京时间同一自然小时内成功下发短信达到该次数时告警。",
     },
     RISK_ONECLICK_DAILY_THRESHOLD_KEY: {
         "default": 20,
@@ -138,7 +266,7 @@ CONFIG_DEFS: dict[str, dict[str, Any]] = {
         "group": "风控",
         "type": "int",
         "min": 1,
-        "max": 100,
+        "max": 100_000,
         "hidden": True,
         "help": "同一账户在北京时间同一自然日内发起比价达到该次数时告警。",
     },

app/core/limit_policy.py

@@ -0,0 +1,623 @@
+"""Unified global limits and per-phone/device policy overrides.
+
+The registry is the single source of truth for the whitelist page.  Existing
+constants remain as backwards-compatible defaults, while business call sites
+resolve an effective value here.
+"""
+from __future__ import annotations
+
+from collections.abc import Iterable
+from dataclasses import dataclass
+from datetime import UTC, datetime
+
+from sqlalchemy import delete, or_, select
+from sqlalchemy.orm import Session
+
+from app.core.config_schema import (
+    AD_FEED_DAILY_LIMIT_KEY,
+    AD_REWARD_VIDEO_DAILY_LIMIT_KEY,
+    COMPARE_DAILY_LIMIT_KEY,
+    GUIDE_VIDEO_MAX_PLAYS_KEY,
+    LIMIT_POLICY_GLOBAL_KEY,
+    PHONE_REBIND_DAYS_KEY,
+    RISK_COMPARE_DAILY_THRESHOLD_KEY,
+    RISK_ONECLICK_DAILY_THRESHOLD_KEY,
+    RISK_SMS_HOURLY_THRESHOLD_KEY,
+    SMS_CODE_MAX_FAILED_ATTEMPTS_KEY,
+    SMS_LOGIN_HOURLY_LIMIT_KEY,
+    SMS_PHONE_COOLDOWN_SECONDS_KEY,
+    SMS_SEND_DAILY_LIMIT_KEY,
+    SMS_SEND_HOURLY_LIMIT_KEY,
+    WECHAT_BIND_SMS_HOURLY_LIMIT_KEY,
+    WECHAT_CONFLICT_HOURLY_LIMIT_KEY,
+)
+from app.models.app_config import AppConfig
+from app.models.limit_policy import LimitPolicyOverride
+from app.models.user import User
+from app.repositories import app_config
+
+MODE_INHERIT = "inherit"
+MODE_OVERRIDE = "override"
+MODE_UNLIMITED = "unlimited"
+MODE_SUPPRESS_ALERT = "suppress_alert"
+
+SUBJECT_PHONE = "phone"
+SUBJECT_DEVICE = "device"
+SUBJECT_TYPES = (SUBJECT_PHONE, SUBJECT_DEVICE)
+SUBJECT_PRECEDENCE = {SUBJECT_DEVICE: 0, SUBJECT_PHONE: 1}
+LEGACY_IP_DEVICE_PREFIX = "legacy-ip:"
+
+
+@dataclass(frozen=True)
+class RuleDefinition:
+    code: str
+    label: str
+    group: str
+    config_key: str
+    default_limit: int
+    window_label: str
+    subject_types: tuple[str, ...]
+    min_value: int = 1
+    max_value: int = 100_000
+    allow_unlimited: bool = True
+    supports_reset: bool = True
+    alert_only: bool = False
+    legacy_config_key: str | None = None
+    legacy_json_field: str | None = None
+
+    @property
+    def allowed_modes(self) -> tuple[str, ...]:
+        if self.alert_only:
+            return (MODE_SUPPRESS_ALERT,)
+        return (MODE_UNLIMITED,) if self.allow_unlimited else ()
+
+
+RULES: tuple[RuleDefinition, ...] = (
+    RuleDefinition(
+        "compare.start.daily",
+        "每日发起比价次数",
+        "比价",
+        COMPARE_DAILY_LIMIT_KEY,
+        100,
+        "北京时间自然日",
+        SUBJECT_TYPES,
+    ),
+    RuleDefinition(
+        "sms.send.hourly",
+        "短信每小时成功发送次数",
+        "短信与登录",
+        SMS_SEND_HOURLY_LIMIT_KEY,
+        5,
+        "固定 1 小时窗口",
+        SUBJECT_TYPES,
+        max_value=10_000,
+    ),
+    RuleDefinition(
+        "sms.send.daily",
+        "短信 24 小时成功发送次数",
+        "短信与登录",
+        SMS_SEND_DAILY_LIMIT_KEY,
+        20,
+        "固定 24 小时窗口",
+        SUBJECT_TYPES,
+    ),
+    RuleDefinition(
+        "sms.phone.cooldown",
+        "同手机号短信发送冷却",
+        "短信与登录",
+        SMS_PHONE_COOLDOWN_SECONDS_KEY,
+        60,
+        "秒",
+        (SUBJECT_PHONE,),
+        min_value=0,
+        max_value=86_400,
+    ),
+    RuleDefinition(
+        "sms.code.failed_attempts",
+        "单验证码最大失败次数",
+        "短信与登录",
+        SMS_CODE_MAX_FAILED_ATTEMPTS_KEY,
+        5,
+        "单个验证码",
+        (SUBJECT_PHONE,),
+        max_value=100,
+    ),
+    RuleDefinition(
+        "sms.login.hourly",
+        "短信登录每小时尝试次数",
+        "短信与登录",
+        SMS_LOGIN_HOURLY_LIMIT_KEY,
+        5,
+        "固定 1 小时窗口",
+        SUBJECT_TYPES,
+        max_value=10_000,
+    ),
+    RuleDefinition(
+        "wechat.bind.hourly",
+        "微信短信绑定每小时尝试次数",
+        "短信与登录",
+        WECHAT_BIND_SMS_HOURLY_LIMIT_KEY,
+        5,
+        "固定 1 小时窗口",
+        SUBJECT_TYPES,
+        max_value=10_000,
+    ),
+    RuleDefinition(
+        "wechat.conflict.hourly",
+        "微信冲突处理每小时尝试次数",
+        "短信与登录",
+        WECHAT_CONFLICT_HOURLY_LIMIT_KEY,
+        5,
+        "固定 1 小时窗口",
+        SUBJECT_TYPES,
+        max_value=10_000,
+    ),
+    RuleDefinition(
+        "ad.reward_video.daily",
+        "激励视频每日发奖次数",
+        "广告",
+        AD_REWARD_VIDEO_DAILY_LIMIT_KEY,
+        500,
+        "北京时间自然日",
+        (SUBJECT_PHONE,),
+        legacy_config_key="ad_daily_limit",
+    ),
+    RuleDefinition(
+        "ad.feed.daily",
+        "Draw 信息流每日发奖次数",
+        "广告",
+        AD_FEED_DAILY_LIMIT_KEY,
+        500,
+        "北京时间自然日",
+        (SUBJECT_PHONE,),
+        legacy_config_key="ad_daily_limit",
+    ),
+    RuleDefinition(
+        "ad.reward_video.cooldown",
+        "激励视频发奖后冷却秒数",
+        "广告",
+        "ad_cooldown_sec",
+        3,
+        "秒",
+        (SUBJECT_PHONE,),
+        min_value=0,
+        max_value=86_400,
+    ),
+    RuleDefinition(
+        "guide.video.lifetime",
+        "领券引导视频最大播放次数",
+        "引导与账号",
+        GUIDE_VIDEO_MAX_PLAYS_KEY,
+        3,
+        "账号生命周期",
+        (SUBJECT_PHONE,),
+        min_value=0,
+        max_value=50,
+        legacy_config_key="coupon_guide_video",
+        legacy_json_field="max_plays",
+    ),
+    RuleDefinition(
+        "phone.rebind.days",
+        "手机/微信换绑冷却天数",
+        "引导与账号",
+        PHONE_REBIND_DAYS_KEY,
+        30,
+        "自然日",
+        (SUBJECT_PHONE,),
+        min_value=0,
+        max_value=3650,
+    ),
+    RuleDefinition(
+        "risk.sms.hourly",
+        "短信设备每小时告警",
+        "风控免告警",
+        RISK_SMS_HOURLY_THRESHOLD_KEY,
+        5,
+        "北京时间自然小时",
+        (SUBJECT_DEVICE,),
+        max_value=100_000,
+        allow_unlimited=False,
+        alert_only=True,
+    ),
+    RuleDefinition(
+        "risk.oneclick.daily",
+        "一键登录设备每日告警",
+        "风控免告警",
+        RISK_ONECLICK_DAILY_THRESHOLD_KEY,
+        20,
+        "北京时间自然日",
+        (SUBJECT_DEVICE,),
+        max_value=100_000,
+        allow_unlimited=False,
+        alert_only=True,
+    ),
+    RuleDefinition(
+        "risk.compare.daily",
+        "比价账号每日告警",
+        "风控免告警",
+        RISK_COMPARE_DAILY_THRESHOLD_KEY,
+        100,
+        "北京时间自然日",
+        (SUBJECT_PHONE,),
+        max_value=100_000,
+        allow_unlimited=False,
+        alert_only=True,
+    ),
+)
+RULE_MAP = {rule.code: rule for rule in RULES}
+RULE_CODE_BY_CONFIG_KEY = {rule.config_key: rule.code for rule in RULES}
+LIMIT_CONFIG_KEYS = tuple(RULE_CODE_BY_CONFIG_KEY)
+
+
+@dataclass(frozen=True)
+class EffectiveLimit:
+    rule_code: str
+    global_limit: int
+    limit: int | None
+    mode: str
+    suppressed: bool
+    override_id: int | None
+    matched_subject_type: str | None
+    matched_subject_value: str | None
+    reset_at: datetime | None
+    bucket_version: str
+
+    @property
+    def unlimited(self) -> bool:
+        return self.limit is None
+
+
+def normalize_subject(subject_type: str, value: str) -> str:
+    value = (value or "").strip()
+    if subject_type == SUBJECT_PHONE:
+        value = "".join(ch for ch in value if ch.isdigit())
+    if subject_type not in SUBJECT_TYPES:
+        raise ValueError(f"unsupported subject type: {subject_type}")
+    if not value:
+        raise ValueError("subject value is empty")
+    return value[:128]
+
+
+def validate_whitelist_subject(subject_type: str, value: str) -> str:
+    """Normalize a whitelist subject and reject unsafe pseudo-devices.
+
+    Old clients without a device ID are grouped by public IP for rate limiting.
+    That fallback remains valid at runtime, but it is not a stable, unique
+    device identity and must never be persisted as a device whitelist target.
+    """
+
+    normalized = normalize_subject(subject_type, value)
+    if (
+        subject_type == SUBJECT_DEVICE
+        and normalized.startswith(LEGACY_IP_DEVICE_PREFIX)
+    ):
+        raise ValueError(
+            "旧客户端未上报真实设备 ID，不能加入设备白名单，请升级客户端后重试"
+        )
+    return normalized
+
+
+def get_rule(rule_code: str) -> RuleDefinition:
+    try:
+        return RULE_MAP[rule_code]
+    except KeyError as exc:
+        raise ValueError(f"unknown rule: {rule_code}") from exc
+
+
+def device_source_scope(rule_code: str) -> str:
+    """返回设备候选数据所属命名空间，防止一个设备 ID 跨来源误套规则。"""
+    rule = get_rule(rule_code)
+    if SUBJECT_DEVICE not in rule.subject_types:
+        raise ValueError("当前限制项不支持设备白名单")
+    return "comparison" if rule_code == "compare.start.daily" else "auth"
+
+
+def default_global_limits() -> dict[str, int]:
+    """Return the complete 16-rule default snapshot keyed by rule code."""
+    return {rule.code: rule.default_limit for rule in RULES}
+
+
+def _normalise_global_limits(value: object) -> dict[str, int]:
+    """Merge a stored JSON object with safe code defaults.
+
+    The migration and every admin write persist all rules.  Defaults are still
+    merged here so a manually damaged/older partial JSON cannot take the
+    service down after deployment.
+    """
+
+    values = default_global_limits()
+    if not isinstance(value, dict):
+        return values
+    for rule_code, raw in value.items():
+        rule = RULE_MAP.get(str(rule_code))
+        if rule is None or isinstance(raw, bool):
+            continue
+        try:
+            parsed = int(raw)
+        except (TypeError, ValueError):
+            continue
+        if rule.min_value <= parsed <= rule.max_value:
+            values[rule.code] = parsed
+    return values
+
+
+def _legacy_global_limit(db: Session, rule: RuleDefinition) -> tuple[int, str]:
+    """Read the pre-bundle representation while upgrading old/test databases."""
+
+    row = db.get(AppConfig, rule.config_key)
+    if row is not None:
+        return int(row.value), "legacy-key"
+
+    if rule.legacy_config_key:
+        legacy = db.get(AppConfig, rule.legacy_config_key)
+        if legacy is not None:
+            value = legacy.value
+            if rule.legacy_json_field:
+                value = value.get(rule.legacy_json_field) if isinstance(value, dict) else None
+            if value is not None:
+                return int(value), "legacy"
+
+    try:
+        return int(app_config.get_value(db, rule.config_key)), "default"
+    except KeyError:
+        return rule.default_limit, "default"
+
+
+def get_global_limits(db: Session) -> dict[str, int]:
+    """Read the complete global-limit JSON, with a pre-migration fallback."""
+
+    row = db.get(AppConfig, LIMIT_POLICY_GLOBAL_KEY)
+    if row is not None:
+        return _normalise_global_limits(row.value)
+    return {rule.code: _legacy_global_limit(db, rule)[0] for rule in RULES}
+
+
+def set_global_limits(
+    db: Session,
+    updates: dict[str, int],
+    *,
+    admin_id: int,
+    commit: bool = True,
+) -> dict[str, int]:
+    """Atomically update selected rules inside the single complete JSON row."""
+
+    parsed_updates: dict[str, int] = {}
+    for rule_code, raw_value in updates.items():
+        rule = get_rule(rule_code)
+        value = int(raw_value)
+        if not rule.min_value <= value <= rule.max_value:
+            raise ValueError(
+                f"limit for {rule_code} must be between "
+                f"{rule.min_value} and {rule.max_value}"
+            )
+        parsed_updates[rule.code] = value
+
+    row = db.scalar(
+        select(AppConfig)
+        .where(AppConfig.key == LIMIT_POLICY_GLOBAL_KEY)
+        .with_for_update()
+    )
+    values = (
+        _normalise_global_limits(row.value)
+        if row is not None
+        else {rule.code: _legacy_global_limit(db, rule)[0] for rule in RULES}
+    )
+    values.update(parsed_updates)
+
+    if row is None:
+        row = AppConfig(
+            key=LIMIT_POLICY_GLOBAL_KEY,
+            value=values,
+            updated_by_admin_id=admin_id,
+        )
+        db.add(row)
+    else:
+        row.value = dict(values)
+        row.updated_by_admin_id = admin_id
+
+    # Once the bundle exists, stale sparse rows must not become a second source
+    # of truth.  The data migration performs the same cleanup for production.
+    db.execute(delete(AppConfig).where(AppConfig.key.in_(LIMIT_CONFIG_KEYS)))
+    if commit:
+        db.commit()
+        db.refresh(row)
+    else:
+        db.flush()
+    return values
+
+
+def _global_limit(db: Session, rule: RuleDefinition) -> tuple[int, str]:
+    row = db.get(AppConfig, LIMIT_POLICY_GLOBAL_KEY)
+    if row is not None:
+        return _normalise_global_limits(row.value)[rule.code], "configured"
+    return _legacy_global_limit(db, rule)
+
+
+def _aware(value: datetime | None) -> datetime | None:
+    if value is None:
+        return None
+    return value.replace(tzinfo=UTC) if value.tzinfo is None else value
+
+
+def _matching_overrides(
+    db: Session,
+    rule: RuleDefinition,
+    subjects: dict[str, str | None],
+    now: datetime,
+) -> list[LimitPolicyOverride]:
+    pairs: list[tuple[str, str]] = []
+    for subject_type in rule.subject_types:
+        raw = subjects.get(subject_type)
+        if raw:
+            pairs.append((subject_type, normalize_subject(subject_type, raw)))
+    if not pairs:
+        return []
+    clauses = [
+        (
+            (LimitPolicyOverride.subject_type == subject_type)
+            & (LimitPolicyOverride.subject_value == subject_value)
+        )
+        for subject_type, subject_value in pairs
+    ]
+    rows = list(
+        db.execute(
+            select(LimitPolicyOverride).where(
+                LimitPolicyOverride.rule_code == rule.code,
+                LimitPolicyOverride.enabled.is_(True),
+                or_(*clauses),
+            )
+        ).scalars()
+    )
+    active = [
+        row
+        for row in rows
+        if (_aware(row.starts_at) is None or _aware(row.starts_at) <= now)
+        and (_aware(row.expires_at) is None or _aware(row.expires_at) > now)
+    ]
+    return sorted(active, key=lambda row: SUBJECT_PRECEDENCE[row.subject_type])
+
+
+def resolve(
+    db: Session,
+    rule_code: str,
+    *,
+    phone: str | None = None,
+    device: str | None = None,
+    now: datetime | None = None,
+) -> EffectiveLimit:
+    """Resolve global config plus the most specific active override.
+
+    Device overrides win over phone overrides when both match.
+    """
+
+    rule = get_rule(rule_code)
+    now = _aware(now) or datetime.now(UTC)
+    global_limit, global_version = _global_limit(db, rule)
+    matches = _matching_overrides(
+        db, rule, {SUBJECT_PHONE: phone, SUBJECT_DEVICE: device}, now
+    )
+    row = matches[0] if matches else None
+    if row is None:
+        return EffectiveLimit(
+            rule.code,
+            global_limit,
+            global_limit,
+            MODE_INHERIT,
+            False,
+            None,
+            None,
+            None,
+            None,
+            global_version,
+        )
+
+    limit: int | None = global_limit
+    suppressed = False
+    if row.mode == MODE_OVERRIDE:
+        limit = int(row.limit_value) if row.limit_value is not None else global_limit
+    elif row.mode == MODE_UNLIMITED:
+        limit = None
+    elif row.mode == MODE_SUPPRESS_ALERT:
+        suppressed = True
+    # 编辑限制值/备注不应隐式清空计数；只有显式“重置状态”才切换桶。
+    reset_version = _aware(row.reset_at)
+    version = (
+        f"{global_version}:o:{row.id}:"
+        f"{reset_version.isoformat() if reset_version else '0'}"
+    )
+    return EffectiveLimit(
+        rule.code,
+        global_limit,
+        limit,
+        row.mode,
+        suppressed,
+        row.id,
+        row.subject_type,
+        row.subject_value,
+        _aware(row.reset_at),
+        version,
+    )
+
+
+def resolve_for_user(
+    db: Session,
+    rule_code: str,
+    user_id: int,
+    *,
+    device: str | None = None,
+) -> EffectiveLimit:
+    user = db.get(User, user_id)
+    return resolve(
+        db,
+        rule_code,
+        phone=user.phone if user is not None else None,
+        device=device,
+    )
+
+
+def rule_catalog(db: Session) -> list[dict]:
+    out: list[dict] = []
+    values = get_global_limits(db)
+    for rule in RULES:
+        out.append(
+            {
+                "code": rule.code,
+                "label": rule.label,
+                "group": rule.group,
+                "global_limit": values[rule.code],
+                "default_limit": rule.default_limit,
+                "window_label": rule.window_label,
+                "subject_types": list(rule.subject_types),
+                "allowed_modes": list(rule.allowed_modes),
+                "min_value": rule.min_value,
+                "max_value": rule.max_value,
+                "supports_reset": rule.supports_reset,
+                "alert_only": rule.alert_only,
+            }
+        )
+    return out
+
+
+def validate_override(
+    rule: RuleDefinition,
+    *,
+    subject_type: str,
+    mode: str,
+    limit_value: int | None,
+    starts_at: datetime | None,
+    expires_at: datetime | None,
+) -> None:
+    if subject_type not in rule.subject_types:
+        raise ValueError("该规则不支持此主体类型")
+    if mode not in rule.allowed_modes:
+        raise ValueError("该规则不支持此策略模式")
+    if limit_value is not None:
+        raise ValueError("白名单不支持覆盖指定值")
+    if mode in {MODE_UNLIMITED, MODE_SUPPRESS_ALERT} and expires_at is None:
+        raise ValueError("临时白名单必须设置失效时间")
+    if starts_at and expires_at and _aware(expires_at) <= _aware(starts_at):
+        raise ValueError("失效时间必须晚于生效时间")
+    if (
+        mode in {MODE_UNLIMITED, MODE_SUPPRESS_ALERT}
+        and expires_at is not None
+        and _aware(expires_at) <= datetime.now(UTC)
+    ):
+        raise ValueError("临时白名单的失效时间必须晚于当前时间")
+    if mode == MODE_SUPPRESS_ALERT and not rule.alert_only:
+        raise ValueError("免告警只支持风控监控的三项规则")
+
+
+def active_overrides(
+    db: Session,
+    *,
+    subject_type: str | None = None,
+    keyword: str | None = None,
+) -> Iterable[LimitPolicyOverride]:
+    stmt = select(LimitPolicyOverride)
+    if subject_type:
+        stmt = stmt.where(LimitPolicyOverride.subject_type == subject_type)
+    if keyword:
+        stmt = stmt.where(LimitPolicyOverride.subject_value.ilike(f"%{keyword.strip()}%"))
+    return db.execute(
+        stmt.order_by(LimitPolicyOverride.updated_at.desc(), LimitPolicyOverride.id.desc())
+    ).scalars()

app/core/media.py

@@ -10,8 +10,13 @@
 """
 from __future__ import annotations
 
+import json
+import os
 import secrets
+import subprocess
+from decimal import ROUND_HALF_UP, Decimal, InvalidOperation
 from pathlib import Path
+from typing import Any
 
 from app.core.config import settings
 
@@ -81,29 +86,91 @@ def _sniff_video_ext(data: bytes) -> str | None:
     """按魔数判定视频类型,返回扩展名;非支持类型返回 None。
 
     只认 MP4 家族(ISO BMFF):`....ftyp` 在偏移 4。Android ExoPlayer 与浏览器 <video>
-    都稳吃 H.264/AAC 的 mp4;放开 mkv/avi 只会让端上放不出来,不如在入口就挡掉。
+    支持 H.264 或 HEVC/H.265 视频及 AAC 音频;放开 mkv/avi 只会让端上放不出来,
+    不如在入口就挡掉。
     """
     if len(data) >= 12 and data[4:8] == b"ftyp":
         return ".mp4"
     return None
 
 
-def save_guide_video(data: bytes) -> str:
-    """保存新手引导视频(运营后台上传的运营素材),返回相对 URL(`/media/guide_video/<file>`)。
+def _probe_guide_video(path: Path) -> dict[str, Any]:
+    """用 ffprobe 读取服务端可信的时长/编码信息。"""
+    try:
+        proc = subprocess.run(
+            [
+                settings.FFPROBE_BINARY,
+                "-v", "error",
+                "-show_format",
+                "-show_streams",
+                "-of", "json",
+                str(path),
+            ],
+            capture_output=True,
+            text=True,
+            timeout=30,
+            check=False,
+        )
+    except FileNotFoundError as exc:
+        raise MediaError("服务器未安装 ffprobe，暂时无法分析视频") from exc
+    except (OSError, subprocess.TimeoutExpired) as exc:
+        raise MediaError("ffprobe 分析视频失败") from exc
+    if proc.returncode != 0:
+        raise MediaError("无法分析视频，请确认文件是有效的 MP4")
+    try:
+        payload = json.loads(proc.stdout)
+        streams = payload.get("streams") or []
+        video = next(s for s in streams if s.get("codec_type") == "video")
+        audio = next((s for s in streams if s.get("codec_type") == "audio"), None)
+        raw_duration = (
+            (payload.get("format") or {}).get("duration")
+            or video.get("duration")
+        )
+        duration_ms = int(
+            (Decimal(str(raw_duration)) * 1000).quantize(Decimal("1"), rounding=ROUND_HALF_UP)
+        )
+    except (StopIteration, TypeError, ValueError, InvalidOperation) as exc:
+        raise MediaError("视频缺少可识别的视频轨或时长") from exc
+    video_codec = str(video.get("codec_name") or "").lower()
+    audio_codec = str(audio.get("codec_name") or "").lower() if audio else None
+    if not 30_000 <= duration_ms <= 180_000:
+        raise MediaError("视频时长必须在 30～180 秒之间")
+    if video_codec not in ("h264", "hevc", "h265"):
+        raise MediaError("视频编码必须为 H.264 或 HEVC/H.265")
+    if audio_codec not in (None, "aac"):
+        raise MediaError("音频编码必须为 AAC")
+    return {
+        "duration_ms": duration_ms,
+        "video_codec": video_codec,
+        "audio_codec": audio_codec,
+        "analysis_status": "valid",
+        "analysis_error": None,
+    }
 
-    与图片分开一套校验:体积上限走 [settings.GUIDE_VIDEO_MAX_BYTES],类型只认 MP4。
-    """
+
+def save_guide_video(data: bytes) -> tuple[str, dict[str, Any]]:
+    """临时落盘、ffprobe 校验后原子发布，返回 URL 与分析结果。"""
     if not data:
         raise MediaError("空文件")
     limit = settings.GUIDE_VIDEO_MAX_BYTES
     if len(data) > limit:
         raise MediaError(f"视频过大(上限 {limit // (1024 * 1024)}MB)")
     if _sniff_video_ext(data) is None:
-        raise MediaError("仅支持 MP4 视频(H.264 编码)")
+        raise MediaError("仅支持 MP4 视频(H.264 或 HEVC/H.265 编码)")
 
-    fname = f"guide_{secrets.token_hex(8)}.mp4"
-    (_media_dir("guide_video") / fname).write_bytes(data)
-    return f"{settings.MEDIA_URL_PREFIX}/guide_video/{fname}"
+    directory = _media_dir("guide_video")
+    stem = f"guide_{secrets.token_hex(8)}"
+    temp_path = directory / f".{stem}.upload"
+    final_path = directory / f"{stem}.mp4"
+    try:
+        temp_path.write_bytes(data)
+        analysis = _probe_guide_video(temp_path)
+        os.replace(temp_path, final_path)
+    except Exception:
+        temp_path.unlink(missing_ok=True)
+        final_path.unlink(missing_ok=True)
+        raise
+    return f"{settings.MEDIA_URL_PREFIX}/guide_video/{final_path.name}", analysis
 
 
 def save_cps_image(admin_id: int, data: bytes) -> str:

app/core/ratelimit.py

@@ -75,10 +75,11 @@ def enforce_rate_limit(
     request: Request,
     scope: str,
     subject: str,
-    limit: int,
+    limit: int | None,
     window_sec: float,
     *,
     detail: str = "操作过于频繁,请稍后再试",
+    bucket_suffix: str = "",
 ) -> None:
     """在路由内部手动限流,按 (subject, 客户端 IP) 计数。
 
@@ -87,9 +88,9 @@ def enforce_rate_limit(
     key = `scope:subject:client_ip`;同一 (subject, IP) 在 window_sec 内超过 limit 次 → 抛 429。
     受 [settings.RATE_LIMIT_ENABLED] 总开关控制(与 [rate_limit] 一致)。
     """
-    if not settings.RATE_LIMIT_ENABLED:
+    if not settings.RATE_LIMIT_ENABLED or limit is None:
         return
-    key = f"{scope}:{subject}:{_client_ip(request)}"
+    key = f"{scope}:{subject}:{_client_ip(request)}:{bucket_suffix}"
     if not _hit(key, limit, window_sec):
         raise HTTPException(
             status_code=status.HTTP_429_TOO_MANY_REQUESTS,
@@ -112,9 +113,10 @@ class RateLimitRule(NamedTuple):
     """
 
     scope: str
-    limit: int
+    limit: int | None
     window_sec: float
     detail: str = "操作过于频繁,请稍后再试"
+    bucket_suffix: str = ""
 
 
 def _peek(key: str, limit: int, window_sec: float) -> bool:
@@ -150,7 +152,13 @@ def check_rate_limits(request: Request, subject: str, rules: list[RateLimitRule]
         return
     ip = _client_ip(request)
     for rule in rules:
-        if not _peek(f"{rule.scope}:{subject}:{ip}", rule.limit, rule.window_sec):
+        if rule.limit is None:
+            continue
+        if not _peek(
+            f"{rule.scope}:{subject}:{ip}:{rule.bucket_suffix}",
+            rule.limit,
+            rule.window_sec,
+        ):
             raise HTTPException(
                 status_code=status.HTTP_429_TOO_MANY_REQUESTS,
                 detail=rule.detail,
@@ -167,4 +175,9 @@ def record_rate_limits(request: Request, subject: str, rules: list[RateLimitRule
         return
     ip = _client_ip(request)
     for rule in rules:
-        _commit(f"{rule.scope}:{subject}:{ip}", rule.window_sec)
+        if rule.limit is None:
+            continue
+        _commit(
+            f"{rule.scope}:{subject}:{ip}:{rule.bucket_suffix}",
+            rule.window_sec,
+        )

app/core/rewards.py

@@ -233,10 +233,16 @@ def calculate_ad_reward_coin(ecpm: str | int | float | None, count_after_this: i
     eCPM 在此先钳到 AD_ECPM_MAX_FEN(¥500 CPM):信息流广告一期 eCPM 由客户端上报,伪造天价值
     会铸天量金币;钳在这唯一入口,feed 与 reward_video 回退客户端 eCPM 的路径都护住,且阈值高于
     所有真实值,不影响正规发奖。
+
+    下限(2026-07):有真实正 eCPM 时单份至少 1 金币——低 eCPM 单份收益四舍五入成 0 时兜底为 1,
+    避免用户看了广告却因数值太小被记 too_short 零发。eCPM 缺失/为 0/非法(没有真实广告价值)仍返 0,
+    不凭空铸币、不破坏 ecpm_missing 语义;防刷上限仍由 AD_ECPM_MAX_FEN 钳顶把守。
     """
     ecpm_yuan = min(parse_ecpm_yuan(ecpm), AD_ECPM_MAX_FEN / 100.0)
+    if ecpm_yuan <= 0:
+        return 0
     yuan = (ecpm_yuan / 1000.0) * ad_ecpm_factor(ecpm_yuan) * ad_lt_factor(count_after_this)
-    return max(0, round(yuan * COIN_PER_YUAN))
+    return max(1, round(yuan * COIN_PER_YUAN))
 
 
 # 签到膨胀(SIGNIN_BOOST_COIN,固定 3000)已于 2026-07 下线:它与广告实际收益脱钩,产品确认

app/core/test_account.py

@@ -9,9 +9,8 @@
   3. **每日使用次数上限**:防被人猜到这个号后写脚本一直刷。当天登录数超过上限即拒绝(429),
      次日自动归零。
 
-手机号与上限都在 .env 配(`TEST_ACCOUNT_PHONE` / `TEST_ACCOUNT_DAILY_LIMIT`),随时可改。
-`TEST_ACCOUNT_PHONE` 留空 = 整个功能关闭(生产默认态),`is_test_account()` 对任何号都返回
-False,登录/短信回到原逻辑,零影响。
+手机号与上限都在 .env 配(`TEST_ACCOUNT_PHONES` / `TEST_ACCOUNT_DAILY_LIMIT`),随时可改。
+兼容旧的单号配置 `TEST_ACCOUNT_PHONE`;两项都留空 = 整个功能关闭(生产默认态)。
 
 计数存**进程内存**(单 worker 够用,与 sms.py 同款约定):重启清零、多 worker 不共享。作为
 一个测试号的粗粒度防滥用闸够用;且因所有登录都落同一个 phone → 同一个 user,滥用面天然只
@@ -27,9 +26,9 @@ from app.core.config import settings
 
 logger = logging.getLogger("shagua.test_account")
 
-# 进程内每日计数:(date_str, 当日已登录次数)。单 worker 有效,重启清零(见模块 docstring)。
+# 进程内每日计数:{手机号: (date_str, 当日已登录次数)}。单 worker 有效,重启清零。
 _lock = Lock()
-_usage: tuple[str, int] = ("", 0)
+_usage: dict[str, tuple[str, int]] = {}
 
 
 def _today() -> str:
@@ -37,16 +36,16 @@ def _today() -> str:
 
 
 def is_enabled() -> bool:
-    """功能总开关:配了 TEST_ACCOUNT_PHONE 才启用(空=关闭)。"""
-    return bool(settings.test_account_phone)
+    """功能总开关:至少配置了一个测试手机号才启用。"""
+    return bool(settings.test_account_phones)
 
 
 def is_test_account(phone: str) -> bool:
-    """该手机号是否为配置的测试账号。功能关闭时对任何号都返回 False。"""
-    return is_enabled() and phone == settings.test_account_phone
+    """该手机号是否在测试账号集合中。"""
+    return phone in settings.test_account_phones
 
 
-def try_consume_quota() -> bool:
+def try_consume_quota(phone: str) -> bool:
     """测试账号登录时调:当日计数 +1。
 
     Returns:
@@ -59,17 +58,17 @@ def try_consume_quota() -> bool:
     limit = settings.TEST_ACCOUNT_DAILY_LIMIT
     with _lock:
         today = _today()
-        day, cnt = _usage
+        day, cnt = _usage.get(phone, ("", 0))
         if day != today:  # 跨天归零
             cnt = 0
         if cnt >= limit:
-            _usage = (today, cnt)  # 已满,保持不变
+            _usage[phone] = (today, cnt)  # 已满,保持不变
             logger.warning(
-                "测试账号 %s 今日登录数已达上限 %d,拒绝", settings.test_account_phone, limit
+                "测试账号 %s 今日登录数已达上限 %d,拒绝", phone, limit
             )
             return False
-        _usage = (today, cnt + 1)
-        logger.info("测试账号 %s 第 %d/%d 次登录", settings.test_account_phone, cnt + 1, limit)
+        _usage[phone] = (today, cnt + 1)
+        logger.info("测试账号 %s 第 %d/%d 次登录", phone, cnt + 1, limit)
         return True
 
 
@@ -77,4 +76,4 @@ def _reset_for_test() -> None:
     """仅供单测:清空进程内计数,隔离用例间状态。"""
     global _usage
     with _lock:
-        _usage = ("", 0)
+        _usage = {}

app/integrations/sms/__init__.py

@@ -1,37 +1,96 @@
-"""短信验证码服务 —— provider 分派入口。
+"""短信验证码服务 —— provider 分派入口（极光主 + 可选创蓝备）。
 
-对外只暴露 `send_code` / `verify_code` / `SmsError`,api 层无需关心用哪个 provider。
-provider 由 `settings.SMS_PROVIDER` 选择(**每次调用读取**,支持运行时切换 + 灰度回退):
-  - `jiguang`(默认):自管码(本服务生成、内存存/校验,极光只发)。见 [jiguang.py](jiguang.py)。
-  - `aliyun`:阿里云号码认证(阿里云生成+下发+校验,Mode A)。见 [aliyun.py](aliyun.py)。
-  - `chuanglan`:创蓝云智(253)模板短信,自管码 Mode B(本服务生成、内存存/校验,创蓝只发)。见 [chuanglan.py](chuanglan.py)。
-
-mock(`SMS_MOCK=true`)与各 provider 的行为差异都封在 provider 内部;本层只做路由。
-拆包前本模块是单文件 `sms.py`;拆包后极光逻辑迁入 `jiguang` 子模块,行为零改动。
+对外暴露 `send_code`（返回 SendResult） / `verify_code` / `SmsError` / `SendResult`。
+主 provider = settings.SMS_PROVIDER（默认 jiguang）；备 = settings.SMS_FALLBACK_PROVIDER
+（默认空=无备）。**每次调用读 settings**，支持运行时切换 / 灰度回退。
 """
 from __future__ import annotations
 
+import logging
+
 from app.core.config import settings
 
 from . import aliyun, chuanglan, jiguang
-from .base import SmsError
+from .base import SendResult, SmsError
 
-__all__ = ["SmsError", "send_code", "verify_code"]
+__all__ = ["SmsError", "SendResult", "send_code", "verify_code"]
 
-# provider 名 -> 模块;未知/缺省值回退 jiguang(默认兜底,防误配把登录打挂)。
-_PROVIDERS = {"aliyun": aliyun, "chuanglan": chuanglan}
+logger = logging.getLogger("shagua.sms")
+
+# provider 名 <-> 模块。未知/缺省主 provider 回退 jiguang（防误配把登录打挂）。
+_ALL = {"jiguang": jiguang, "aliyun": aliyun, "chuanglan": chuanglan}
+_NAME = {jiguang: "jiguang", aliyun: "aliyun", chuanglan: "chuanglan"}
 
 
-def _provider():
-    """按配置选 provider 模块(每次调用读 settings,支持运行时切换 / 测试注入)。"""
-    return _PROVIDERS.get(settings.SMS_PROVIDER, jiguang)
+def _primary():
+    """当前主 provider 模块（未知/缺省回退 jiguang）。"""
+    return _ALL.get(settings.SMS_PROVIDER, jiguang)
 
 
-def send_code(phone: str) -> int:
-    """发送验证码,返回距下次可发的冷却秒数;失败抛 SmsError。委托给当前 provider。"""
-    return _provider().send_code(phone)
+def _fallback():
+    """启用的备 provider 模块；未配置 / 同主 / 未知名 → None（= 不启用 fallback）。"""
+    name = (settings.SMS_FALLBACK_PROVIDER or "").strip()
+    if not name or name == settings.SMS_PROVIDER:
+        return None
+    return _ALL.get(name)
 
 
-def verify_code(phone: str, code: str) -> bool:
-    """校验验证码,返回是否通过;provider 异常降级抛 SmsError。委托给当前 provider。"""
-    return _provider().verify_code(phone, code)
+def _send_with_cooldown(provider, phone: str, cooldown_sec: int | None) -> int:
+    if cooldown_sec is None:
+        return provider.send_code(phone)
+    return provider.send_code(phone, cooldown_sec=cooldown_sec)
+
+
+def send_code(
+    phone: str,
+    *,
+    cooldown_sec: int | None = None,
+) -> SendResult:
+    """发码：主成功即返回；仅主「供应商不可用(503)」且配置了备时转备补发。
+
+    429（本地冷却/超频）、400（手机号无效）不转——不绕过防刷、不为无效号白烧。
+    备也失败则抛备的 SmsError。调用方传入的动态冷却值在主、备渠道保持一致。
+    """
+    primary = _primary()
+    fb = _fallback()
+    try:
+        cooldown = _send_with_cooldown(primary, phone, cooldown_sec)
+        return SendResult(cooldown_sec=cooldown, provider=_NAME[primary], fallback=False)
+    except SmsError as e:
+        if fb is not None and e.status_code == 503:
+            logger.warning("[SMS] primary=%s 不可用(%s)，fallback→%s",
+                           _NAME[primary], e, _NAME[fb])
+            cooldown = _send_with_cooldown(fb, phone, cooldown_sec)
+            return SendResult(cooldown_sec=cooldown, provider=_NAME[fb], fallback=True)
+        raise
+
+
+def verify_code(
+    phone: str,
+    code: str,
+    *,
+    max_failed_attempts: int | None = None,
+) -> bool:
+    """校验：try-both，遍历「启用的 fallback 链」（主→备），任一命中即 True。
+
+    码只存在实际发码那家（fallback 前主已 pop 掉自己的码），另一家 rec is None 即 False、
+    不误判、不累加其防爆破计数。动态失败次数上限会一致传给主备渠道。
+    """
+    chain = [_primary()]
+    fb = _fallback()
+    if fb is not None:
+        chain.append(fb)
+    for prov in chain:
+        verified = (
+            prov.verify_code(phone, code)
+            if max_failed_attempts is None
+            else prov.verify_code(
+                phone,
+                code,
+                max_failed_attempts=max_failed_attempts,
+            )
+        )
+        if verified:
+            logger.info("[SMS] verify hit provider=%s", _NAME[prov])
+            return True
+    return False

app/integrations/sms/aliyun.py

@@ -35,7 +35,7 @@ _GC_THRESHOLD = 10000                    # 超此阈值,send 时顺手清老于
 
 # 发码错误码 → (HTTP 码, 用户提示)。未列出的一律 503(供应商不可用)。
 _SEND_ERRORS: dict[str, tuple[int, str]] = {
-    "MOBILE_NUMBER_ILLEGAL": (400, "手机号无效"),
+    "MOBILE_NUMBER_ILLEGAL": (400, "请输入有效的手机号"),
     "BUSINESS_LIMIT_CONTROL": (429, "今日发送次数过多,请明天再试"),
     "FREQUENCY_FAIL": (429, "发送过于频繁,请稍后再试"),
 }
@@ -47,19 +47,26 @@ _client = None  # 惰性构建的 SDK client(模块级缓存)
 
 # ============================ 对外:发码 / 校验 ============================
 
-def send_code(phone: str) -> int:
+def send_code(phone: str, *, cooldown_sec: int | None = None) -> int:
     """发送验证码(阿里云生成+下发)。
 
     Returns: 距下次可发的秒数(= ALIYUN_SMS_INTERVAL_SEC,冷却由阿里云 Interval 侧执行)。
     Raises: SmsError(手机号无效 400 / 过频·天级流控 429 / 未配置·未开通·其他 503)。
     """
+    effective_cooldown = (
+        settings.ALIYUN_SMS_INTERVAL_SEC if cooldown_sec is None else cooldown_sec
+    )
     if settings.SMS_MOCK:
         logger.info("[SMS-aliyun-MOCK] to %s**** (不真发)", phone[:3])
-        return settings.ALIYUN_SMS_INTERVAL_SEC
+        return effective_cooldown
     if not settings.aliyun_sms_configured:
         raise SmsError("短信服务未配置(缺阿里云凭证)", status_code=503)
 
-    result = _call_send(phone)  # 传输/SDK 异常在内部抛 SmsError(503)
+    result = (
+        _call_send(phone)
+        if cooldown_sec is None
+        else _call_send(phone, cooldown_sec=effective_cooldown)
+    )
 
     if result["success"] and result["code"] == "OK":
         now = time.time()
@@ -68,7 +75,7 @@ def send_code(phone: str) -> int:
             _verify_attempts.pop(phone, None)  # 新码 = 新失败预算
             _verify_seen.pop(phone, None)
         logger.info("[SMS-aliyun] sent to %s****", phone[:3])
-        return settings.ALIYUN_SMS_INTERVAL_SEC
+        return effective_cooldown
 
     code = result["code"]
     logger.error("[SMS-aliyun] send failed code=%s msg=%s", code, result["message"])
@@ -78,7 +85,12 @@ def send_code(phone: str) -> int:
     raise SmsError(msg, status_code=status)
 
 
-def verify_code(phone: str, code: str) -> bool:
+def verify_code(
+    phone: str,
+    code: str,
+    *,
+    max_failed_attempts: int | None = None,
+) -> bool:
     """校验验证码(阿里云裁决)。
 
     - **mock**:放行任意 N 位数字(provider 无关,同极光)。
@@ -92,8 +104,13 @@ def verify_code(phone: str, code: str) -> bool:
 
     # 失败计数是 best-effort:网络调用不持锁(不能锁跨 IO),故并发下同号可能多放行个位数次。
     # 无碍——API 层登录频控(设备+IP 5/时)是硬上限,阿里云码有效期 + DuplicatePolicy 亦兜底。
+    effective_max_attempts = (
+        settings.SMS_MAX_VERIFY_ATTEMPTS
+        if max_failed_attempts is None
+        else max_failed_attempts
+    )
     with _lock:
-        if _verify_attempts.get(phone, 0) >= settings.SMS_MAX_VERIFY_ATTEMPTS:
+        if _verify_attempts.get(phone, 0) >= effective_max_attempts:
             return False  # 已作废:保持计数(直到 send_code 重置),与极光「达上限即作废」一致
 
     result = _call_check(phone, code)  # 传输/SDK 异常在内部抛 SmsError(503)
@@ -146,7 +163,7 @@ def _get_client():
     return _client
 
 
-def _call_send(phone: str) -> dict:
+def _call_send(phone: str, *, cooldown_sec: int | None = None) -> dict:
     """调 SendSmsVerifyCode。返回归一化 {success, code, message};import/建 client/调用 任一失败抛 SmsError(503)。"""
     valid_min = max(1, settings.ALIYUN_SMS_VALID_TIME_SEC // 60)
     template_param = json.dumps({"code": "##code##", "min": str(valid_min)}, ensure_ascii=False)
@@ -160,7 +177,11 @@ def _call_send(phone: str) -> dict:
             template_param=template_param,
             code_length=settings.ALIYUN_SMS_CODE_LENGTH,
             valid_time=settings.ALIYUN_SMS_VALID_TIME_SEC,
-            interval=settings.ALIYUN_SMS_INTERVAL_SEC,
+            interval=(
+                settings.ALIYUN_SMS_INTERVAL_SEC
+                if cooldown_sec is None
+                else cooldown_sec
+            ),
             scheme_name=settings.ALIYUN_SMS_SCHEME_NAME or None,
         )
         body = _get_client().send_sms_verify_code(req).body

app/integrations/sms/base.py

@@ -2,9 +2,12 @@
 
 各 provider(jiguang / aliyun)都 `from .base import SmsError`,api 层也从包入口拿到同一个
 `SmsError` —— 保证无论用哪个 provider,异常类型与 HTTP 码映射语义一致。
+`SendResult` 是分派层 `send_code` 的返回类型,携带 cooldown、实际 provider 及是否走了 fallback。
 """
 from __future__ import annotations
 
+from dataclasses import dataclass
+
 from app.core.config import settings
 
 
@@ -18,6 +21,19 @@ class SmsError(Exception):
         self.status_code = status_code
 
 
+@dataclass(frozen=True)
+class SendResult:
+    """发码结果：距下次可发秒数 + 实际发码的 provider 名 + 是否走了备(fallback)。
+
+    分派层 send_code 的返回类型（provider 各自的 send_code 仍返回 int cooldown，
+    由分派层包装）。auth 层据此把渠道记入风控流水 details。
+    """
+
+    cooldown_sec: int
+    provider: str
+    fallback: bool = False
+
+
 def mock_verify(code: str) -> bool:
     """mock 模式校验:放行任意 SMS_CODE_LENGTH 位数字(provider 无关,测试/开发便利,不真校验)。"""
     return len(code) == settings.SMS_CODE_LENGTH and code.isdigit()

app/integrations/sms/chuanglan.py

@@ -53,7 +53,7 @@ _GC_THRESHOLD = 10000  # 任一内存 dict 超此阈值,send 时顺手清过期
 # 发码错误码(创蓝 `code`)→ (HTTP 码, 用户提示)。未列出的一律 503(供应商不可用)。
 _SEND_ERRORS: dict[str, tuple[int, str]] = {
     "103": (429, "发送过于频繁，请稍后再试"),   # 提交速度过快
-    "107": (400, "手机号无效"),                 # 手机号码错误
+    "107": (400, "请输入有效的手机号"),                 # 手机号码错误
 }
 # 需运维介入的配置/开通/余额类错误:打 critical 日志(仍归 503)。
 _SEND_CRITICAL_CODES = frozenset({
@@ -84,20 +84,23 @@ def _gc(now: float) -> None:
             _last_sent.pop(p, None)
 
 
-def send_code(phone: str) -> int:
+def send_code(phone: str, *, cooldown_sec: int | None = None) -> int:
     """发送验证码。
 
     Returns: 距下次可发的秒数(= SMS_SEND_INTERVAL_SEC)
     Raises: SmsError(过频 429 / 手机号无效 400 / 供应商失败 503)
     """
     now = time.time()
+    effective_cooldown = (
+        settings.SMS_SEND_INTERVAL_SEC if cooldown_sec is None else cooldown_sec
+    )
 
     # --- lock 内:防刷检查 + 预占(防并发重复发烧钱)---
     with _lock:
         _gc(now)  # 顺手清过期内存(超阈值才扫)
         elapsed = now - _last_sent.get(phone, 0.0)
-        if elapsed < settings.SMS_SEND_INTERVAL_SEC:
-            remain = int(settings.SMS_SEND_INTERVAL_SEC - elapsed)
+        if elapsed < effective_cooldown:
+            remain = int(effective_cooldown - elapsed)
             raise SmsError(f"发送过于频繁,请 {remain}s 后再试")
 
         code = _gen_code()
@@ -122,10 +125,15 @@ def send_code(phone: str) -> int:
         logger.exception("[SMS-chuanglan] send failed phone=%s****", phone[:3])
         raise SmsError("验证码发送失败,请稍后重试", status_code=503) from e
 
-    return settings.SMS_SEND_INTERVAL_SEC
+    return effective_cooldown
 
 
-def verify_code(phone: str, code: str) -> bool:
+def verify_code(
+    phone: str,
+    code: str,
+    *,
+    max_failed_attempts: int | None = None,
+) -> bool:
     """校验验证码。
 
     - **mock 模式**:放行任意 N 位数字(测试/开发便利,不真校验)。
@@ -136,6 +144,11 @@ def verify_code(phone: str, code: str) -> bool:
         logger.info("[SMS-chuanglan-MOCK] verify %s for %s****", "ok" if ok else "fail", phone[:3])
         return ok
 
+    effective_max_attempts = (
+        settings.SMS_MAX_VERIFY_ATTEMPTS
+        if max_failed_attempts is None
+        else max_failed_attempts
+    )
     with _lock:
         rec = _codes.get(phone)
         if rec is None:
@@ -143,7 +156,7 @@ def verify_code(phone: str, code: str) -> bool:
         if time.time() > rec.expires_at:
             _codes.pop(phone, None)
             return False
-        if rec.attempts >= settings.SMS_MAX_VERIFY_ATTEMPTS:
+        if rec.attempts >= effective_max_attempts:
             _codes.pop(phone, None)  # 试错过多,作废
             return False
         if secrets.compare_digest(code.encode("utf-8"), rec.code.encode("utf-8")):

app/integrations/sms/jiguang.py

@@ -70,20 +70,23 @@ def _gc(now: float) -> None:
             _last_sent.pop(p, None)
 
 
-def send_code(phone: str) -> int:
+def send_code(phone: str, *, cooldown_sec: int | None = None) -> int:
     """发送验证码。
 
     Returns: 距下次可发的秒数(= SMS_SEND_INTERVAL_SEC)
     Raises: SmsError(过频 429 / 当日超限 429 / 供应商失败 503 / 手机号无效 400)
     """
     now = time.time()
+    effective_cooldown = (
+        settings.SMS_SEND_INTERVAL_SEC if cooldown_sec is None else cooldown_sec
+    )
 
     # --- lock 内:防刷检查 + 预占(防并发重复发烧钱)---
     with _lock:
         _gc(now)  # 顺手清过期内存(超阈值才扫)
         elapsed = now - _last_sent.get(phone, 0.0)
-        if elapsed < settings.SMS_SEND_INTERVAL_SEC:
-            remain = int(settings.SMS_SEND_INTERVAL_SEC - elapsed)
+        if elapsed < effective_cooldown:
+            remain = int(effective_cooldown - elapsed)
             raise SmsError(f"发送过于频繁,请 {remain}s 后再试")
 
         code = _gen_code()
@@ -108,10 +111,15 @@ def send_code(phone: str) -> int:
         logger.exception("[SMS] send failed phone=%s****", phone[:3])
         raise SmsError("验证码发送失败,请稍后重试", status_code=503) from e
 
-    return settings.SMS_SEND_INTERVAL_SEC
+    return effective_cooldown
 
 
-def verify_code(phone: str, code: str) -> bool:
+def verify_code(
+    phone: str,
+    code: str,
+    *,
+    max_failed_attempts: int | None = None,
+) -> bool:
     """校验验证码。
 
     - **mock 模式**:放行任意 N 位数字(测试/开发便利,不真校验)。
@@ -122,6 +130,11 @@ def verify_code(phone: str, code: str) -> bool:
         logger.info("[SMS-MOCK] verify %s for %s****", "ok" if ok else "fail", phone[:3])
         return ok
 
+    effective_max_attempts = (
+        settings.SMS_MAX_VERIFY_ATTEMPTS
+        if max_failed_attempts is None
+        else max_failed_attempts
+    )
     with _lock:
         rec = _codes.get(phone)
         if rec is None:
@@ -129,7 +142,7 @@ def verify_code(phone: str, code: str) -> bool:
         if time.time() > rec.expires_at:
             _codes.pop(phone, None)
             return False
-        if rec.attempts >= settings.SMS_MAX_VERIFY_ATTEMPTS:
+        if rec.attempts >= effective_max_attempts:
             _codes.pop(phone, None)  # 试错过多,作废
             return False
         if secrets.compare_digest(code.encode("utf-8"), rec.code.encode("utf-8")):
@@ -185,5 +198,5 @@ def _send_via_jiguang(phone: str, code: str) -> None:
     if ecode == 50009:  # 极光侧超频
         raise SmsError("发送过于频繁,请稍后再试", status_code=429)
     if ecode == 50006:  # 手机号无效(schema 已挡格式,这里多是空号/停机)
-        raise SmsError("手机号无效", status_code=400)
+        raise SmsError("请输入有效的手机号", status_code=400)
     raise SmsError(f"短信发送失败(code={ecode})", status_code=503)

app/integrations/vendor_push.py

@@ -22,7 +22,7 @@ import uuid
 from collections.abc import Callable
 from dataclasses import dataclass
 from typing import Any
-from urllib.parse import quote
+from urllib.parse import quote, urlsplit
 
 import httpx
 
@@ -31,6 +31,7 @@ from app.core.config import settings
 logger = logging.getLogger("shagua.vendor_push")
 
 TYPE_ACCESSIBILITY_DISABLED = "accessibility_disabled"
+DATA_EVENT_NOTIFICATION_CREATED = "notification_created"
 SUPPORTED_VENDORS = frozenset({"honor", "huawei", "vivo", "xiaomi", "oppo"})
 
 # vendor key → 中文名(测试/配置状态接口展示用)
@@ -69,6 +70,18 @@ class _CachedToken:
 
 _token_cache: dict[str, _CachedToken] = {}
 
+_LOG_SUMMARY_MAX_CHARS = 1500
+_LOG_STRING_MAX_CHARS = 200
+_SENSITIVE_LOG_KEYS = {
+    "accesstoken",
+    "appkey",
+    "authtoken",
+    "authorization",
+    "clientsecret",
+    "mastersecret",
+    "sign",
+}
+
 
 def normalize_vendor(push_vendor: str | None) -> str | None:
     if not push_vendor:
@@ -115,8 +128,16 @@ def send_notification(
 
     if mock:
         logger.info(
-            "[mock push] vendor=%s token=%s... title=%s body=%s extras=%s",
-            vendor, token[:12], title, body, extras,
+            "vendor push mock vendor=%s request=%s",
+            vendor,
+            _log_summary(
+                {
+                    "push_token": token,
+                    "title": title,
+                    "body": body,
+                    "extras": extras,
+                }
+            ),
         )
         return {
             "mock": True,
@@ -133,7 +154,28 @@ def send_notification(
         "xiaomi": _send_xiaomi,
         "oppo": _send_oppo,
     }
-    return dispatch[vendor](token, title, body, extras)
+    started = time.perf_counter()
+    try:
+        result = dispatch[vendor](token, title, body, extras)
+    except Exception:
+        logger.exception(
+            "vendor push dispatch failed vendor=%s token=%s notification_type=%s "
+            "elapsed_ms=%.1f",
+            vendor,
+            token,
+            extras.get("type", ""),
+            _elapsed_ms(started),
+        )
+        raise
+    logger.info(
+        "vendor push dispatch succeeded vendor=%s token=%s notification_type=%s "
+        "elapsed_ms=%.1f",
+        vendor,
+        token,
+        extras.get("type", ""),
+        _elapsed_ms(started),
+    )
+    return result
 
 
 def send_accessibility_disabled(
@@ -153,19 +195,191 @@ def send_accessibility_disabled(
     )
 
 
+def send_data_event(
+    push_vendor: str,
+    push_token: str,
+    *,
+    event: str,
+    notification_id: str,
+    mock: bool = False,
+) -> dict[str, Any]:
+    """发送无界面的轻量事件；当前只允许站内消息创建事件。"""
+    vendor = normalize_vendor(push_vendor)
+    token = push_token.strip() if push_token else ""
+    if not vendor or vendor not in SUPPORTED_VENDORS:
+        raise VendorPushError(f"unsupported push vendor: {push_vendor}")
+    if not token:
+        raise VendorPushError("push token is empty")
+    if event != DATA_EVENT_NOTIFICATION_CREATED:
+        raise VendorPushError(f"unsupported data event: {event}")
+    if not notification_id:
+        raise VendorPushError("notification_id is empty")
+
+    payload = {"event": event, "notificationId": str(notification_id)}
+    if mock:
+        return {"mock": True, "vendor": vendor, "payload": payload}
+
+    # vivo 的通知已设置 foregroundShow=false：App 在前台时必走
+    # onForegroundMessageArrived，正好就是本事件需要的刷新信号；不重复占用一次推送配额。
+    if vendor == "vivo":
+        return {"skipped": True, "reason": "foreground notification callback"}
+
+    # 小米服务端不再发送透传消息；通知栏消息仍照常发送，前台角标由客户端主动刷新。
+    if vendor == "xiaomi":
+        return {"skipped": True, "reason": "xiaomi data messages disabled"}
+
+    # OPush 当前只支持通知栏消息，没有服务端透传单推接口。
+    # 客户端在 OPPO 首页可见时轮询未读数；这里必须安全跳过，不能请求不存在的
+    # /message/transparent/unicast（该地址会稳定返回 HTTP 404）。
+    if vendor == "oppo":
+        return {"skipped": True, "reason": "oppo does not support data messages"}
+
+    dispatch: dict[str, Callable[[str, dict[str, str]], dict[str, Any]]] = {
+        "honor": _send_honor_data,
+        "huawei": _send_huawei_data,
+    }
+    return dispatch[vendor](token, payload)
+
+
 def _require(value: str, name: str) -> str:
     if not value:
         raise VendorPushError(f"{name} not configured")
     return value
 
 
+def _elapsed_ms(started: float) -> float:
+    return (time.perf_counter() - started) * 1000
+
+
+def _redacted_value(value: Any) -> str:
+    raw = json.dumps(value, ensure_ascii=False, sort_keys=True, default=str)
+    digest = hashlib.sha256(raw.encode()).hexdigest()[:10]
+    return f"<redacted len={len(raw)} sha256={digest}>"
+
+
+def _sanitize_for_log(value: Any, *, key: str = "", depth: int = 0) -> Any:
+    """生成有排障价值、但不暴露服务端鉴权凭据的紧凑日志摘要。"""
+    normalized_key = "".join(char for char in key.lower() if char.isalnum())
+    if normalized_key in _SENSITIVE_LOG_KEYS or normalized_key.endswith("secret"):
+        return _redacted_value(value)
+    if depth >= 10:
+        return f"<{type(value).__name__}>"
+    if isinstance(value, dict):
+        items = list(value.items())
+        sanitized = {
+            str(item_key): _sanitize_for_log(item_value, key=str(item_key), depth=depth + 1)
+            for item_key, item_value in items[:30]
+        }
+        if len(items) > 30:
+            sanitized["<omitted_keys>"] = len(items) - 30
+        return sanitized
+    if isinstance(value, (list, tuple)):
+        items = list(value)
+        sanitized = [_sanitize_for_log(item, key=key, depth=depth + 1) for item in items[:20]]
+        if len(items) > 20:
+            sanitized.append(f"<omitted_items={len(items) - 20}>")
+        return sanitized
+    if isinstance(value, str):
+        if value.lstrip().startswith(("{", "[")):
+            try:
+                decoded = json.loads(value)
+            except ValueError:
+                pass
+            else:
+                return _sanitize_for_log(decoded, key=key, depth=depth + 1)
+        if len(value) > _LOG_STRING_MAX_CHARS:
+            return f"{value[:_LOG_STRING_MAX_CHARS]}…<len={len(value)}>"
+    return value
+
+
+def _log_summary(value: Any) -> str:
+    rendered = json.dumps(
+        _sanitize_for_log(value),
+        ensure_ascii=False,
+        sort_keys=True,
+        separators=(",", ":"),
+        default=str,
+    )
+    if len(rendered) > _LOG_SUMMARY_MAX_CHARS:
+        return f"{rendered[:_LOG_SUMMARY_MAX_CHARS]}…<len={len(rendered)}>"
+    return rendered
+
+
+def _raw_log_summary(value: Any) -> str:
+    """厂商响应摘要：保留原始字段和值，仅限制单条日志长度。"""
+    rendered = json.dumps(
+        value,
+        ensure_ascii=False,
+        sort_keys=True,
+        separators=(",", ":"),
+        default=str,
+    )
+    if len(rendered) > _LOG_SUMMARY_MAX_CHARS:
+        return f"{rendered[:_LOG_SUMMARY_MAX_CHARS]}…<len={len(rendered)}>"
+    return rendered
+
+
+def _endpoint_for_log(url: str) -> str:
+    parsed = urlsplit(url)
+    return f"{parsed.scheme}://{parsed.netloc}{parsed.path}"
+
+
+def _request_summary(kwargs: dict[str, Any]) -> str:
+    summary: dict[str, Any] = {}
+    if "json" in kwargs:
+        summary["body_type"] = "json"
+        summary["request_payload"] = kwargs["json"]
+    elif "data" in kwargs:
+        summary["body_type"] = "form"
+        summary["request_payload"] = kwargs["data"]
+    if kwargs.get("params"):
+        summary["params"] = kwargs["params"]
+    if kwargs.get("headers"):
+        summary["headers"] = kwargs["headers"]
+    return _log_summary(summary)
+
+
+def _response_summary(resp: Any, parsed: Any | None = None) -> str:
+    if parsed is None:
+        try:
+            parsed = resp.json()
+        except ValueError:
+            pass
+    if parsed is not None:
+        return _raw_log_summary(parsed)
+    return _raw_log_summary(getattr(resp, "text", ""))
+
+
+def _vendor_response_failed(vendor: str, data: dict[str, Any]) -> bool:
+    """识别 HTTP 200 中明确的厂商业务失败，避免将失败请求记录成 success。"""
+    if data.get("error") or data.get("success") is False:
+        return True
+    if vendor == "xiaomi":
+        code = data.get("code")
+        result = str(data.get("result", "ok")).lower()
+        return code not in (0, "0", None) or result not in ("ok", "success")
+    if vendor == "oppo" and data.get("code") is not None:
+        return int(data["code"]) != 0
+    if vendor == "vivo" and data.get("result") is not None:
+        return int(data["result"]) != 0
+    if vendor == "honor" and data.get("code") is not None:
+        return int(data["code"]) != 200
+    if vendor == "huawei" and data.get("code") is not None:
+        return str(data["code"]) != "80000000"
+    return False
+
+
 def _request_json(
     method: str,
     url: str,
     *,
+    vendor: str,
+    operation: str,
     expected_status: tuple[int, ...] = (200,),
     **kwargs: Any,
 ) -> dict[str, Any]:
+    started = time.perf_counter()
+    request_summary = _request_summary(kwargs)
     try:
         resp = httpx.request(
             method,
@@ -174,41 +388,85 @@ def _request_json(
             **kwargs,
         )
     except httpx.HTTPError as e:
+        logger.error(
+            "vendor push http completed vendor=%s operation=%s method=%s endpoint=%s "
+            "outcome=network_error request=%s response=%s elapsed_ms=%.1f error=%s",
+            vendor,
+            operation,
+            method,
+            _endpoint_for_log(url),
+            request_summary,
+            "<no_response>",
+            _elapsed_ms(started),
+            type(e).__name__,
+        )
         raise VendorPushError(f"push http error: {e}") from e
 
     if resp.status_code not in expected_status:
-        logger.error("vendor push http failed url=%s http=%s body=%s", url, resp.status_code, resp.text[:500])
+        logger.error(
+            "vendor push http completed vendor=%s operation=%s method=%s endpoint=%s "
+            "outcome=http_error http_status=%s request=%s response=%s elapsed_ms=%.1f",
+            vendor,
+            operation,
+            method,
+            _endpoint_for_log(url),
+            resp.status_code,
+            request_summary,
+            _response_summary(resp),
+            _elapsed_ms(started),
+        )
         raise VendorPushError(f"push http {resp.status_code}")
     try:
-        return resp.json()
+        data = resp.json()
     except ValueError as e:
+        logger.error(
+            "vendor push http completed vendor=%s operation=%s method=%s endpoint=%s "
+            "outcome=invalid_json http_status=%s request=%s response=%s elapsed_ms=%.1f",
+            vendor,
+            operation,
+            method,
+            _endpoint_for_log(url),
+            resp.status_code,
+            request_summary,
+            _response_summary(resp),
+            _elapsed_ms(started),
+        )
         raise VendorPushError(f"push invalid json: {resp.text[:200]}") from e
+    vendor_failed = _vendor_response_failed(vendor, data)
+    log = logger.warning if vendor_failed else logger.info
+    log(
+        "vendor push http completed vendor=%s operation=%s method=%s endpoint=%s "
+        "outcome=%s http_status=%s request=%s response=%s elapsed_ms=%.1f",
+        vendor,
+        operation,
+        method,
+        _endpoint_for_log(url),
+        "vendor_error" if vendor_failed else "success",
+        resp.status_code,
+        request_summary,
+        _response_summary(resp, data),
+        _elapsed_ms(started),
+    )
+    return data
 
 
 def _request_form(
     method: str,
     url: str,
     *,
+    vendor: str,
+    operation: str,
     expected_status: tuple[int, ...] = (200,),
     **kwargs: Any,
 ) -> dict[str, Any]:
-    try:
-        resp = httpx.request(
-            method,
-            url,
-            timeout=settings.PUSH_REQUEST_TIMEOUT_SEC,
-            **kwargs,
-        )
-    except httpx.HTTPError as e:
-        raise VendorPushError(f"push http error: {e}") from e
-
-    if resp.status_code not in expected_status:
-        logger.error("vendor push http failed url=%s http=%s body=%s", url, resp.status_code, resp.text[:500])
-        raise VendorPushError(f"push http {resp.status_code}")
-    try:
-        return resp.json()
-    except ValueError as e:
-        raise VendorPushError(f"push invalid json: {resp.text[:200]}") from e
+    return _request_json(
+        method,
+        url,
+        vendor=vendor,
+        operation=operation,
+        expected_status=expected_status,
+        **kwargs,
+    )
 
 
 def _cache_get(key: str) -> str | None:
@@ -234,6 +492,8 @@ def _honor_access_token() -> str:
     data = _request_form(
         "POST",
         settings.HONOR_PUSH_TOKEN_ENDPOINT,
+        vendor="honor",
+        operation="authenticate",
         data={
             "grant_type": "client_credentials",
             "client_id": client_id,
@@ -243,25 +503,31 @@ def _honor_access_token() -> str:
     )
     token = data.get("access_token")
     if not token:
-        raise VendorPushError(f"honor auth failed: {data}")
+        raise VendorPushError(f"honor auth failed: {_raw_log_summary(data)}")
     return _cache_put(cache_key, str(token), data.get("expires_in"))
 
 
 def _send_honor(token: str, title: str, body: str, extras: dict[str, str]) -> dict[str, Any]:
     app_id = _require(settings.HONOR_PUSH_APP_ID, "HONOR_PUSH_APP_ID")
     access_token = _honor_access_token()
+    click_action: dict[str, Any]
+    if extras.get("notificationId"):
+        # type=3 只负责打开首页，不保证 data 会变成目标 Activity 的 extras。消息中心通知必须
+        # 用自定义页面(type=1)+ intent URI，把 notificationId/type/feedbackId 等直接带给
+        # MainActivity；否则华为/荣耀系统能展示通知，但用户点击后客户端收不到任何导航参数。
+        click_action = {"type": 1, "intent": _click_intent_uri(extras)}
+    else:
+        click_action = {"type": 3}
     payload = {
-        # clickAction type=3(打开应用首页)时,荣耀点击会把 data JSON 的键值对注入启动 intent 的
-        # extras(与 HMS 同机制)→ MainActivity.consumeNavTarget 读 notif_id/notif_type 直达落地。
         "data": json.dumps(_click_extras(extras), ensure_ascii=False),
         "notification": {"title": title, "body": body},
         "android": {
             "ttl": f"{settings.PUSH_TIME_TO_LIVE_SEC}s",
-            "targetUserType": 1,
+            "targetUserType": settings.HONOR_PUSH_TARGET_USER_TYPE,
             "notification": {
                 "title": title,
                 "body": body,
-                "clickAction": {"type": 3},
+                "clickAction": click_action,
                 "importance": "NORMAL",
             },
         },
@@ -270,6 +536,8 @@ def _send_honor(token: str, title: str, body: str, extras: dict[str, str]) -> di
     data = _request_json(
         "POST",
         settings.HONOR_PUSH_SEND_ENDPOINT_TEMPLATE.format(app_id=app_id),
+        vendor="honor",
+        operation="send_notification",
         json=payload,
         headers={
             "Content-Type": "application/json; charset=UTF-8",
@@ -279,7 +547,28 @@ def _send_honor(token: str, title: str, body: str, extras: dict[str, str]) -> di
     )
     code = data.get("code")
     if code is not None and int(code) != 200:
-        raise VendorPushError(f"honor push failed: {data}")
+        raise VendorPushError(f"honor push failed: {_raw_log_summary(data)}")
+    return data
+
+
+def _send_honor_data(token: str, payload: dict[str, str]) -> dict[str, Any]:
+    app_id = _require(settings.HONOR_PUSH_APP_ID, "HONOR_PUSH_APP_ID")
+    access_token = _honor_access_token()
+    data = _request_json(
+        "POST",
+        settings.HONOR_PUSH_SEND_ENDPOINT_TEMPLATE.format(app_id=app_id),
+        vendor="honor",
+        operation="send_data_event",
+        json={"data": json.dumps(payload, ensure_ascii=False), "token": [token]},
+        headers={
+            "Content-Type": "application/json; charset=UTF-8",
+            "Authorization": f"Bearer {access_token}",
+            "timestamp": str(int(time.time() * 1000)),
+        },
+    )
+    code = data.get("code")
+    if code is not None and int(code) != 200:
+        raise VendorPushError(f"honor data push failed: {_raw_log_summary(data)}")
     return data
 
 
@@ -294,6 +583,8 @@ def _huawei_access_token() -> str:
     data = _request_form(
         "POST",
         settings.HUAWEI_PUSH_TOKEN_ENDPOINT,
+        vendor="huawei",
+        operation="authenticate",
         data={
             "grant_type": "client_credentials",
             "client_id": app_id,
@@ -303,7 +594,7 @@ def _huawei_access_token() -> str:
     )
     token = data.get("access_token")
     if not token:
-        raise VendorPushError(f"huawei auth failed: {data}")
+        raise VendorPushError(f"huawei auth failed: {_raw_log_summary(data)}")
     return _cache_put(cache_key, str(token), data.get("expires_in"))
 
 
@@ -312,18 +603,24 @@ def _send_huawei(token: str, title: str, body: str, extras: dict[str, str]) -> d
     '80100000' 为部分成功(单 token 场景仍视为失败,错误里带原始响应便于排障)。"""
     app_id = _require(settings.HUAWEI_PUSH_APP_ID, "HUAWEI_PUSH_APP_ID")
     access_token = _huawei_access_token()
+    click_action: dict[str, Any]
+    if extras.get("notificationId"):
+        # type=3 只打开首页，Mate 20/EMUI 不会把 message.data 自动拆成启动 Intent extras。
+        # type=1 的自定义 intent 才能稳定携带反馈记录 id，且冷启动/onNewIntent 都走同一路由。
+        click_action = {"type": 1, "intent": _click_intent_uri(extras)}
+    else:
+        click_action = {"type": 3}
     payload = {
         "validate_only": False,
         "message": {
-            # click_action type=3(打开应用首页)时,HMS 点击会把 data JSON 的键值对注入启动 intent
-            # 的 extras → MainActivity.consumeNavTarget 读 notif_id/notif_type 直达落地。
             "data": json.dumps(_click_extras(extras), ensure_ascii=False),
             "android": {
+                "target_user_type": settings.HUAWEI_PUSH_TARGET_USER_TYPE,
                 "ttl": f"{settings.PUSH_TIME_TO_LIVE_SEC}s",
                 "notification": {
                     "title": title,
                     "body": body,
-                    "click_action": {"type": 3},
+                    "click_action": click_action,
                     "importance": "NORMAL",
                 },
             },
@@ -333,6 +630,8 @@ def _send_huawei(token: str, title: str, body: str, extras: dict[str, str]) -> d
     data = _request_json(
         "POST",
         settings.HUAWEI_PUSH_SEND_ENDPOINT_TEMPLATE.format(app_id=app_id),
+        vendor="huawei",
+        operation="send_notification",
         json=payload,
         headers={
             "Content-Type": "application/json; charset=UTF-8",
@@ -340,7 +639,32 @@ def _send_huawei(token: str, title: str, body: str, extras: dict[str, str]) -> d
         },
     )
     if str(data.get("code", "")) != "80000000":
-        raise VendorPushError(f"huawei push failed: {data}")
+        raise VendorPushError(f"huawei push failed: {_raw_log_summary(data)}")
+    return data
+
+
+def _send_huawei_data(token: str, payload: dict[str, str]) -> dict[str, Any]:
+    app_id = _require(settings.HUAWEI_PUSH_APP_ID, "HUAWEI_PUSH_APP_ID")
+    access_token = _huawei_access_token()
+    data = _request_json(
+        "POST",
+        settings.HUAWEI_PUSH_SEND_ENDPOINT_TEMPLATE.format(app_id=app_id),
+        vendor="huawei",
+        operation="send_data_event",
+        json={
+            "validate_only": False,
+            "message": {
+                "data": json.dumps(payload, ensure_ascii=False),
+                "token": [token],
+            },
+        },
+        headers={
+            "Content-Type": "application/json; charset=UTF-8",
+            "Authorization": f"Bearer {access_token}",
+        },
+    )
+    if str(data.get("code", "")) != "80000000":
+        raise VendorPushError(f"huawei data push failed: {_raw_log_summary(data)}")
     return data
 
 
@@ -357,6 +681,8 @@ def _vivo_auth_token() -> str:
     data = _request_json(
         "POST",
         settings.VIVO_PUSH_AUTH_ENDPOINT,
+        vendor="vivo",
+        operation="authenticate",
         json={
             "appId": app_id,
             "appKey": app_key,
@@ -366,10 +692,10 @@ def _vivo_auth_token() -> str:
         headers={"Content-Type": "application/json"},
     )
     if int(data.get("result", -1)) != 0:
-        raise VendorPushError(f"vivo auth failed: {data}")
+        raise VendorPushError(f"vivo auth failed: {_raw_log_summary(data)}")
     token = data.get("authToken")
     if not token:
-        raise VendorPushError(f"vivo auth missing authToken: {data}")
+        raise VendorPushError(f"vivo auth missing authToken: {_raw_log_summary(data)}")
     return _cache_put(cache_key, str(token), 24 * 3600)
 
 
@@ -385,15 +711,22 @@ def _send_vivo(token: str, title: str, body: str, extras: dict[str, str]) -> dic
         "timeToLive": settings.PUSH_TIME_TO_LIVE_SEC,
         "requestId": uuid.uuid4().hex,
         "pushMode": settings.VIVO_PUSH_MODE,
+        # vivo SDK 4.1.5 只有在前台展示关闭时才调用
+        # OpenClientPushMessageReceiver.onForegroundMessageArrived。
+        # App 在该回调中刷新服务端未读数并补发一条本地通知；后台/锁屏时仍由
+        # vivo 系统展示，因而任何场景都只会出现一条通知。
+        "foregroundShow": False,
+        # vivo 官方 VPush 角标字段：离线收到通知时先由桌面自动 +1；
+        # App 启动/同步后再由统一未读数通过系统 API 精确校准。
+        "addBadge": True,
         "clientCustomMap": extras,
     }
-    # 点击落地:消息中心推送(带 notificationId)→ skipType=4 + skipContent=intent uri,由 vivo
-    # 系统直启 MainActivity 并携带 S. extras(与小米 notify_effect=2 同机制)。不依赖客户端
-    # VivoPushReceiver.onNotificationMessageClicked 里的后台 startActivity——Android 10+ BAL
-    # 会静默拦掉,receiver 路径仅作兜底。无 notificationId 的召回类保持 skipType=1 仅打开首页。
+    # vivo Push SDK 480+ 已不再回调 skipType=3；必须使用 skipType=4 的完整 Intent URI。
+    # URI 同时包含 data/scheme、显式 component 和 S. extras，OriginOS 才会把业务参数
+    # 原样交给 MainActivity（仅靠隐式 deeplink 会打开 App，但可能剥掉 extras）。
     if extras.get("notificationId"):
         payload["skipType"] = 4
-        payload["skipContent"] = _click_intent_uri(extras)
+        payload["skipContent"] = _vivo_click_intent_uri(extras)
     else:
         payload["skipType"] = 1
     if settings.VIVO_PUSH_CATEGORY:
@@ -401,14 +734,37 @@ def _send_vivo(token: str, title: str, body: str, extras: dict[str, str]) -> dic
     data = _request_json(
         "POST",
         settings.VIVO_PUSH_SEND_ENDPOINT,
+        vendor="vivo",
+        operation="send_notification",
         json=payload,
         headers={
             "Content-Type": "application/json",
             "authToken": auth_token,
         },
     )
+    # 10089 = 当前 vivo 应用尚未开通 VPush「离线自动角标」能力。
+    # 角标只是通知中心未读数的镜像，不能因此让整条通知发送失败：去掉 addBadge
+    # 原样重试，通知到达/应用同步后仍由客户端系统 API 按统一未读数精确设置。
+    if int(data.get("result", -1)) == 10089:
+        logger.warning("vivo VPush badge not enabled (10089); retrying without addBadge")
+        fallback_payload = dict(payload)
+        fallback_payload.pop("addBadge", None)
+        fallback_payload["requestId"] = uuid.uuid4().hex
+        data = _request_json(
+            "POST",
+            settings.VIVO_PUSH_SEND_ENDPOINT,
+            vendor="vivo",
+            operation="send_notification",
+            json=fallback_payload,
+            headers={
+                "Content-Type": "application/json",
+                "authToken": auth_token,
+            },
+        )
+        if int(data.get("result", -1)) == 0:
+            data = {**data, "badgeFallback": True}
     if int(data.get("result", -1)) != 0:
-        raise VendorPushError(f"vivo push failed: {data}")
+        raise VendorPushError(f"vivo push failed: {_raw_log_summary(data)}")
     return data
 
 
@@ -446,14 +802,43 @@ def _send_xiaomi(token: str, title: str, body: str, extras: dict[str, str]) -> d
     data = _request_form(
         "POST",
         settings.XIAOMI_PUSH_SEND_ENDPOINT,
+        vendor="xiaomi",
+        operation="send_notification",
         data=form,
         headers={"Authorization": f"key={app_secret}"},
     )
     code = data.get("code")
     if code not in (0, "0", None):
-        raise VendorPushError(f"xiaomi push failed: {data}")
+        raise VendorPushError(f"xiaomi push failed: {_raw_log_summary(data)}")
     if str(data.get("result", "ok")).lower() not in ("ok", "success"):
-        raise VendorPushError(f"xiaomi push failed: {data}")
+        raise VendorPushError(f"xiaomi push failed: {_raw_log_summary(data)}")
+    return data
+
+
+def _send_xiaomi_data(token: str, payload: dict[str, str]) -> dict[str, Any]:
+    app_secret = _require(settings.XIAOMI_PUSH_APP_SECRET, "XIAOMI_PUSH_APP_SECRET")
+    form = {
+        "registration_id": token,
+        "restricted_package_name": settings.ANDROID_PACKAGE_NAME,
+        "payload": json.dumps(payload, ensure_ascii=False),
+        "pass_through": "1",
+        "time_to_live": str(settings.PUSH_TIME_TO_LIVE_SEC * 1000),
+    }
+    if settings.XIAOMI_PUSH_CHANNEL_ID.strip():
+        form["extra.channel_id"] = settings.XIAOMI_PUSH_CHANNEL_ID.strip()
+    data = _request_form(
+        "POST",
+        settings.XIAOMI_PUSH_SEND_ENDPOINT,
+        vendor="xiaomi",
+        operation="send_data_event",
+        data=form,
+        headers={"Authorization": f"key={app_secret}"},
+    )
+    code = data.get("code")
+    if code not in (0, "0", None):
+        raise VendorPushError(f"xiaomi data push failed: {_raw_log_summary(data)}")
+    if str(data.get("result", "ok")).lower() not in ("ok", "success"):
+        raise VendorPushError(f"xiaomi data push failed: {_raw_log_summary(data)}")
     return data
 
 
@@ -490,6 +875,24 @@ def _click_intent_uri(extras: dict[str, str]) -> str:
     return ";".join(parts)
 
 
+def _vivo_click_intent_uri(extras: dict[str, str]) -> str:
+    """vivo skipType=4 使用官方要求的完整 intent deeplink 形态。
+
+    `?#Intent` 中的问号不能省；OriginOS 对缺少它的 URI 会退化为“仅打开应用”，
+    不把 S. 参数交给目标 Activity。
+    """
+    pkg = settings.ANDROID_PACKAGE_NAME
+    parts = [
+        "intent://push/detail?#Intent",
+        "scheme=shaguabijia",
+        f"component={pkg}/{pkg}.MainActivity",
+        "launchFlags=0x14000000",
+    ]
+    parts += [f"S.{key}={quote(str(value), safe='')}" for key, value in _click_extras(extras).items()]
+    parts.append("end")
+    return ";".join(parts)
+
+
 def _xiaomi_template_param(title: str, alert: str) -> str:
     rendered = (
         settings.XIAOMI_PUSH_TEMPLATE_PARAM_JSON
@@ -522,6 +925,8 @@ def _oppo_auth_token() -> str:
     data = _request_form(
         "POST",
         settings.OPPO_PUSH_AUTH_ENDPOINT,
+        vendor="oppo",
+        operation="authenticate",
         data={
             "app_key": app_key,
             "timestamp": timestamp,
@@ -530,10 +935,10 @@ def _oppo_auth_token() -> str:
         headers={"Content-Type": "application/x-www-form-urlencoded"},
     )
     if int(data.get("code", -1)) != 0:
-        raise VendorPushError(f"oppo auth failed: {data}")
+        raise VendorPushError(f"oppo auth failed: {_raw_log_summary(data)}")
     token = (data.get("data") or {}).get("auth_token") or data.get("auth_token")
     if not token:
-        raise VendorPushError(f"oppo auth missing auth_token: {data}")
+        raise VendorPushError(f"oppo auth missing auth_token: {_raw_log_summary(data)}")
     return _cache_put(cache_key, str(token), 24 * 3600)
 
 
@@ -548,6 +953,11 @@ def _send_oppo(token: str, title: str, body: str, extras: dict[str, str]) -> dic
         "off_line_ttl": ttl_hours,
         "action_parameters": json.dumps(_click_extras(extras), ensure_ascii=False),
     }
+    if extras.get("notificationId"):
+        # 只有已落库的站内消息才计入角标；保活提醒等临时推送不应留下无法消除的未读数。
+        # 客户端打开/已读后会按服务端真实未读总数覆盖，避免长期漂移。
+        notification["badge_operation_type"] = 1
+        notification["badge_message_count"] = 1
     # 点击落地:OPPO SDK 没有点击回调,参数只能靠服务端点击动作配置送达——action_parameters 的
     # 键值对仅在 click_action_type=1/4 时才会注入目标 Activity 的 intent extras(type=0「启动应用」
     # 会忽略它,extras 全丢 → 点了没反应,与小米 notify_effect=1 同款坑)。
@@ -573,6 +983,8 @@ def _send_oppo(token: str, title: str, body: str, extras: dict[str, str]) -> dic
     data = _request_form(
         "POST",
         settings.OPPO_PUSH_SEND_ENDPOINT,
+        vendor="oppo",
+        operation="send_notification",
         data={
             "auth_token": auth_token,
             "message": json.dumps(message, ensure_ascii=False),
@@ -580,5 +992,5 @@ def _send_oppo(token: str, title: str, body: str, extras: dict[str, str]) -> dic
         headers={"Content-Type": "application/x-www-form-urlencoded"},
     )
     if int(data.get("code", -1)) != 0:
-        raise VendorPushError(f"oppo push failed: {data}")
+        raise VendorPushError(f"oppo push failed: {_raw_log_summary(data)}")
     return data

app/models/__init__.py

@@ -36,6 +36,7 @@ from app.models.inactivity import (  # noqa: F401
 from app.models.invite import InviteRelation  # noqa: F401
 from app.models.invite_fingerprint import InviteFingerprint  # noqa: F401
 from app.models.launch_confirm_sample import LaunchConfirmSample  # noqa: F401
+from app.models.limit_policy import LimitPolicyOverride  # noqa: F401
 from app.models.meituan_coupon import MeituanCoupon  # noqa: F401
 from app.models.notification import Notification  # noqa: F401
 from app.models.onboarding import OnboardingCompletion  # noqa: F401

app/models/comparison.py

@@ -97,7 +97,7 @@ class ComparisonRecord(Base):
     total_dish_count: Mapped[int | None] = mapped_column(Integer, nullable=True)
     skipped_dish_count: Mapped[int | None] = mapped_column(Integer, nullable=True)
 
-    # success(拿到有效对比)/ failed(出错或没采到目标价)
+    # success(流程正常完成,含 below_minimum)/ failed(技术异常或未形成可信结论)
     status: Mapped[str] = mapped_column(String(16), nullable=False, default="success")
     # done 帧 information 文案。成功:"在美团找到同店,到手价 ¥X…";
     # 失败:具体原因(如"美团、京东外卖均未找到该商品")。前端在比价失败时当原因展示。
@@ -113,6 +113,10 @@ class ComparisonRecord(Base):
     items: Mapped[list] = mapped_column(_JSON, nullable=False, default=list)
     # 逐平台对比 [{platform_id, platform_name, package, price, is_source, rank, coupon_saved, coupon_name, applied_coupons}](price/coupon_saved 单位:元,原样存;coupon_name=优惠来源名;applied_coupons=[{name,amount}] 多券明细)
     comparison_results: Mapped[list] = mapped_column(_JSON, nullable=False, default=list)
+    # 展示模型统一数组(pricebot done.params.platforms 原样存): 每平台一行、自带
+    # status/is_best/display/display_order, 记录页据此直接渲染, 不再靠 comparison_results
+    # + 客户端合并 + 前端派生。老记录/旧客户端为空 → 前端回退老 comparison_results 渲染。
+    platforms: Mapped[list] = mapped_column(_JSON, nullable=False, default=list)
     # 目标平台未找到、跳过的菜名
     skipped_dish_names: Mapped[list] = mapped_column(_JSON, nullable=False, default=list)
     # 客户端上报的原始 payload(calibration + done.params 全量),未来取数兜底

app/models/device.py

@@ -34,6 +34,11 @@ class DeviceLiveness(Base):
     __tablename__ = "device_liveness"
     __table_args__ = (
         UniqueConstraint("user_id", "device_id", name="uq_device_liveness_user_device"),
+        UniqueConstraint(
+            "push_vendor",
+            "push_token",
+            name="uq_device_liveness_vendor_token",
+        ),
     )
 
     id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
@@ -48,6 +53,11 @@ class DeviceLiveness(Base):
     push_vendor: Mapped[str | None] = mapped_column(String(32), nullable=True)
     # 厂商 push token / regId / registration_id;不同厂商命名不同,后端统一存这里。
     push_token: Mapped[str | None] = mapped_column(String(256), nullable=True)
+    # 客户端每次登录生成的高熵随机值。离线退出后只凭此值精确撤销旧会话绑定，
+    # 新账号/新会话注册时会替换它，故旧解绑任务绝不能误解绑新登录。
+    push_binding_id: Mapped[str | None] = mapped_column(String(128), index=True, nullable=True)
+    # 被退出/转移的 binding 留作墓碑，阻止旧的在途 register/heartbeat 把 token 抢回。
+    push_binding_revoked: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)
     platform: Mapped[str] = mapped_column(String(16), nullable=False, default="android")
     app_version: Mapped[str | None] = mapped_column(String(32), nullable=True)
 

app/models/guide_video.py

@@ -1,20 +1,4 @@
-"""新手引导视频播放记录(领券浮层前 N 次用它替代广告)。
-
-产品规则(2026-07 拍板):新用户点「一键自动领取」后的等候浮层,**前 3 次**不放广告,
-改放运营后台上传的引导视频;每次固定 120 金币,中途关闭也算看完照发。
-
-口径:
-- **计次按账号**(user_id),与设备无关 —— 换设备不重新送 3 次。
-- **开播即计数**:客户端每次要展示浮层时调 `/api/v1/guide-video/start`,服务端当场
-  写一行(status='playing')并返回 play_token;`COUNT(*)` 即已用次数。用户中途 kill
-  App 也算用掉一次(产品选定口径,防反复进出刷金币)。
-- **发币幂等**靠 play_token 定位 + `status='playing'` 条件更新:并发两次上报只有一次
-  改到行(另一次 rowcount=0),所以只发一次币。光有 play_token 唯一键挡不住 —— 发币走的是
-  UPDATE,不 INSERT,撞不到任何唯一键。
-- **次数上限**靠 (user_id, seq) 唯一键兜底,防并发 /start 绕过 COUNT 判定(见下)。
-
-与广告收益(ad_feed_reward_record)彻底分离:引导视频不是广告,不该进广告收益报表。
-"""
+"""引导视频待播放计划、起播快照与逐圈结算状态。"""
 from __future__ import annotations
 
 from datetime import datetime
@@ -26,48 +10,48 @@ from app.db.base import Base
 
 
 class GuideVideoPlay(Base):
-    """一次引导视频播放一行。开播时建(status='playing'),发币后置 'granted'。"""
-
     __tablename__ = "guide_video_play"
     __table_args__ = (
-        # 客户端幂等键:同一次播放重复上报奖励只发一次。
         UniqueConstraint("play_token", name="uq_guide_video_play_token"),
-        # 次数上限的**硬约束**:start_play 是无锁 check-then-insert(读 COUNT 算 seq 再插),
-        # N 个并发 /start 会都读到同一个已用次数、算出同一个 seq,不拦就能各拿一个 token、
-        # 各发一次金币,3 次上限形同虚设(改包即可无限刷)。seq 唯一 → 并发同 seq 必撞,
-        # start_play 捕获 IntegrityError 降级成"这次不放视频"。
-        # 用 unique Index 而非 UniqueConstraint:与迁移里的 create_index 对齐(SQLite 加约束
-        # 要整表重建),autogenerate 才不会每次报一条假 diff。
-        Index("uq_guide_video_play_user_seq", "user_id", "seq", unique=True),
+        Index(
+            "uq_guide_video_play_user_scene_seq",
+            "user_id",
+            "scene",
+            "seq",
+            unique=True,
+        ),
     )
 
     id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
     user_id: Mapped[int] = mapped_column(
         Integer, ForeignKey("user.id"), index=True, nullable=False
     )
-    # 服务端生成下发给客户端的幂等键(uuid hex)。
     play_token: Mapped[str] = mapped_column(String(64), nullable=False)
-    # 触发场景:目前只有 coupon(领券等候浮层);留字段以便日后比价等场景复用。
     scene: Mapped[str] = mapped_column(String(16), nullable=False, default="coupon")
-    # 本账号第几次(1-based),= 建行时已有行数 + 1。日常判定仍以 COUNT 为准,但 (user_id, seq)
-    # 唯一键让并发 /start 只能成一个 —— 见 __table_args__。
-    seq: Mapped[int] = mapped_column(Integer, nullable=False, default=1)
-    # 当次下发的视频地址(运营换片后能回溯用户当时看的是哪支)。
+    # prepare 不占次数，seq=NULL；start 成功才写入 1-based seq。
+    seq: Mapped[int | None] = mapped_column(Integer, nullable=True)
     video_url: Mapped[str | None] = mapped_column(String(512), nullable=True)
-    # 实发金币;未发时 0。
     coin: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
-    # playing(已开播未发币) / granted(已发币)。
-    status: Mapped[str] = mapped_column(String(16), nullable=False, default="playing")
-    # 客户端上报时是否播完(true=自然播完 / false=中途关闭)。仅留痕:两者都发币。
+    duration_ms: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    config_version: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    settled_circles: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
+    # prepared / started / completed / legacy_completed / legacy_closed
+    status: Mapped[str] = mapped_column(String(24), nullable=False, default="prepared")
     completed: Mapped[int] = mapped_column(Integer, nullable=False, default=0)
 
-    started_at: Mapped[datetime] = mapped_column(
-        DateTime(timezone=True), server_default=func.now(), index=True, nullable=False
+    prepared_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), server_default=func.now(), nullable=False
+    )
+    expires_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), index=True, nullable=True
+    )
+    started_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), index=True, nullable=True
     )
     granted_at: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
 
     def __repr__(self) -> str:  # pragma: no cover
         return (
-            f"<GuideVideoPlay user={self.user_id} seq={self.seq} "
-            f"{self.status} coin={self.coin}>"
+            f"<GuideVideoPlay user={self.user_id} scene={self.scene} seq={self.seq} "
+            f"{self.status} circles={self.settled_circles}>"
         )

app/models/limit_policy.py

@@ -0,0 +1,77 @@
+"""Per-subject limit policy overrides used by the admin whitelist page."""
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import (
+    Boolean,
+    DateTime,
+    Index,
+    Integer,
+    String,
+    UniqueConstraint,
+    func,
+    true,
+)
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.db.base import Base
+
+
+class LimitPolicyOverride(Base):
+    """One rule override for one phone or device.
+
+    ``reset_at`` is a non-destructive usage baseline.  Business records and
+    security events remain intact; quota readers only count rows at or after
+    this timestamp.
+    """
+
+    __tablename__ = "limit_policy_override"
+    __table_args__ = (
+        UniqueConstraint(
+            "subject_type",
+            "subject_value",
+            "rule_code",
+            name="uq_limit_policy_subject_rule",
+        ),
+        Index(
+            "ix_limit_policy_lookup",
+            "subject_type",
+            "subject_value",
+            "rule_code",
+            "enabled",
+        ),
+        Index("ix_limit_policy_expires", "expires_at"),
+    )
+
+    id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)
+    subject_type: Mapped[str] = mapped_column(String(16), nullable=False)
+    subject_value: Mapped[str] = mapped_column(String(128), nullable=False)
+    rule_code: Mapped[str] = mapped_column(String(64), nullable=False)
+    # 产品只保留“临时不限/免告警”。即使有内部脚本绕过 API 直接建 ORM
+    # 对象，也不能再悄悄落成已经下线的 override 模式。
+    mode: Mapped[str] = mapped_column(String(24), nullable=False, default="unlimited")
+    limit_value: Mapped[int | None] = mapped_column(Integer, nullable=True)
+    enabled: Mapped[bool] = mapped_column(
+        Boolean, nullable=False, default=True, server_default=true()
+    )
+    starts_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )
+    expires_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )
+    reset_at: Mapped[datetime | None] = mapped_column(
+        DateTime(timezone=True), nullable=True
+    )
+    reason: Mapped[str | None] = mapped_column(String(256), nullable=True)
+    created_by_admin_id: Mapped[int | None] = mapped_column(Integer, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True), server_default=func.now(), nullable=False
+    )
+    updated_at: Mapped[datetime] = mapped_column(
+        DateTime(timezone=True),
+        server_default=func.now(),
+        onupdate=func.now(),
+        nullable=False,
+    )

app/models/wallet.py

@@ -60,6 +60,15 @@ class CoinTransaction(Base):
             sqlite_where=text("biz_type LIKE 'task%' AND ref_id IS NOT NULL"),
             postgresql_where=text("biz_type LIKE 'task%' AND ref_id IS NOT NULL"),
         ),
+        Index(
+            "ux_coin_transaction_guide_video_ref",
+            "user_id",
+            "biz_type",
+            "ref_id",
+            unique=True,
+            sqlite_where=text("biz_type = 'guide_video' AND ref_id IS NOT NULL"),
+            postgresql_where=text("biz_type = 'guide_video' AND ref_id IS NOT NULL"),
+        ),
     )
 
     id: Mapped[int] = mapped_column(Integer, primary_key=True, autoincrement=True)

app/repositories/ad_feed_reward.py

@@ -5,11 +5,13 @@
 """
 from __future__ import annotations
 
+from datetime import datetime
+
 from sqlalchemy import func, select
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
-from app.core import rewards
+from app.core import limit_policy, rewards
 from app.core.rewards import cn_today
 from app.models.ad_feed_reward import AdFeedRewardRecord
 from app.repositories import wallet as crud_wallet
@@ -29,8 +31,14 @@ def _find_by_event(db: Session, client_event_id: str) -> AdFeedRewardRecord | No
     ).scalar_one_or_none()
 
 
-def _granted_today(db: Session, user_id: int, reward_date: str) -> int:
-    return db.execute(
+def _granted_today(
+    db: Session,
+    user_id: int,
+    reward_date: str,
+    *,
+    reset_at: datetime | None = None,
+) -> int:
+    stmt = (
         select(func.count())
         .select_from(AdFeedRewardRecord)
         .where(
@@ -38,7 +46,10 @@ def _granted_today(db: Session, user_id: int, reward_date: str) -> int:
             AdFeedRewardRecord.reward_date == reward_date,
             AdFeedRewardRecord.status == "granted",
         )
-    ).scalar_one()
+    )
+    if reset_at is not None:
+        stmt = stmt.where(AdFeedRewardRecord.created_at >= reset_at)
+    return db.execute(stmt).scalar_one()
 
 
 def granted_unit_total(db: Session, user_id: int) -> int:
@@ -121,7 +132,27 @@ def grant_feed_reward(
         )
         return _commit_record(db, rec, client_event_id)
 
-    if _granted_today(db, user_id, today) >= rewards.get_ad_daily_limit(db):
+    daily_policy = limit_policy.resolve_for_user(
+        db,
+        "ad.feed.daily",
+        user_id,
+    )
+    daily_limit = (
+        rewards.get_ad_daily_limit(db)
+        if daily_policy.override_id is None
+        and daily_policy.bucket_version == "default"
+        else daily_policy.limit
+    )
+    if (
+        daily_limit is not None
+        and _granted_today(
+            db,
+            user_id,
+            today,
+            reset_at=daily_policy.reset_at,
+        )
+        >= daily_limit
+    ):
         rec = AdFeedRewardRecord(
             client_event_id=client_event_id,
             user_id=user_id,

app/repositories/ad_reward.py

@@ -10,13 +10,13 @@
 """
 from __future__ import annotations
 
-from datetime import datetime, timezone
+from datetime import UTC, datetime
 
 from sqlalchemy import func, select
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
-from app.core import rewards
+from app.core import limit_policy, rewards
 from app.core.ad_cooldown import compute_cooldown
 from app.core.rewards import DAILY_AD_WATCH_SECONDS_LIMIT, cn_today
 from app.models.ad_reward import AdRewardRecord
@@ -96,8 +96,14 @@ def round_coin_total(db: Session, user_id: int, boost_round_id: str) -> int:
     )
 
 
-def _granted_today(db: Session, user_id: int, reward_date: str) -> int:
-    return db.execute(
+def _granted_today(
+    db: Session,
+    user_id: int,
+    reward_date: str,
+    *,
+    reset_at: datetime | None = None,
+) -> int:
+    stmt = (
         select(func.count())
         .select_from(AdRewardRecord)
         .where(
@@ -106,7 +112,10 @@ def _granted_today(db: Session, user_id: int, reward_date: str) -> int:
             AdRewardRecord.status == "granted",
             AdRewardRecord.reward_scene == "reward_video",
         )
-    ).scalar_one()
+    )
+    if reset_at is not None:
+        stmt = stmt.where(AdRewardRecord.created_at >= reset_at)
+    return db.execute(stmt).scalar_one()
 
 
 def _granted_cumulative(db: Session, user_id: int) -> int:
@@ -165,7 +174,27 @@ def grant_ad_reward(
         DAILY_AD_WATCH_SECONDS_LIMIT > 0
         and watched_seconds_today(db, user_id, today=today) >= DAILY_AD_WATCH_SECONDS_LIMIT
     )
-    over_count = _granted_today(db, user_id, today) >= rewards.get_ad_daily_limit(db)
+    daily_policy = limit_policy.resolve_for_user(
+        db,
+        "ad.reward_video.daily",
+        user_id,
+    )
+    daily_limit = (
+        rewards.get_ad_daily_limit(db)
+        if daily_policy.override_id is None
+        and daily_policy.bucket_version == "default"
+        else daily_policy.limit
+    )
+    over_count = (
+        daily_limit is not None
+        and _granted_today(
+            db,
+            user_id,
+            today,
+            reset_at=daily_policy.reset_at,
+        )
+        >= daily_limit
+    )
     if over_time or over_count:
         rec = AdRewardRecord(
             trans_id=trans_id, user_id=user_id, coin=0, status="capped",
@@ -321,19 +350,24 @@ def _commit_record(db: Session, rec: AdRewardRecord, trans_id: str) -> AdRewardR
     return rec
 
 
-def _granted_times_today_desc(db: Session, user_id: int, reward_date: str) -> list[datetime]:
+def _granted_times_today_desc(
+    db: Session,
+    user_id: int,
+    reward_date: str,
+    *,
+    reset_at: datetime | None = None,
+) -> list[datetime]:
     """当日 status=granted 记录的 created_at,按时间倒序(最新在前)——冷却策略的输入数据。"""
-    return list(
-        db.execute(
-            select(AdRewardRecord.created_at)
-            .where(
+    stmt = select(AdRewardRecord.created_at).where(
                 AdRewardRecord.user_id == user_id,
                 AdRewardRecord.reward_date == reward_date,
                 AdRewardRecord.status == "granted",
                 AdRewardRecord.reward_scene == "reward_video",
             )
-            .order_by(AdRewardRecord.created_at.desc())
-        ).scalars()
+    if reset_at is not None:
+        stmt = stmt.where(AdRewardRecord.created_at >= reset_at)
+    return list(
+        db.execute(stmt.order_by(AdRewardRecord.created_at.desc())).scalars()
     )
 
 
@@ -349,16 +383,47 @@ def today_status(
     旧客户端兼容,当前 DAILY_AD_WATCH_SECONDS_LIMIT=0 表示不启用时长闸。
     """
     today = cn_today().isoformat()
-    granted_desc = _granted_times_today_desc(db, user_id, today)
+    daily_policy = limit_policy.resolve_for_user(
+        db,
+        "ad.reward_video.daily",
+        user_id,
+    )
+    cooldown_policy = limit_policy.resolve_for_user(
+        db,
+        "ad.reward_video.cooldown",
+        user_id,
+    )
+    daily_limit = (
+        rewards.get_ad_daily_limit(db)
+        if daily_policy.override_id is None
+        and daily_policy.bucket_version == "default"
+        else daily_policy.limit
+    )
+    cooldown_seconds = (
+        rewards.get_ad_cooldown_sec(db)
+        if cooldown_policy.override_id is None
+        and cooldown_policy.bucket_version == "default"
+        else (cooldown_policy.limit or 0)
+    )
+    granted_desc = _granted_times_today_desc(
+        db,
+        user_id,
+        today,
+        reset_at=daily_policy.reset_at,
+    )
     state = compute_cooldown(
         granted_desc,
-        datetime.now(timezone.utc),
+        datetime.now(UTC),
         round_size=rewards.get_ad_round_count(db),
-        cooldown_seconds=rewards.get_ad_cooldown_sec(db),
+        cooldown_seconds=cooldown_seconds,
     )
     return (
         len(granted_desc),
-        rewards.get_ad_daily_limit(db),
+        (
+            daily_limit
+            if daily_limit is not None
+            else limit_policy.get_rule("ad.reward_video.daily").max_value
+        ),
         0,
         state.round_count,
         state.cooldown_until,

app/repositories/comparison.py

@@ -76,6 +76,19 @@ _BIZ_STATUS_PRIORITY = (
 )
 
 
+def _normalize_record_status(status: str | None) -> str | None:
+    """Map a completed below-minimum outcome onto persisted ``success``.
+
+    ``record_status`` describes the business outcome, while
+    ``comparison_record.status`` is also the completed-comparison flag used by
+    milestones, stats and idempotent rewards. Reaching the target cart and
+    conclusively reading that it is below the delivery minimum is a completed
+    comparison, not a technical failure. The granular outcome remains in
+    ``raw_payload`` and ``platform_results`` for result rendering.
+    """
+    return "success" if status == "below_minimum" else status
+
+
 def _store_closed_text(reason: str | None) -> str:
     """打烊/暂停营业/休息类 reason 常带脏店名元数据 → 只留结论,套简短模板。"""
     r = reason or ""
@@ -164,8 +177,10 @@ def _derive(payload: ComparisonRecordIn) -> dict:
 
     is_source_best = best.is_source if best is not None else None
 
-    # status:客户端显式给了就用;否则有"非源且有价"的结果=success,否则 failed
-    status = payload.status
+    # status:优先 pricebot record_status → 客户端显式 status → 兜底派生。
+    # below_minimum 是已形成可信结论的正常完成态,记录级归 success;细分结局仍完整保留在
+    # raw_payload/platform_results,供结果卡展示"未满起送"。
+    status = _normalize_record_status(payload.record_status or payload.status)
     if status is None:
         has_valid_target = any(
             (not r.is_source) and r.price is not None for r in results
@@ -197,16 +212,37 @@ def upsert_record(
     灰度期老客户端 POST /compare/record 走这条,与后端 harvest 按 trace_id reconcile;
     新客户端不再 POST(改由 compare.py 透传壳 harvest 落库)。
     """
-    derived = _derive(payload)
+    # 单源派生: 与 harvest_done 一致, payload 带 platforms 时从它派生(唯一真相源
+    # _derive_from_platforms), 老客户端不带 platforms 时回退 _derive(从 comparison_results)。
+    if payload.platforms:
+        derived = _derive_from_platforms(
+            payload.platforms, payload.record_status or payload.status
+        )
+        # 对齐 _derive 返回键(#189 fail_reason): 两路径 fields 键集一致, 覆盖已有行时不残留旧值
+        derived["fail_reason"] = (
+            _derive_fail_display(payload.information, payload.platform_results or {})
+            if derived["status"] == "failed"
+            else None
+        )
+        # 单源派生取自 platforms 源行(常无源平台元数据/店名)→ 空则用 payload 兜底不丢字段。
+        # 下面 fields 不再显式写这四个键, 统一由 derived 提供(否则 dict(store_name=..., **derived)
+        # 与 _derive_from_platforms 同名键撞键 TypeError)。
+        for _k in ("store_name", "source_platform_id", "source_platform_name", "source_package"):
+            if not derived.get(_k):
+                derived[_k] = getattr(payload, _k)
+    else:
+        derived = _derive(payload)
+        # _derive 只从 comparison_results 派生, 不含源平台四件套 / store_name → 从 payload 补,
+        # 与上面 platforms 分支键集对齐(fields 统一靠 **derived 提供这些列)。
+        for _k in ("store_name", "source_platform_id", "source_platform_name", "source_package"):
+            derived[_k] = getattr(payload, _k)
     items = [it.model_dump(exclude_none=True) for it in payload.items]
     fields = dict(
         device_id=payload.device_id,
         business_type=payload.business_type,
-        store_name=payload.store_name,
         product_names=_product_names_from_items(items),
-        source_platform_id=payload.source_platform_id,
-        source_platform_name=payload.source_platform_name,
-        source_package=payload.source_package,
+        # store_name / source_platform_id / source_platform_name / source_package 统一由
+        # derived 提供(见上方两分支补齐), 不在此显式写 —— 否则与 _derive_from_platforms 撞键。
         information=payload.information,
         best_deeplink=payload.best_deeplink,
         trace_url=payload.trace_url,
@@ -214,6 +250,7 @@ def upsert_record(
         skipped_dish_count=payload.skipped_dish_count,
         items=items,
         comparison_results=[r.model_dump() for r in payload.comparison_results],
+        platforms=list(payload.platforms or []),
         skipped_dish_names=list(payload.skipped_dish_names),
         # 客户端环境 / 性能(debug,客户端上报;旧客户端为 None)
         device_model=payload.device_model,
@@ -286,7 +323,8 @@ def upsert_record(
 
 
 def _derive_from_results(
-    results: list[dict], platform_results: dict | None = None
+    results: list[dict], platform_results: dict | None = None,
+    record_status: str | None = None,
 ) -> dict:
     """从 done 帧 comparison_results(pricebot 原始 dict 列表)派生结构化列。
     等价 _derive,但吃原始字段(is_source/price/rank/platform_id/store_name...)而非 pydantic 对象。
@@ -334,7 +372,64 @@ def _derive_from_results(
         "saved_amount_cents": saved_amount_cents,
         "is_source_best": best.get("is_source") if best else None,
         "store_name": (src_row or {}).get("store_name") or None,
-        "status": "success" if has_valid_target else "failed",
+        # below_minimum 已完成到购物车并形成可信结论,记录级计 success;细分结局仍在
+        # raw_payload/platform_results。旧 pricebot 未下发 record_status 时回退二态派生。
+        "status": _normalize_record_status(record_status)
+        or ("success" if has_valid_target else "failed"),
+    }
+
+
+def _derive_from_platforms(
+    platforms: list, record_status: str | None = None,
+) -> dict:
+    """从 done 帧 platforms(每平台一行、渲染就绪)派生结构化列——**单一真相源**。
+
+    best_* 直接取 platforms 里 is_best 的那一行、source_* 取 role=source 行,与前端读的
+    platforms 天然一致(不再像 _derive_from_results 那样从 comparison_results 二次评最优,
+    消除"标量列 vs platforms"双源不一致)。platforms 非空时优先走这里;老 pricebot 无
+    platforms 时调用方回退 _derive_from_results(向后兼容)。"""
+    rows = [p for p in (platforms or []) if isinstance(p, dict)]
+    src = next((p for p in rows if p.get("role") == "source"), None)
+    best = next((p for p in rows if p.get("is_best")), None)
+    if best is None:
+        # pricebot 没标 is_best(如全平台 has_dish_diff「相似替换/仅供参考」→ 不认定权威最低价)
+        # 但仍有有价行 → 兜底取有价行里最低价当参考 best,避免记录级 best_*/saved 整条落 NULL
+        # (否则首页价 0.00 / 记录页无最低红框 / 省额丢失 /「累计发现可省」漏计)。**候选含源**:
+        # 源常年全菜、价可信,源本身最便宜时 best 回落源(saved=0、is_source_best=True),与 _derive
+        # 「全目标缺菜回落源、不虚报省」同一语义;若排除源强选更贵目标,saved 会变负、倒扣「累计
+        # 发现可省」(get_stats 对 saved_amount_cents 求和不带 >0 过滤)。
+        priced = [p for p in rows if p.get("price") is not None]
+        if priced:
+            best = min(priced, key=lambda p: (p["price"], p.get("display_order") or 0))
+    source_price_cents = _yuan_to_cents(src.get("price")) if src else None
+    best_price_cents = _yuan_to_cents(best.get("price")) if best else None
+    saved_amount_cents = None
+    if source_price_cents is not None and best_price_cents is not None:
+        saved_amount_cents = source_price_cents - best_price_cents
+    has_valid_target = any(
+        p.get("role") != "source" and p.get("price") is not None for p in rows
+    )
+    # store_name: 优先源行; recompare 场景源平台自己当目标、源行被目标覆盖(pricebot
+    # _build_platform_rows 有意去重, platforms 无 role=source 行)→ 回退 best 行 → 首个有店名
+    # 的行(显示现场实际比到的店), 免得记录页店名空掉兜底显示成"比价"。正常比价有源行不走回退。
+    store_name = (
+        (src or {}).get("store_name")
+        or (best or {}).get("store_name")
+        or next((p.get("store_name") for p in rows if p.get("store_name")), None)
+    )
+    return {
+        "source_platform_id": (src or {}).get("platform_id"),
+        "source_platform_name": (src or {}).get("platform_name"),
+        "source_package": (src or {}).get("package"),
+        "source_price_cents": source_price_cents,
+        "best_platform_id": (best or {}).get("platform_id"),
+        "best_platform_name": (best or {}).get("platform_name"),
+        "best_price_cents": best_price_cents,
+        "saved_amount_cents": saved_amount_cents,
+        "is_source_best": (best.get("role") == "source") if best else None,
+        "store_name": store_name or None,
+        "status": _normalize_record_status(record_status)
+        or ("success" if has_valid_target else "failed"),
     }
 
 
@@ -366,6 +461,8 @@ def reserve_daily_start(
     business_type: str = "food",
     device_id: str | None = None,
     now: datetime | None = None,
+    limit: int | None = DAILY_COMPARE_START_LIMIT,
+    reset_at: datetime | None = None,
 ) -> tuple[ComparisonRecord, int]:
     """Atomically reserve one of a user's 100 Beijing-day comparison starts.
 
@@ -391,6 +488,11 @@ def reserve_daily_start(
         if existing_at.tzinfo is not None:
             existing_at = existing_at.astimezone(CN_TZ).replace(tzinfo=None)
         day_start = existing_at.replace(hour=0, minute=0, second=0, microsecond=0)
+        if reset_at is not None:
+            reset_start = reset_at
+            if reset_start.tzinfo is not None:
+                reset_start = reset_start.astimezone(CN_TZ).replace(tzinfo=None)
+            day_start = max(day_start, reset_start)
         day_end = day_start + timedelta(days=1)
         used = db.scalar(
             select(func.count(ComparisonRecord.id)).where(
@@ -405,6 +507,11 @@ def reserve_daily_start(
     if current.tzinfo is not None:
         current = current.astimezone(CN_TZ).replace(tzinfo=None)
     day_start = current.replace(hour=0, minute=0, second=0, microsecond=0)
+    if reset_at is not None:
+        reset_start = reset_at
+        if reset_start.tzinfo is not None:
+            reset_start = reset_start.astimezone(CN_TZ).replace(tzinfo=None)
+        day_start = max(day_start, reset_start)
     day_end = day_start + timedelta(days=1)
     used = db.scalar(
         select(func.count(ComparisonRecord.id)).where(
@@ -413,7 +520,7 @@ def reserve_daily_start(
             ComparisonRecord.created_at < day_end,
         )
     ) or 0
-    if used >= DAILY_COMPARE_START_LIMIT:
+    if limit is not None and used >= limit:
         raise DailyCompareStartLimitExceeded
 
     rec = ComparisonRecord(
@@ -493,7 +600,30 @@ def harvest_done(
     返回 (记录, 是否本次**新**落成 success)——供调用方据此幂等发一次邀请奖。
     行不存在(理论上帧0已建;防御)则新建。"""
     results = done_params.get("comparison_results") or []
-    derived = _derive_from_results(results, done_params.get("platform_results"))
+    # 展示模型统一数组(pricebot 新增, 每平台一行自带 status/is_best): 原样存, 记录页据此直渲染。
+    # record_status: 记录级业务结局(success/below_minimum/store_closed/failed)。其中
+    # below_minimum 是正常完成态,持久化 status 归 success,原值仍随 done_params 落 raw_payload。
+    platforms = done_params.get("platforms") or []
+    record_status = done_params.get("record_status")
+    # 单源派生: platforms(含 pricebot 权威 is_best)是唯一真相源, best_*/source_*/saved/status
+    # 全从它取 → 与前端读的 platforms 天然一致; 菜品也取 platforms 源行。老 pricebot 无
+    # platforms 时回退从 comparison_results 派生(向后兼容)。
+    if platforms:
+        derived = _derive_from_platforms(platforms, record_status)
+        # 菜品优先源行; recompare 无源行 → 回退 best 行 → 首个有菜品的行(同 store_name 回退)
+        _item_row = (
+            next((p for p in platforms if isinstance(p, dict) and p.get("role") == "source"), None)
+            or next((p for p in platforms if isinstance(p, dict) and p.get("is_best")), None)
+            or next((p for p in platforms if isinstance(p, dict) and p.get("items")), None)
+        )
+        items = (_item_row or {}).get("items") or []
+    else:
+        derived = _derive_from_results(
+            results, done_params.get("platform_results"), record_status
+        )
+        # pricebot 已把源单菜品塞进 comparison_results[源行].items
+        items = next((r.get("items") or [] for r in results if r.get("is_source")), [])
+    # 失败展示原因(#189): platforms / results 两个派生分支的 status 都可能 failed, 统一在此算
     fail_reason = (
         _derive_fail_display(
             done_params.get("information"), done_params.get("platform_results")
@@ -501,8 +631,6 @@ def harvest_done(
         if derived["status"] == "failed"
         else None
     )
-    # 菜品:pricebot 已把源单菜品塞进 comparison_results[源行].items
-    items = next((r.get("items") or [] for r in results if r.get("is_source")), [])
     fields = dict(
         business_type=business_type or "food",
         information=done_params.get("information") or None,
@@ -514,6 +642,7 @@ def harvest_done(
         skipped_dish_count=done_params.get("skipped_dish_count"),
         skipped_dish_names=list(done_params.get("skipped_dish_names") or []),
         comparison_results=results,
+        platforms=platforms,
         items=items,
         product_names=_product_names_from_items(items),
         raw_payload=done_params,

app/repositories/device.py

@@ -1,9 +1,9 @@
 """device 表读写(设备注册 / 心跳 / 超时扫描)。"""
 from __future__ import annotations
 
-from datetime import datetime, timedelta, timezone
+from datetime import UTC, datetime, timedelta
 
-from sqlalchemy import select
+from sqlalchemy import and_, or_, select
 from sqlalchemy.orm import Session
 
 from app.models.device import DeviceLiveness
@@ -24,13 +24,29 @@ def register_or_update(
     registration_id: str | None = None,
     push_vendor: str | None = None,
     push_token: str | None = None,
+    push_binding_id: str | None = None,
     platform: str = "android",
     app_version: str | None = None,
 ) -> DeviceLiveness:
-    """注册设备或更新其厂商 push token / 元信息。upsert by (user_id, device_id)。"""
+    """注册设备或更新推送目标；同一设备/token 的旧用户绑定在同一事务内失效。"""
     normalized_vendor = _normalize_push_vendor(push_vendor)
     normalized_token = push_token.strip() if push_token else None
+    normalized_binding = push_binding_id.strip() if push_binding_id else None
     device = _get(db, user_id=user_id, device_id=device_id)
+    stale_binding = bool(
+        device
+        and normalized_binding
+        and device.push_binding_id == normalized_binding
+        and device.push_binding_revoked
+    )
+    if not stale_binding:
+        _release_conflicting_push_bindings(
+            db,
+            user_id=user_id,
+            device_id=device_id,
+            push_vendor=normalized_vendor,
+            push_token=normalized_token,
+        )
     if device is None:
         device = DeviceLiveness(
             user_id=user_id,
@@ -38,17 +54,22 @@ def register_or_update(
             registration_id=registration_id,
             push_vendor=normalized_vendor,
             push_token=normalized_token,
+            push_binding_id=normalized_binding,
+            push_binding_revoked=False,
             platform=platform or "android",
             app_version=app_version,
         )
         db.add(device)
     else:
-        if registration_id:
+        if registration_id and not stale_binding:
             device.registration_id = registration_id
-        if normalized_vendor:
+        if normalized_vendor and not stale_binding:
             device.push_vendor = normalized_vendor
-        if normalized_token:
+        if normalized_token and not stale_binding:
             device.push_token = normalized_token
+        if normalized_binding and not stale_binding:
+            device.push_binding_id = normalized_binding
+            device.push_binding_revoked = False
         if platform:
             device.platform = platform
         if app_version:
@@ -67,26 +88,45 @@ def touch_heartbeat(
     registration_id: str | None = None,
     push_vendor: str | None = None,
     push_token: str | None = None,
+    push_binding_id: str | None = None,
 ) -> DeviceLiveness:
     """处理一次心跳(心跳也能自注册)。
 
     service 心跳或 accessibility_enabled=true 时,刷新存活并把状态机重置回 alive、
     清掉 notified_at(掉线恢复 → 下次再断才会再推一条)。
     """
-    now = datetime.now(timezone.utc)
+    now = datetime.now(UTC)
     device = _get(db, user_id=user_id, device_id=device_id)
+    normalized_vendor = _normalize_push_vendor(push_vendor)
+    normalized_token = push_token.strip() if push_token else None
+    normalized_binding = push_binding_id.strip() if push_binding_id else None
+    stale_binding = bool(
+        device
+        and normalized_binding
+        and device.push_binding_id == normalized_binding
+        and device.push_binding_revoked
+    )
+    if (normalized_token or normalized_binding) and not stale_binding:
+        _release_conflicting_push_bindings(
+            db,
+            user_id=user_id,
+            device_id=device_id,
+            push_vendor=normalized_vendor,
+            push_token=normalized_token,
+        )
     if device is None:
         device = DeviceLiveness(user_id=user_id, device_id=device_id)
         db.add(device)
 
-    if registration_id:
+    if registration_id and not stale_binding:
         device.registration_id = registration_id
-    normalized_vendor = _normalize_push_vendor(push_vendor)
-    normalized_token = push_token.strip() if push_token else None
-    if normalized_vendor:
+    if normalized_vendor and not stale_binding:
         device.push_vendor = normalized_vendor
-    if normalized_token:
+    if normalized_token and not stale_binding:
         device.push_token = normalized_token
+    if normalized_binding and not stale_binding:
+        device.push_binding_id = normalized_binding
+        device.push_binding_revoked = False
     device.last_report_protection_on = accessibility_enabled
 
     if accessibility_enabled:
@@ -107,7 +147,7 @@ def list_overdue(db: Session, *, timeout_minutes: int) -> list[DeviceLiveness]:
 
     即使没有厂商 token 也要检出,后续由 kill_alert_pending 走客户端进 App 后兜底提醒。
     """
-    cutoff = datetime.now(timezone.utc) - timedelta(minutes=timeout_minutes)
+    cutoff = datetime.now(UTC) - timedelta(minutes=timeout_minutes)
     stmt = select(DeviceLiveness).where(
         DeviceLiveness.ever_protected.is_(True),
         DeviceLiveness.liveness_state == "alive",
@@ -126,7 +166,7 @@ def mark_notified(db: Session, *, device_id_pk: int) -> None:
     device = db.get(DeviceLiveness, device_id_pk)
     if device is not None:
         device.liveness_state = "notified"
-        device.notified_at = datetime.now(timezone.utc)
+        device.notified_at = datetime.now(UTC)
         device.kill_alert_pending = True
         db.commit()
 
@@ -177,6 +217,69 @@ def list_push_targets(db: Session, *, user_id: int) -> list[DeviceLiveness]:
     return targets
 
 
+def unregister_push_binding(db: Session, *, user_id: int, device_id: str) -> None:
+    """当前用户主动退出：清推送归属但保留设备存活历史。"""
+    device = _get(db, user_id=user_id, device_id=device_id)
+    if device is not None:
+        _clear_push_binding(device)
+        db.commit()
+
+
+def unregister_push_binding_deferred(
+    db: Session, *, device_id: str, push_binding_id: str
+) -> None:
+    """离线退出补偿；只撤销完全匹配的旧登录会话，且始终幂等。"""
+    stmt = select(DeviceLiveness).where(
+        DeviceLiveness.device_id == device_id,
+        DeviceLiveness.push_binding_id == push_binding_id,
+    )
+    device = db.execute(stmt).scalar_one_or_none()
+    if device is not None:
+        _clear_push_binding(device)
+        db.commit()
+
+
+def _release_conflicting_push_bindings(
+    db: Session,
+    *,
+    user_id: int,
+    device_id: str,
+    push_vendor: str | None,
+    push_token: str | None,
+) -> None:
+    """把设备或 token 从其他记录转移走，当前 (user,device) 行除外。"""
+    conflicts = [DeviceLiveness.device_id == device_id]
+    if push_vendor and push_token:
+        conflicts.append(
+            and_(
+                DeviceLiveness.push_vendor == push_vendor,
+                DeviceLiveness.push_token == push_token,
+            )
+        )
+    stmt = select(DeviceLiveness).where(
+        or_(*conflicts),
+        ~and_(
+            DeviceLiveness.user_id == user_id,
+            DeviceLiveness.device_id == device_id,
+        ),
+    )
+    released = False
+    for old in db.execute(stmt).scalars():
+        _clear_push_binding(old)
+        released = True
+    if released:
+        # 唯一约束下必须先落旧行清理，再把同一 token 写给新行。
+        db.flush()
+
+
+def _clear_push_binding(device: DeviceLiveness) -> None:
+    device.registration_id = None
+    device.push_vendor = None
+    device.push_token = None
+    # binding_id 留作撤销墓碑；同一旧会话的在途请求不能重新认领 token。
+    device.push_binding_revoked = bool(device.push_binding_id)
+
+
 def _normalize_push_vendor(push_vendor: str | None) -> str | None:
     if not push_vendor:
         return None

app/repositories/guide_video.py

@@ -1,95 +1,136 @@
-"""新手引导视频:运营配置读写 + 播放计次 + 发币。
-
-**配置**(开关 / 视频地址 / 前几次 / 每次金币)整体作为一个 JSON 存进通用 app_config 表
-(key=coupon_guide_video),写法完全对齐 feedback_qr —— 不进 CONFIG_DEFS,所以不会污染
-系统配置页的通用列表,由本模块独占维护。
-
-**计次**按账号(user_id)、**开播即计数**:客户端每次要展示领券等候浮层时调
-`/api/v1/guide-video/start`,命中则当场写一行 guide_video_play(status='playing')。
-已用次数 = 该账号的行数,达到 max_plays 后不再下发,客户端改放广告(原逻辑)。
-COUNT 判定本身无锁,真正卡住次数上限的是 (user_id, seq) 唯一键:并发 /start 只能成一个。
-
-**发币**幂等键是 play_token,落地方式是 `status='playing' → 'granted'` 的**条件更新**:
-同一次播放重复上报只入账一次(网络重试 / 关闭与播完同时触发都靠它挡住)。
-中途关闭也照发 —— 产品拍板「中途关闭也算看完」。
-
-两处都是直接铸币的路径,改动前先看 `start_play` / `grant_play` 上的并发注释。
-"""
+"""双场景引导视频配置与 prepare/start/reward 状态机。"""
 from __future__ import annotations
 
 import uuid
-from datetime import datetime
+from datetime import datetime, timedelta
+from pathlib import Path
 from typing import Any
 
 from sqlalchemy import func, select, update
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
-from app.core import rewards
+from app.core import limit_policy, media, rewards
+from app.core.config import settings
 from app.models.app_config import AppConfig
 from app.models.guide_video import GuideVideoPlay
+from app.models.wallet import CoinAccount
 from app.repositories import wallet as crud_wallet
 
-_KEY = "coupon_guide_video"
-
-#: 金币流水 biz_type。客户端收益明细按它显示「新手引导视频奖励」。
-BIZ_TYPE = "guide_video"
-
-# 默认值 = 「运营还没配」时的行为:video_url 为空 → 一律不下发引导视频,浮层维持现状(放广告)。
-# 所以本功能上线后**不配视频就等于没上线**,不会影响存量用户。
-_DEFAULTS: dict[str, Any] = {
-    "enabled": True,
-    "video_url": None,   # None/空 = 未配片 → 不下发,浮层照旧放广告
-    "max_plays": 3,      # 每个账号前 N 次浮层放引导视频
-    "reward_coin": 120,  # 每次固定金币
+SCENES = ("coupon", "comparison")
+_KEY_BY_SCENE = {
+    "coupon": "coupon_guide_video",
+    "comparison": "comparison_guide_video",
 }
-
-_FIELDS = tuple(_DEFAULTS.keys())
-
-# 后台可配范围的护栏:防手滑把次数/金币填成天文数字(配置直接决定发币)。
+BIZ_TYPE = "guide_video"
+CIRCLE_COUNT = 10
+PLAN_TTL = timedelta(minutes=10)
+# 0 表示全局暂停播放；白名单页与旧引导视频配置页必须接受同一口径，
+# 否则在白名单页设为 0 后，旧页面连金币/开关等无关字段也无法保存。
+MIN_PLAYS = 0
 MAX_PLAYS_LIMIT = 50
+MIN_REWARD_COIN = 10
 REWARD_COIN_LIMIT = 10_000
 
+_DEFAULTS: dict[str, Any] = {
+    "enabled": False,
+    "video_url": None,
+    "max_plays": 3,
+    "reward_coin": 100,
+    "duration_ms": None,
+    "video_codec": None,
+    "audio_codec": None,
+    "analysis_status": "missing",
+    "analysis_error": None,
+    "config_version": 0,
+}
+_FIELDS = tuple(_DEFAULTS)
 
-# ===== 配置 =====
+
+class PlayStateError(Exception):
+    def __init__(self, code: str, message: str, *, reprepare_required: bool = False):
+        super().__init__(message)
+        self.code = code
+        self.message = message
+        self.reprepare_required = reprepare_required
+
+    def detail(self) -> dict[str, Any]:
+        return {
+            "code": self.code,
+            "message": self.message,
+            "retryable": False,
+            "reprepare_required": self.reprepare_required,
+        }
+
+
+def _now() -> datetime:
+    return datetime.now(rewards.CN_TZ).replace(tzinfo=None)
+
+
+def _config_key(scene: str) -> str:
+    if scene not in _KEY_BY_SCENE:
+        raise ValueError(f"unsupported guide video scene: {scene}")
+    return _KEY_BY_SCENE[scene]
 
 
 def _merge(raw: Any) -> dict[str, Any]:
-    """DB 里(可能不全的)dict 叠加到默认上,得到完整配置(4 个字段,无 updated_at)。"""
     out = dict(_DEFAULTS)
     if isinstance(raw, dict):
-        for k in _FIELDS:
-            v = raw.get(k)
-            if v is not None:
-                out[k] = v
+        for key in _FIELDS:
+            if key in raw and raw[key] is not None:
+                out[key] = raw[key]
     return out
 
 
-def get_config(db: Session) -> dict[str, Any]:
-    """完整配置 + updated_at(admin 读 / 业务读共用)。"""
-    row = db.get(AppConfig, _KEY)
-    cfg = _merge(row.value if row is not None else None)
+def _public_config(cfg: dict[str, Any]) -> dict[str, Any]:
+    duration = int(cfg.get("duration_ms") or 0)
+    reward = int(cfg.get("reward_coin") or 0)
+    return {
+        **cfg,
+        "duration_ms": duration or None,
+        "circle_count": CIRCLE_COUNT,
+        "circle_duration_ms": duration / CIRCLE_COUNT if duration else None,
+        "reward_per_circle": reward // CIRCLE_COUNT,
+    }
+
+
+def get_config(db: Session, scene: str = "coupon") -> dict[str, Any]:
+    row = db.get(AppConfig, _config_key(scene))
+    cfg = _public_config(_merge(row.value if row is not None else None))
+    if scene == "coupon":
+        cfg["max_plays"] = limit_policy.resolve(
+            db, "guide.video.lifetime"
+        ).global_limit
     cfg["updated_at"] = row.updated_at.isoformat() if row is not None and row.updated_at else None
     return cfg
 
 
-def _write(db: Session, value: dict[str, Any], *, admin_id: int, commit: bool) -> dict[str, Any]:
-    """整体覆写该行(value 须为完整字段 dict),返回合并后的完整配置(含 updated_at)。"""
-    row = db.get(AppConfig, _KEY)
+def _validate_settings(max_plays: int, reward_coin: int) -> None:
+    if not MIN_PLAYS <= max_plays <= MAX_PLAYS_LIMIT:
+        raise ValueError(f"播放次数必须在 {MIN_PLAYS}～{MAX_PLAYS_LIMIT} 之间")
+    if not MIN_REWARD_COIN <= reward_coin <= REWARD_COIN_LIMIT:
+        raise ValueError(f"金币总价必须在 {MIN_REWARD_COIN}～{REWARD_COIN_LIMIT} 之间")
+    if reward_coin % CIRCLE_COUNT:
+        raise ValueError("金币总价必须是 10 的倍数")
+
+
+def _write(
+    db: Session, value: dict[str, Any], *, scene: str, admin_id: int, commit: bool
+) -> dict[str, Any]:
+    key = _config_key(scene)
+    row = db.get(AppConfig, key)
     if row is None:
-        row = AppConfig(key=_KEY, value=value, updated_by_admin_id=admin_id)
+        row = AppConfig(key=key, value=value, updated_by_admin_id=admin_id)
         db.add(row)
     else:
-        row.value = value  # 整体重新赋值,SQLAlchemy 才侦测得到变更
+        row.value = value
         row.updated_by_admin_id = admin_id
     if commit:
         db.commit()
         db.refresh(row)
     else:
         db.flush()
-    out = _merge(row.value)
-    out["updated_at"] = row.updated_at.isoformat() if row.updated_at else None
-    return out
+    return get_config(db, scene)
 
 
 def update_config(
@@ -98,193 +139,431 @@ def update_config(
     enabled: bool | None = None,
     max_plays: int | None = None,
     reward_coin: int | None = None,
+    scene: str = "coupon",
     admin_id: int,
     commit: bool = True,
 ) -> tuple[dict[str, Any], dict[str, Any]]:
-    """改开关 / 次数 / 金币(只改传了的字段;视频走 set_video)。返回 (before, after) 供审计。"""
-    row = db.get(AppConfig, _KEY)
-    before = _merge(row.value if row is not None else None)
-    new_value = {k: before[k] for k in _FIELDS}
+    before = get_config(db, scene)
+    raw = {key: before[key] for key in _FIELDS}
+    candidate_plays = int(max_plays if max_plays is not None else raw["max_plays"])
+    candidate_reward = int(reward_coin if reward_coin is not None else raw["reward_coin"])
+    _validate_settings(candidate_plays, candidate_reward)
+    if enabled is True and (
+        not raw.get("video_url")
+        or raw.get("analysis_status") != "valid"
+        or not raw.get("duration_ms")
+    ):
+        raise ValueError("请先上传并通过分析的视频，再启用")
     if enabled is not None:
-        new_value["enabled"] = enabled
-    if max_plays is not None:
-        new_value["max_plays"] = max(0, min(int(max_plays), MAX_PLAYS_LIMIT))
-    if reward_coin is not None:
-        new_value["reward_coin"] = max(0, min(int(reward_coin), REWARD_COIN_LIMIT))
-    after = _write(db, new_value, admin_id=admin_id, commit=commit)
+        raw["enabled"] = enabled
+    raw["max_plays"] = candidate_plays
+    raw["reward_coin"] = candidate_reward
+    raw["config_version"] = int(raw.get("config_version") or 0) + 1
+    after = _write(
+        db,
+        raw,
+        scene=scene,
+        admin_id=admin_id,
+        commit=False,
+    )
+    if scene == "coupon" and max_plays is not None:
+        # 兼容仍调用旧专用接口的客户端/脚本，并把统一策略全局值一并更新。
+        limit_policy.set_global_limits(
+            db,
+            {"guide.video.lifetime": candidate_plays},
+            admin_id=admin_id,
+            commit=False,
+        )
+        # set_global_limits 已 flush；即使由 admin 路由统一在外层提交，也要把
+        # 同一事务内的最新统一策略值写进审计 after 快照。
+        after = get_config(db, scene)
+    if commit:
+        db.commit()
+        after = get_config(db, scene)
     return before, after
 
 
 def set_video(
-    db: Session, video_url: str | None, *, admin_id: int, commit: bool = True
+    db: Session,
+    video_url: str | None,
+    *,
+    analysis: dict[str, Any] | None = None,
+    scene: str = "coupon",
+    admin_id: int,
+    commit: bool = True,
 ) -> tuple[dict[str, Any], dict[str, Any]]:
-    """设置/清空引导视频地址。返回 (before, after);before['video_url'] 供调用方删旧文件。"""
-    row = db.get(AppConfig, _KEY)
-    before = _merge(row.value if row is not None else None)
-    new_value = {k: before[k] for k in _FIELDS}
-    new_value["video_url"] = video_url
-    after = _write(db, new_value, admin_id=admin_id, commit=commit)
+    before = get_config(db, scene)
+    raw = {key: before[key] for key in _FIELDS}
+    raw["video_url"] = video_url
+    if video_url:
+        if not analysis or analysis.get("analysis_status") != "valid":
+            raise ValueError("视频必须先完成服务端分析")
+        for key in (
+            "duration_ms", "video_codec", "audio_codec",
+            "analysis_status", "analysis_error",
+        ):
+            raw[key] = analysis.get(key)
+    else:
+        raw.update(
+            duration_ms=None,
+            video_codec=None,
+            audio_codec=None,
+            analysis_status="missing",
+            analysis_error=None,
+        )
+        raw["enabled"] = False
+    raw["config_version"] = int(raw.get("config_version") or 0) + 1
+    after = _write(db, raw, scene=scene, admin_id=admin_id, commit=commit)
     return before, after
 
 
-# ===== 播放计次 =====
-
-
-def used_plays(db: Session, user_id: int) -> int:
-    """该账号已用掉的引导视频次数(开播即算,含未发币的)。"""
-    return int(
-        db.execute(
-            select(func.count()).select_from(GuideVideoPlay).where(
-                GuideVideoPlay.user_id == user_id
-            )
-        ).scalar_one()
+def used_plays(
+    db: Session,
+    user_id: int,
+    scene: str = "coupon",
+    *,
+    reset_at: datetime | None = None,
+) -> int:
+    stmt = select(func.count()).select_from(GuideVideoPlay).where(
+        GuideVideoPlay.user_id == user_id,
+        GuideVideoPlay.scene == scene,
+        GuideVideoPlay.status != "prepared",
     )
+    if reset_at is not None:
+        # guide_video_play 使用北京时间 naive 墙钟；白名单重置点是带时区时间。
+        reset_value = reset_at.astimezone(rewards.CN_TZ).replace(tzinfo=None)
+        stmt = stmt.where(GuideVideoPlay.started_at >= reset_value)
+    return int(db.execute(stmt).scalar_one())
 
 
-def play_stats(db: Session) -> dict[str, int]:
-    """全站播放统计(admin 页展示):总播放次数 / 其中已发币次数。"""
-    total = int(
-        db.execute(select(func.count()).select_from(GuideVideoPlay)).scalar_one()
+def _effective_quota(
+    db: Session,
+    user_id: int,
+    scene: str,
+    cfg: dict[str, Any],
+) -> tuple[int | None, datetime | None]:
+    if scene != "coupon":
+        return int(cfg.get("max_plays") or 0), None
+    policy = limit_policy.resolve_for_user(
+        db, "guide.video.lifetime", user_id
     )
-    granted = int(
-        db.execute(
-            select(func.count()).select_from(GuideVideoPlay).where(
-                GuideVideoPlay.status == "granted"
-            )
-        ).scalar_one()
-    )
-    return {"total_plays": total, "granted_plays": granted}
+    return policy.limit, policy.reset_at
 
 
-def start_play(
-    db: Session, user_id: int, *, scene: str = "coupon", commit: bool = True
+def _remaining(maximum: int | None, used: int) -> int:
+    # API 字段保持整数兼容；不限时使用足够大的展示值，不参与服务端判定。
+    exposed_maximum = maximum if maximum is not None else 1_000_000
+    return max(0, exposed_maximum - used)
+
+
+def _prepare_miss(
+    scene: str,
+    reason: str,
+    cfg: dict[str, Any],
+    used: int,
+    maximum: int | None,
 ) -> dict[str, Any]:
-    """决定这次浮层是否放引导视频;命中则**当场计次**并返回 play_token。
+    return {
+        "should_play": False,
+        "reason": reason,
+        "scene": scene,
+        "video_url": None,
+        "play_token": "",
+        "config_version": int(cfg.get("config_version") or 0),
+        "duration_ms": 0,
+        "circle_count": CIRCLE_COUNT,
+        "circle_duration_ms": 0,
+        "reward_coin": int(cfg.get("reward_coin") or 0),
+        "reward_per_circle": int(cfg.get("reward_coin") or 0) // CIRCLE_COUNT,
+        "seq": used,
+        "remaining": _remaining(maximum, used),
+        "expires_at": None,
+    }
 
-    返回 dict:
-      should_play  是否放引导视频(False → 客户端照旧放广告)
-      video_url    相对地址(/media/...);客户端自行拼 BASE_URL
-      play_token   发币幂等键(should_play=False 时为空串)
-      reward_coin  播完/中途关闭都发的固定金币
-      seq / remaining  第几次 / 发完这次还剩几次(仅展示与排查用)
-    """
-    cfg = get_config(db)
-    video_url = (cfg.get("video_url") or "").strip()
-    max_plays = int(cfg.get("max_plays") or 0)
-    reward_coin = int(cfg.get("reward_coin") or 0)
-    used = used_plays(db, user_id)
 
-    def _miss(used_now: int) -> dict[str, Any]:
-        return {
-            "should_play": False,
-            "video_url": None,
-            "play_token": "",
-            "reward_coin": reward_coin,
-            "seq": used_now,
-            "remaining": max(0, max_plays - used_now),
-        }
+def prepare_play(db: Session, user_id: int, *, scene: str = "coupon") -> dict[str, Any]:
+    cfg = get_config(db, scene)
+    maximum, reset_at = _effective_quota(db, user_id, scene, cfg)
+    used = used_plays(db, user_id, scene, reset_at=reset_at)
+    video_url = str(cfg.get("video_url") or "").strip()
+    duration = int(cfg.get("duration_ms") or 0)
+    if not cfg.get("enabled"):
+        return _prepare_miss(scene, "disabled", cfg, used, maximum)
+    if not video_url or cfg.get("analysis_status") != "valid" or duration <= 0:
+        return _prepare_miss(
+            scene, "video_unavailable", cfg, used, maximum
+        )
+    if maximum is not None and used >= maximum:
+        return _prepare_miss(
+            scene, "play_limit_reached", cfg, used, maximum
+        )
 
-    if not cfg.get("enabled") or not video_url or max_plays <= 0 or used >= max_plays:
-        return _miss(used)
-
-    seq = used + 1
+    now = _now()
     play = GuideVideoPlay(
         user_id=user_id,
         play_token=uuid.uuid4().hex,
         scene=scene,
-        seq=seq,
+        seq=None,
         video_url=video_url,
-        coin=0,
-        status="playing",
+        coin=int(cfg["reward_coin"]),
+        duration_ms=duration,
+        config_version=int(cfg["config_version"]),
+        settled_circles=0,
+        status="prepared",
         completed=0,
-        started_at=datetime.now(rewards.CN_TZ).replace(tzinfo=None),
+        prepared_at=now,
+        expires_at=now + PLAN_TTL,
+        started_at=None,
     )
     db.add(play)
-    # 上面的 COUNT 判定是无锁 check-then-insert:并发 /start 会都算出同一个 seq。
-    # (user_id, seq) 唯一键让只有一个能落库,其余撞键 → 回滚后按"这次不放视频"降级,
-    # 客户端照旧走广告链路。没有它,并发就能绕过 max_plays 无限刷金币。
-    try:
-        if commit:
-            db.commit()
-        else:
-            db.flush()
-    except IntegrityError:
-        db.rollback()
-        return _miss(used_plays(db, user_id))
+    db.commit()
     return {
         "should_play": True,
+        "reason": "ready",
+        "scene": scene,
         "video_url": video_url,
         "play_token": play.play_token,
-        "reward_coin": reward_coin,
-        "seq": seq,
-        "remaining": max(0, max_plays - seq),
+        "config_version": play.config_version,
+        "duration_ms": duration,
+        "circle_count": CIRCLE_COUNT,
+        "circle_duration_ms": duration / CIRCLE_COUNT,
+        "reward_coin": play.coin,
+        "reward_per_circle": play.coin // CIRCLE_COUNT,
+        "seq": used + 1,
+        "remaining": _remaining(maximum, used + 1),
+        "expires_at": play.expires_at.isoformat(),
     }
 
 
 def _find_play(db: Session, user_id: int, token: str) -> GuideVideoPlay | None:
-    """按 (play_token, user_id) 取播放行 —— 带 user_id 是防拿别人的 token 来兑。"""
     return db.execute(
         select(GuideVideoPlay).where(
-            GuideVideoPlay.play_token == token,
+            GuideVideoPlay.play_token == token.strip(),
             GuideVideoPlay.user_id == user_id,
         )
     ).scalar_one_or_none()
 
 
-def grant_play(
-    db: Session, user_id: int, *, play_token: str, completed: bool
+def _start_out(
+    play: GuideVideoPlay,
+    maximum: int | None,
+    used: int,
+    status: str,
 ) -> dict[str, Any]:
-    """按 play_token 发这次引导视频的金币(幂等)。播完 / 中途关闭都发。
+    assert play.started_at is not None and play.seq is not None and play.video_url
+    return {
+        "started": True,
+        "status": status,
+        "play_token": play.play_token,
+        "scene": play.scene,
+        "video_url": play.video_url,
+        "config_version": play.config_version,
+        "duration_ms": play.duration_ms,
+        "circle_count": CIRCLE_COUNT,
+        "circle_duration_ms": play.duration_ms / CIRCLE_COUNT,
+        "reward_coin": play.coin,
+        "reward_per_circle": play.coin // CIRCLE_COUNT,
+        "seq": play.seq,
+        "remaining": _remaining(maximum, used),
+        "started_at": play.started_at.isoformat(),
+    }
 
-    返回 {granted, coin, status}:granted=True 表示**本次调用真的入账了**;
-    重复上报返回 granted=False + 已发金币(客户端据此不重复累加 toast 金额)。
-    """
-    token = (play_token or "").strip()
-    # 金币额度以**服务端配置**为准,不信客户端(客户端只上报"播完/关闭")。
-    coin = int(get_config(db).get("reward_coin") or 0)
 
-    # 幂等核心:把 status 放进 WHERE 做条件更新(compare-and-set),而不是"先读再判再写"。
-    # 「播完」与「✕ 关闭」抢跑、或客户端超时重试时,两个请求会都读到 status='playing',
-    # 无锁的话就都往下发币、都 commit,金币入账两次(不用恶意,重试就会中招)。改成条件更新后
-    # 并发里只有一条 rowcount=1,另一条拿 0 → 按已发返回,不二次铸币。
-    # (PG READ COMMITTED 下后到的 UPDATE 阻塞到对手提交,再按新版本重判 status;SQLite 写串行。)
-    #
-    # 别指望 IntegrityError 兜底:这里只 UPDATE 不 INSERT,撞不到 uq_guide_video_play_token;
-    # 而 biz_type='guide_video' 的金币流水也不在 ux_coin_transaction_task_ref 的谓词
-    # (biz_type LIKE 'task%')覆盖范围内 —— 两个唯一键在这条路径上都是不生效的。
+def start_play(db: Session, user_id: int, *, play_token: str) -> dict[str, Any]:
+    play = _find_play(db, user_id, play_token)
+    if play is None:
+        raise PlayStateError("play_not_found", "播放计划不存在")
+    cfg = get_config(db, play.scene)
+    maximum, reset_at = _effective_quota(
+        db, user_id, play.scene, cfg
+    )
+    used = used_plays(
+        db, user_id, play.scene, reset_at=reset_at
+    )
+    if play.status in {"started", "completed"}:
+        return _start_out(play, maximum, used, "already_started")
+    if play.status != "prepared":
+        raise PlayStateError("play_not_found", "播放计划不可用")
+    now = _now()
+    if play.expires_at is None or now > play.expires_at:
+        raise PlayStateError(
+            "plan_expired", "播放计划已过期，请重新获取",
+            reprepare_required=True,
+        )
+    if int(cfg["config_version"]) != play.config_version:
+        raise PlayStateError(
+            "config_changed", "视频配置已变化，请重新获取",
+            reprepare_required=True,
+        )
+    if maximum is not None and used >= maximum:
+        raise PlayStateError("play_limit_reached", "播放次数已用完")
+    # seq 在数据库中是账号+场景生命周期唯一值；白名单重置只重置额度，
+    # 不能从 1 重新编号，否则会与历史记录冲突。
+    play.seq = used_plays(db, user_id, play.scene) + 1
+    play.status = "started"
+    play.started_at = now
+    try:
+        db.commit()
+        db.refresh(play)
+    except IntegrityError as exc:
+        db.rollback()
+        raise PlayStateError("play_limit_reached", "并发起播冲突，请重新获取") from exc
+    return _start_out(play, maximum, used + 1, "started")
+
+
+def _coin_balance(db: Session, user_id: int) -> int:
+    account = db.get(CoinAccount, user_id)
+    return int(account.coin_balance if account else 0)
+
+
+def grant_circle(
+    db: Session, user_id: int, *, play_token: str, circle: int
+) -> dict[str, Any]:
+    play = _find_play(db, user_id, play_token)
+    balance = _coin_balance(db, user_id)
+    base = {
+        "granted": False,
+        "retryable": False,
+        "retry_after_ms": 0,
+        "circle": circle,
+        "granted_coin": 0,
+        "settled_circles": int(play.settled_circles if play else 0),
+        "coin_balance": balance,
+    }
+    if play is None:
+        return {**base, "status": "not_found"}
+    if play.status not in {"started", "completed"} or play.started_at is None:
+        return {**base, "status": "not_started"}
+    settled = int(play.settled_circles)
+    if circle <= settled:
+        return {
+            **base,
+            "status": "already_granted",
+            "granted_coin": play.coin // CIRCLE_COUNT,
+            "settled_circles": settled,
+        }
+    if settled >= CIRCLE_COUNT:
+        return {**base, "status": "finished", "settled_circles": settled}
+    if circle != settled + 1:
+        return {
+            **base,
+            "status": "out_of_order",
+            "retryable": True,
+            "settled_circles": settled,
+        }
+    # 向上取整到毫秒，绝不让第 N 圈早于精确 N/10 边界发奖。
+    boundary_ms = (circle * play.duration_ms + CIRCLE_COUNT - 1) // CIRCLE_COUNT
+    elapsed_ms = max(0, int((_now() - play.started_at).total_seconds() * 1000))
+    if elapsed_ms < boundary_ms:
+        return {
+            **base,
+            "status": "too_early",
+            "retryable": True,
+            "retry_after_ms": boundary_ms - elapsed_ms,
+            "settled_circles": settled,
+        }
+
+    final = circle == CIRCLE_COUNT
     won = db.execute(
         update(GuideVideoPlay)
         .where(
-            GuideVideoPlay.play_token == token,
-            GuideVideoPlay.user_id == user_id,
-            GuideVideoPlay.status == "playing",
+            GuideVideoPlay.id == play.id,
+            GuideVideoPlay.status == "started",
+            GuideVideoPlay.settled_circles == settled,
         )
         .values(
-            status="granted",
-            coin=coin,
-            completed=1 if completed else 0,
-            granted_at=datetime.now(rewards.CN_TZ).replace(tzinfo=None),
+            settled_circles=circle,
+            status="completed" if final else "started",
+            completed=1 if final else 0,
+            granted_at=_now() if final else None,
         )
         .execution_options(synchronize_session=False)
     ).rowcount
-
     if not won:
-        # 没抢到:token 不存在 / 不是本人的 / 已被另一次上报发过。回滚拿干净快照再区分两者
-        # (对手此时必然已提交 —— 我们就是被它挡下的,所以读得到它写的 coin)。
         db.rollback()
-        play = _find_play(db, user_id, token)
-        if play is None:
-            return {"granted": False, "coin": 0, "status": "not_found"}
-        return {"granted": False, "coin": play.coin, "status": "already_granted"}
+        fresh = _find_play(db, user_id, play_token)
+        fresh_settled = int(fresh.settled_circles if fresh else 0)
+        return {
+            **base,
+            "status": "already_granted" if circle <= fresh_settled else "out_of_order",
+            "retryable": circle > fresh_settled,
+            "granted_coin": (
+                play.coin // CIRCLE_COUNT if circle <= fresh_settled else 0
+            ),
+            "settled_circles": fresh_settled,
+            "coin_balance": _coin_balance(db, user_id),
+        }
 
-    if coin > 0:
-        crud_wallet.grant_coins(
-            db,
-            user_id,
-            coin,
-            biz_type=BIZ_TYPE,
-            ref_id=token,
-            remark="新手引导视频奖励",
-        )
-    db.commit()
-    return {"granted": True, "coin": coin, "status": "granted"}
+    per_circle = play.coin // CIRCLE_COUNT
+    account, _ = crud_wallet.grant_coins(
+        db,
+        user_id,
+        per_circle,
+        biz_type=BIZ_TYPE,
+        ref_id=f"{play.play_token}:{circle}",
+        remark=f"新手引导视频第{circle}圈奖励",
+    )
+    try:
+        db.commit()
+    except IntegrityError:
+        # 流水唯一键是第二道幂等保险；若撞键，回滚后返回权威状态。
+        db.rollback()
+        fresh = _find_play(db, user_id, play_token)
+        return {
+            **base,
+            "status": "already_granted",
+            "granted_coin": play.coin // CIRCLE_COUNT,
+            "settled_circles": int(fresh.settled_circles if fresh else settled),
+            "coin_balance": _coin_balance(db, user_id),
+        }
+    return {
+        **base,
+        "granted": True,
+        "status": "granted",
+        "granted_coin": per_circle,
+        "settled_circles": circle,
+        "coin_balance": int(account.coin_balance),
+    }
+
+
+def cleanup_old_videos(db: Session, *, protection: timedelta = PLAN_TTL) -> int:
+    """清理超过保护期且未被当前配置/有效计划/近期起播引用的托管旧片。"""
+    now = _now()
+    protected_urls: set[str] = set()
+    for scene in SCENES:
+        url = get_config(db, scene).get("video_url")
+        if url:
+            protected_urls.add(str(url))
+    protected_urls.update(
+        str(url)
+        for url in db.execute(
+            select(GuideVideoPlay.video_url).where(
+                GuideVideoPlay.video_url.is_not(None),
+                (
+                    (
+                        (GuideVideoPlay.status == "prepared")
+                        & (GuideVideoPlay.expires_at >= now)
+                    )
+                    | (
+                        (GuideVideoPlay.status == "started")
+                        & (GuideVideoPlay.started_at >= now - protection)
+                    )
+                ),
+            )
+        ).scalars()
+        if url
+    )
+    directory = Path(settings.MEDIA_ROOT) / "guide_video"
+    if not directory.is_dir():
+        return 0
+    cutoff = now.timestamp() - protection.total_seconds()
+    removed = 0
+    for path in directory.glob("guide_*.mp4"):
+        url = f"{settings.MEDIA_URL_PREFIX}/guide_video/{path.name}"
+        try:
+            if url not in protected_urls and path.stat().st_mtime <= cutoff:
+                media.delete_guide_video(url)
+                removed += 1
+        except OSError:
+            continue
+    return removed

app/repositories/phone_rebind.py

@@ -2,7 +2,7 @@
 from __future__ import annotations
 
 import math
-from datetime import datetime, timedelta, timezone
+from datetime import UTC, datetime, timedelta
 
 from sqlalchemy import func, select
 from sqlalchemy.orm import Session
@@ -10,9 +10,22 @@ from sqlalchemy.orm import Session
 from app.models.phone_rebind_log import PhoneRebindLog
 
 
-def rebound_within_days(db: Session, phone: str, days: int) -> bool:
+def rebound_within_days(
+    db: Session,
+    phone: str,
+    days: int,
+    *,
+    reset_at: datetime | None = None,
+) -> bool:
     """该手机号在最近 days 天内是否换绑过(命中 → 禁止再次换绑)。"""
-    since = datetime.now(timezone.utc) - timedelta(days=days)
+    since = datetime.now(UTC) - timedelta(days=days)
+    if reset_at is not None:
+        reset_value = (
+            reset_at.replace(tzinfo=UTC)
+            if reset_at.tzinfo is None
+            else reset_at.astimezone(UTC)
+        )
+        since = max(since, reset_value)
     stmt = (
         select(PhoneRebindLog.id)
         .where(PhoneRebindLog.phone == phone, PhoneRebindLog.rebound_at >= since)
@@ -21,16 +34,25 @@ def rebound_within_days(db: Session, phone: str, days: int) -> bool:
     return db.execute(stmt).first() is not None
 
 
-def remaining_block_days(db: Session, phone: str, days: int) -> int:
+def remaining_block_days(
+    db: Session,
+    phone: str,
+    days: int,
+    *,
+    reset_at: datetime | None = None,
+) -> int:
     """距离该手机号可再次换绑还剩几天(向上取整;无记录返回 0)。"""
+    conditions = [PhoneRebindLog.phone == phone]
+    if reset_at is not None:
+        conditions.append(PhoneRebindLog.rebound_at >= reset_at)
     last = db.execute(
-        select(func.max(PhoneRebindLog.rebound_at)).where(PhoneRebindLog.phone == phone)
+        select(func.max(PhoneRebindLog.rebound_at)).where(*conditions)
     ).scalar_one_or_none()
     if last is None:
         return 0
     if last.tzinfo is None:  # SQLite 取回 naive datetime,按 UTC 归一
-        last = last.replace(tzinfo=timezone.utc)
-    remaining = (last + timedelta(days=days) - datetime.now(timezone.utc)).total_seconds()
+        last = last.replace(tzinfo=UTC)
+    remaining = (last + timedelta(days=days) - datetime.now(UTC)).total_seconds()
     return max(0, math.ceil(remaining / 86400))
 
 

app/repositories/risk.py

@@ -9,15 +9,10 @@ from sqlalchemy import func, select
 from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
-from app.core.config_schema import (
-    RISK_COMPARE_DAILY_THRESHOLD_KEY,
-    RISK_ONECLICK_DAILY_THRESHOLD_KEY,
-    RISK_SMS_HOURLY_THRESHOLD_KEY,
-)
+from app.core import limit_policy
 from app.models.app_config import AppConfig
 from app.models.comparison import ComparisonRecord
 from app.models.risk import BehaviorEvent, RiskIncident, SubjectRestriction
-from app.repositories import app_config
 
 CN_TZ = ZoneInfo("Asia/Shanghai")
 
@@ -41,7 +36,6 @@ class RuleSpec:
     code: str
     event_type: str
     subject_type: str
-    threshold_key: str
     window: str
     count_outcomes: tuple[str, ...]
 
@@ -51,7 +45,6 @@ RULES: dict[str, RuleSpec] = {
         code=RULE_SMS_HOURLY,
         event_type=EVENT_SMS_SEND,
         subject_type="device",
-        threshold_key=RISK_SMS_HOURLY_THRESHOLD_KEY,
         window="hour",
         count_outcomes=("success",),
     ),
@@ -59,22 +52,23 @@ RULES: dict[str, RuleSpec] = {
         code=RULE_ONECLICK_DAILY,
         event_type=EVENT_ONECLICK_LOGIN,
         subject_type="device",
-        threshold_key=RISK_ONECLICK_DAILY_THRESHOLD_KEY,
         window="day",
         count_outcomes=("success", "failed"),
     ),
 }
 
-RULE_THRESHOLD_KEYS: dict[str, str] = {
-    RULE_SMS_HOURLY: RISK_SMS_HOURLY_THRESHOLD_KEY,
-    RULE_ONECLICK_DAILY: RISK_ONECLICK_DAILY_THRESHOLD_KEY,
-    RULE_COMPARE_DAILY: RISK_COMPARE_DAILY_THRESHOLD_KEY,
+RISK_LIMIT_RULE_CODES: dict[str, str] = {
+    RULE_SMS_HOURLY: "risk.sms.hourly",
+    RULE_ONECLICK_DAILY: "risk.oneclick.daily",
+    RULE_COMPARE_DAILY: "risk.compare.daily",
 }
 
 
 def get_rule_threshold(db: Session, rule_code: str) -> int:
     """读取规则当前阈值；配置表为空时回退上线前的 5/20/100 默认值。"""
-    return int(app_config.get_value(db, RULE_THRESHOLD_KEYS[rule_code]))
+    return limit_policy.resolve(
+        db, RISK_LIMIT_RULE_CODES[rule_code]
+    ).global_limit
 
 
 def utcnow() -> datetime:
@@ -257,22 +251,34 @@ def _upsert_incident(
 
 
 def evaluate_behavior_rule(
-    db: Session, *, rule_code: str, subject_id: str, at: datetime
+    db: Session,
+    *,
+    rule_code: str,
+    subject_id: str,
+    at: datetime,
+    threshold: int | None = None,
+    subject_reset_at: datetime | None = None,
 ) -> RiskIncident | None:
     spec = RULES[rule_code]
-    threshold = get_rule_threshold(db, rule_code)
+    effective_threshold = threshold or get_rule_threshold(db, rule_code)
     window_key, window_start, end = _window_bounds(at, spec.window)
     reset_at = get_rule_reset_at(db, rule_code)
-    start = max(window_start, reset_at) if reset_at else window_start
+    baselines = [value for value in (reset_at, subject_reset_at) if value is not None]
+    start = max(window_start, *baselines) if baselines else window_start
     count, first_at, last_at, triggered_at = _event_stats(
         db,
         spec,
         subject_id=subject_id,
         start=start,
         end=end,
-        threshold=threshold,
+        threshold=effective_threshold,
     )
-    if count < threshold or first_at is None or last_at is None or triggered_at is None:
+    if (
+        count < effective_threshold
+        or first_at is None
+        or last_at is None
+        or triggered_at is None
+    ):
         return None
     return _upsert_incident(
         db,
@@ -300,7 +306,6 @@ def reconcile_behavior_rule(
     """按当前阈值重算短信/一键登录当前窗口，并收起已不再命中的待处理告警。"""
     spec = RULES[rule_code]
     current = at or utcnow()
-    threshold = get_rule_threshold(db, rule_code)
     window_key, window_start, end = _window_bounds(current, spec.window)
     reset_at = get_rule_reset_at(db, rule_code)
     start = max(window_start, reset_at) if reset_at else window_start
@@ -311,23 +316,39 @@ def reconcile_behavior_rule(
         BehaviorEvent.occurred_at >= start,
         BehaviorEvent.occurred_at < end,
     )
-    qualifying_rows = db.execute(
+    subject_rows = db.execute(
         select(
             BehaviorEvent.subject_id,
             func.max(BehaviorEvent.occurred_at),
+            func.max(BehaviorEvent.phone),
         )
         .where(*filters)
         .group_by(BehaviorEvent.subject_id)
-        .having(func.count(BehaviorEvent.id) >= threshold)
     ).all()
-    qualifying = {str(subject_id) for subject_id, _ in qualifying_rows}
-    for subject_id, last_at in qualifying_rows:
-        evaluate_behavior_rule(
+    qualifying: set[str] = set()
+    policy_code = {
+        RULE_SMS_HOURLY: "risk.sms.hourly",
+        RULE_ONECLICK_DAILY: "risk.oneclick.daily",
+    }[rule_code]
+    for subject_id, last_at, phone in subject_rows:
+        policy = limit_policy.resolve(
+            db,
+            policy_code,
+            phone=phone,
+            device=str(subject_id),
+        )
+        if policy.suppressed or policy.limit is None:
+            continue
+        incident = evaluate_behavior_rule(
             db,
             rule_code=rule_code,
             subject_id=str(subject_id),
             at=last_at or current,
+            threshold=policy.limit,
+            subject_reset_at=policy.reset_at,
         )
+        if incident is not None:
+            qualifying.add(str(subject_id))
 
     open_incidents = db.scalars(
         select(RiskIncident).where(
@@ -383,7 +404,29 @@ def record_behavior_event(
     db.add(event)
     db.flush()
     if evaluate_rule:
-        evaluate_behavior_rule(db, rule_code=evaluate_rule, subject_id=subject_id, at=at)
+        policy_code = {
+            RULE_SMS_HOURLY: "risk.sms.hourly",
+            RULE_ONECLICK_DAILY: "risk.oneclick.daily",
+        }.get(evaluate_rule)
+        policy = (
+            limit_policy.resolve(
+                db,
+                policy_code,
+                phone=phone,
+                device=device_id or subject_id,
+            )
+            if policy_code
+            else None
+        )
+        if policy is None or not policy.suppressed:
+            evaluate_behavior_rule(
+                db,
+                rule_code=evaluate_rule,
+                subject_id=subject_id,
+                at=at,
+                threshold=policy.limit if policy else None,
+                subject_reset_at=policy.reset_at if policy else None,
+            )
     if commit:
         db.commit()
         db.refresh(event)
@@ -396,20 +439,33 @@ def sync_compare_incident(
     user_id: int,
     at: datetime,
     threshold: int | None = None,
+    device_id: str | None = None,
     commit: bool = True,
 ) -> RiskIncident | None:
-    effective_threshold = threshold or get_rule_threshold(db, RULE_COMPARE_DAILY)
+    policy = limit_policy.resolve_for_user(
+        db,
+        "risk.compare.daily",
+        user_id,
+        device=device_id,
+    )
+    if policy.suppressed:
+        return None
+    effective_threshold = threshold or policy.limit or get_rule_threshold(
+        db, RULE_COMPARE_DAILY
+    )
     # comparison_record 的既有写入口统一落“北京时间 naive”时间；这里必须沿用同一
     # 口径，否则 SQLite/PG session timezone 不同时会把凌晨记录算到前一天。
     local = at.astimezone(CN_TZ).replace(tzinfo=None) if at.tzinfo else at
     window_start = local.replace(hour=0, minute=0, second=0, microsecond=0)
     end = window_start + timedelta(days=1)
     window_key = window_start.strftime("%Y-%m-%d")
-    reset_at = get_rule_reset_at(db, RULE_COMPARE_DAILY)
-    reset_local = (
-        reset_at.astimezone(CN_TZ).replace(tzinfo=None) if reset_at else None
-    )
-    start = max(window_start, reset_local) if reset_local else window_start
+    global_reset_at = get_rule_reset_at(db, RULE_COMPARE_DAILY)
+    baselines = [
+        value.astimezone(CN_TZ).replace(tzinfo=None)
+        for value in (global_reset_at, policy.reset_at)
+        if value is not None
+    ]
+    start = max(window_start, *baselines) if baselines else window_start
     filters = (
         ComparisonRecord.user_id == user_id,
         ComparisonRecord.created_at >= start,
@@ -464,7 +520,6 @@ def reconcile_compare_rule(
         reset_at.astimezone(CN_TZ).replace(tzinfo=None) if reset_at else None
     )
     start = max(window_start, reset_local) if reset_local else window_start
-    threshold = get_rule_threshold(db, RULE_COMPARE_DAILY)
     rows = db.execute(
         select(
             ComparisonRecord.user_id,
@@ -476,17 +531,17 @@ def reconcile_compare_rule(
             ComparisonRecord.created_at < end,
         )
         .group_by(ComparisonRecord.user_id)
-        .having(func.count(ComparisonRecord.id) >= threshold)
     ).all()
-    qualifying = {str(user_id) for user_id, _ in rows}
+    qualifying: set[str] = set()
     for user_id, last_at in rows:
-        sync_compare_incident(
+        incident = sync_compare_incident(
             db,
             user_id=int(user_id),
             at=last_at or current,
-            threshold=threshold,
             commit=False,
         )
+        if incident is not None:
+            qualifying.add(str(user_id))
 
     open_incidents = db.scalars(
         select(RiskIncident).where(

app/repositories/wallet.py

@@ -132,10 +132,20 @@ def get_or_create_account(
 ) -> CoinAccount:
     """取用户金币账户,不存在则建一个空账户。
 
-    lock=True 时对已存在的账户行加 SELECT FOR UPDATE(读-算-写余额的调用方串行化,防并发
-    双写余额错位,如 admin set 模式连点);默认 False 不改 C 端发奖行为。SQLite 下为 no-op。
+    lock=True 时对已存在的账户行加 SELECT FOR UPDATE，并强制刷新 identity map 中可能
+    缓存的旧余额；所有金币写入口都会使用它。SQLite 下 FOR UPDATE 为 no-op。
     """
-    acc = db.get(CoinAccount, user_id, with_for_update=True) if lock else db.get(CoinAccount, user_id)
+    if lock:
+        # populate_existing 很关键：同一 Session 可能早已缓存旧余额。只加 FOR UPDATE 而
+        # 继续复用 identity map 里的旧对象，仍会在另一笔发奖提交后把余额覆盖回去。
+        acc = db.execute(
+            select(CoinAccount)
+            .where(CoinAccount.user_id == user_id)
+            .with_for_update()
+            .execution_options(populate_existing=True)
+        ).scalar_one_or_none()
+    else:
+        acc = db.get(CoinAccount, user_id)
     if acc is None:
         acc = CoinAccount(
             user_id=user_id,
@@ -165,7 +175,9 @@ def grant_coins(
 
     返回 (account, transaction)。调用方负责 commit。
     """
-    acc = get_or_create_account(db, user_id, commit=False)
+    # 所有金币来源统一遵守账户行锁协议；否则 guide 锁了账户，广告/签到等未锁路径
+    # 仍可拿旧余额做 ORM 读改写，最终余额会小于流水累计。
+    acc = get_or_create_account(db, user_id, commit=False, lock=True)
     acc.coin_balance += amount
     if amount > 0:
         acc.total_coin_earned += amount
@@ -198,7 +210,7 @@ def grant_cash(
     与 [grant_coins] 同模式(运营手动调现金 / 测试发现金用)。返回 (account, transaction),
     调用方负责 commit。不在此校验扣成负——由调用方(admin router)按业务保护。
     """
-    acc = get_or_create_account(db, user_id, commit=False)
+    acc = get_or_create_account(db, user_id, commit=False, lock=True)
     acc.cash_balance_cents += amount_cents
 
     txn = CashTransaction(
@@ -228,7 +240,7 @@ def grant_invite_cash(
     invite_cash_transaction,不 commit。与金币兑换的 cash_balance_cents **物理隔离**
     (产品红线:邀请奖励金 ≠ 金币现金,两本账不可累加)。返回 (account, transaction),
     调用方负责 commit。不在此校验扣成负——由调用方按业务保护。"""
-    acc = get_or_create_account(db, user_id, commit=False)
+    acc = get_or_create_account(db, user_id, commit=False, lock=True)
     acc.invite_cash_balance_cents += amount_cents
 
     txn = InviteCashTransaction(
@@ -286,7 +298,8 @@ def exchange_coins_to_cash(
     if coin_amount < floor_min or coin_amount % COIN_PER_CENT != 0:
         raise InvalidExchangeAmountError
 
-    acc = get_or_create_account(db, user_id, commit=False)
+    # 余额充足校验必须和扣减遵守同一账户锁，否则并发消费都可能通过旧余额校验。
+    acc = get_or_create_account(db, user_id, commit=False, lock=True)
     if acc.coin_balance < coin_amount:
         raise InsufficientCoinError
 

app/schemas/auth.py

@@ -7,9 +7,19 @@
 """
 from __future__ import annotations
 
+import re
 from datetime import datetime
 
-from pydantic import BaseModel, ConfigDict, Field
+from pydantic import BaseModel, ConfigDict, Field, field_validator
+
+from app.core.config import settings
+
+
+def _validate_login_phone(phone: str) -> str:
+    """普通号保持大陆手机号格式；仅显式配置的测试号允许例外。"""
+    if re.fullmatch(r"^1\d{10}$", phone) or phone in settings.test_account_phones:
+        return phone
+    raise ValueError("invalid phone")
 
 # ===== 用户对外信息 =====
 
@@ -72,7 +82,7 @@ class JverifyLoginRequest(BaseModel):
 # ===== 短信验证码 =====
 
 class SmsSendRequest(BaseModel):
-    phone: str = Field(..., min_length=11, max_length=11, pattern=r"^1\d{10}$")
+    phone: str = Field(..., min_length=11, max_length=11)
     device_id: str = Field(
         "", max_length=64,
         description="硬件级设备标识(Android ANDROID_ID),用于发码防刷按 设备+IP 限流;空=按 IP 聚一桶",
@@ -80,6 +90,7 @@ class SmsSendRequest(BaseModel):
     device_model: str = Field(
         "", max_length=128, description="客户端设备型号快照,用于短信安全审计"
     )
+    _valid_phone = field_validator("phone")(_validate_login_phone)
 
 
 class SmsSendResponse(BaseModel):
@@ -89,7 +100,7 @@ class SmsSendResponse(BaseModel):
 
 
 class SmsLoginRequest(BaseModel):
-    phone: str = Field(..., min_length=11, max_length=11, pattern=r"^1\d{10}$")
+    phone: str = Field(..., min_length=11, max_length=11)
     code: str = Field(..., min_length=4, max_length=8)
     device_id: str = Field(
         "", max_length=64,
@@ -98,6 +109,7 @@ class SmsLoginRequest(BaseModel):
     device_model: str = Field(
         "", max_length=128, description="客户端设备型号快照,用于短信验证安全审计"
     )
+    _valid_phone = field_validator("phone")(_validate_login_phone)
 
 
 # ===== Refresh =====

app/schemas/compare_record.py

@@ -107,6 +107,14 @@ class ComparisonRecordIn(BaseModel):
     # 明细
     items: list[ComparisonItemIn] = Field(default_factory=list)
     comparison_results: list[ComparisonResultIn] = Field(default_factory=list)
+    # 展示模型统一数组(pricebot done.params.platforms 原样透传): 每平台一行、自带
+    # status/is_best/display/display_order,记录页据此直渲染。宽松 list[dict] 存(结构由
+    # pricebot 定,server 只原样落库),前端读它、老记录空时回退 comparison_results。
+    platforms: list[dict] = Field(default_factory=list)
+    # 记录级业务结局(pricebot 下发): success/below_minimum/store_closed/failed。
+    # below_minimum 表示流程正常完成,持久化主状态归 success;原值仍随 raw_payload 落库,
+    # admin/记录页从 platform_results 展示"未满起送"的细分结论。
+    record_status: str | None = None
     # 逐平台结局摘要(含失败平台的细分原因 status: store_not_found/items_not_found/below_minimum/
     # unsupported/...)。来自 done.params.platform_results,客户端透传;落 raw_payload(不单列),
     # admin「卡在哪一步」从这里读。dict{platform_id: {...}} 宽松存(结构由 pricebot 定——是
@@ -177,6 +185,9 @@ class ComparisonRecordOut(BaseModel):
     fail_reason: str | None = None
     items: list = []
     comparison_results: list = []
+    # 展示模型统一数组(每平台一行、自带 status/is_best/display/display_order): 记录页据此
+    # 直渲染, 不再靠 comparison_results + 前端派生。老记录为空 → 前端回退 comparison_results。
+    platforms: list = []
     skipped_dish_names: list = []
     total_ms: int | None = None
     # 「已下单」(店级):该店名在该用户真实下单(source='compare')里出现过即 True。
@@ -212,9 +223,9 @@ class CompareStartReserveIn(BaseModel):
 
 
 class CompareStartReserveOut(BaseModel):
-    limit: int
+    limit: int | None
     used: int
-    remaining: int
+    remaining: int | None
 
 
 class CompareStatsOut(BaseModel):

app/schemas/device.py

@@ -12,6 +12,7 @@ class DeviceRegisterRequest(BaseModel):
     registration_id: str | None = None
     push_vendor: str | None = None
     push_token: str | None = None
+    push_binding_id: str | None = Field(default=None, max_length=128)
     platform: str = "android"
     app_version: str | None = None
 
@@ -23,6 +24,7 @@ class HeartbeatRequest(BaseModel):
     registration_id: str | None = None
     push_vendor: str | None = None
     push_token: str | None = None
+    push_binding_id: str | None = Field(default=None, max_length=128)
 
 
 class DeviceOut(BaseModel):
@@ -33,6 +35,7 @@ class DeviceOut(BaseModel):
     registration_id: str | None
     push_vendor: str | None
     push_token: str | None
+    push_binding_id: str | None
     ever_protected: bool
     liveness_state: str
     last_heartbeat_at: datetime | None
@@ -43,6 +46,14 @@ class OkResponse(BaseModel):
     ok: bool = True
 
 
+class DeviceUnregisterRequest(BaseModel):
+    device_id: str = Field(min_length=1, max_length=128)
+
+
+class DeferredDeviceUnregisterRequest(DeviceUnregisterRequest):
+    push_binding_id: str = Field(min_length=32, max_length=128)
+
+
 class LivenessOut(BaseModel):
     """本机掉线告警状态(后置检测 pull 版)。客户端只需这一个布尔判断要不要弹「开启自启动」引导,
     故只返回 kill_alert_pending(不暴露设备详情 / 内部 liveness_state 等)。从未注册过 → 默认 False(无告警)。"""

app/schemas/guide_video.py

@@ -1,36 +1,72 @@
-"""新手引导视频(领券等候浮层前 N 次替代广告)的客户端请求/响应契约。"""
-from __future__ import annotations
+"""客户端引导视频三阶段协议。"""
+from typing import Literal
 
 from pydantic import BaseModel, Field
 
+GuideScene = Literal["coupon", "comparison"]
+
+
+class GuideVideoPrepareIn(BaseModel):
+    scene: GuideScene = "coupon"
+
+
+class GuideVideoPrepareOut(BaseModel):
+    should_play: bool
+    reason: str
+    scene: GuideScene
+    video_url: str | None = None
+    play_token: str = ""
+    config_version: int = 0
+    duration_ms: int = 0
+    circle_count: int = 10
+    circle_duration_ms: float = 0
+    reward_coin: int = 0
+    reward_per_circle: int = 0
+    seq: int = 0
+    remaining: int = 0
+    expires_at: str | None = None
+
 
 class GuideVideoStartIn(BaseModel):
-    """开播询问。scene 目前只有 coupon(领券浮层);预留给日后比价等场景。"""
-
-    scene: str = Field(default="coupon", max_length=16)
+    play_token: str = Field(min_length=1, max_length=64)
 
 
 class GuideVideoStartOut(BaseModel):
-    """should_play=False 时客户端照旧走广告链路,其余字段无意义。"""
-
-    should_play: bool
-    video_url: str | None = None   # 相对地址 /media/...;客户端自行拼 BASE_URL
-    play_token: str = ""           # 发奖幂等键
-    reward_coin: int = 0           # 播完/中途关闭都发的固定金币
-    seq: int = 0                   # 本账号第几次
-    remaining: int = 0             # 发完这次还剩几次
+    started: bool
+    status: Literal["started", "already_started"]
+    play_token: str
+    scene: GuideScene
+    video_url: str
+    config_version: int
+    duration_ms: int
+    circle_count: int = 10
+    circle_duration_ms: float
+    reward_coin: int
+    reward_per_circle: int
+    seq: int
+    remaining: int
+    started_at: str
 
 
 class GuideVideoRewardIn(BaseModel):
-    """播完或中途关闭都调这个;completed 只做留痕,两者都发币。"""
-
     play_token: str = Field(min_length=1, max_length=64)
-    completed: bool = False
+    circle: int = Field(ge=1, le=10)
 
 
 class GuideVideoRewardOut(BaseModel):
-    """granted=True 表示本次调用真的入账(重复上报为 False,coin 是已发金额)。"""
-
     granted: bool
-    coin: int
-    status: str
+    status: Literal[
+        "granted",
+        "already_granted",
+        "too_early",
+        "out_of_order",
+        "not_started",
+        "not_found",
+        "finished",
+    ]
+    retryable: bool = False
+    retry_after_ms: int = 0
+    circle: int
+    granted_coin: int = 0
+    settled_circles: int = 0
+    coin_balance: int = 0

app/services/notification_events.py

@@ -83,8 +83,13 @@ def _dispatch(
     extra: dict[str, str] | None = None,
     dedup_key: str | None = None,
     push_vars: dict[str, str] | None = None,
+    push_token_contains: str | None = None,
 ) -> Notification | None:
     """落一条站内消息并向该用户设备直推。返回落库行;去重命中/失败返回 None。"""
+    logger.info(
+        "notification dispatch started user_id=%s type=%s dedup_key=%s",
+        user_id, type_key, dedup_key,
+    )
     try:
         row = notif_repo.create_notification(
             db,
@@ -111,11 +116,20 @@ def _dispatch(
             logger.exception("rollback after notification failure also failed")
         return None
 
-    _push_to_user_devices(db, row, push_vars)
+    logger.info(
+        "notification created user_id=%s type=%s notification_id=%s dedup_key=%s",
+        user_id, type_key, row.id, dedup_key,
+    )
+    _push_to_user_devices(db, row, push_vars, push_token_contains)
     return row
 
 
-def _push_to_user_devices(db: Session, row: Notification, push_vars: dict[str, str] | None) -> None:
+def _push_to_user_devices(
+    db: Session,
+    row: Notification,
+    push_vars: dict[str, str] | None,
+    push_token_contains: str | None = None,
+) -> None:
     """向消息归属用户的全部厂商推送目标直推(best-effort,单设备失败不影响其余)。"""
     try:
         title, body = catalog.render_push(row.type, push_vars)
@@ -124,29 +138,103 @@ def _push_to_user_devices(db: Session, row: Notification, push_vars: dict[str, s
         extras.update({str(k): str(v) for k, v in (row.extra or {}).items()})
         extras["notificationId"] = str(row.id)
 
-        for dev in device_repo.list_push_targets(db, user_id=row.user_id):
+        targets = device_repo.list_push_targets(db, user_id=row.user_id)
+        if push_token_contains:
+            normalized_filter = push_token_contains.casefold()
+            targets = [
+                target
+                for target in targets
+                if normalized_filter in (target.push_token or "").casefold()
+            ]
+        logger.info(
+            "push targets resolved user_id=%s type=%s notification_id=%s "
+            "token_filter=%s target_count=%s",
+            row.user_id, row.type, row.id, bool(push_token_contains), len(targets),
+        )
+        if not targets:
+            logger.warning(
+                "push skipped no targets user_id=%s type=%s notification_id=%s",
+                row.user_id, row.type, row.id,
+            )
+            return
+
+        sent = failed = skipped = data_sent = data_failed = 0
+        for dev in targets:
             vendor = vendor_push.normalize_vendor(dev.push_vendor)
             if not vendor or vendor not in vendor_push.SUPPORTED_VENDORS:
-                continue
-            if vendor_push.missing_settings(vendor):
-                # 本地/测试环境凭据不齐 → 只落站内消息,不发真推送(与 push/vendors 的报缺口径一致)
-                logger.info(
-                    "skip push (vendor %s not configured) user_id=%s type=%s",
-                    vendor, row.user_id, row.type,
+                skipped += 1
+                logger.warning(
+                    "push target skipped unsupported vendor user_id=%s type=%s "
+                    "notification_id=%s device_id=%s raw_vendor=%s normalized_vendor=%s",
+                    row.user_id, row.type, row.id, dev.device_id, dev.push_vendor, vendor,
                 )
                 continue
+            missing = vendor_push.missing_settings(vendor)
+            if missing:
+                skipped += 1
+                # 本地/测试环境凭据不齐 → 只落站内消息,不发真推送(与 push/vendors 的报缺口径一致)
+                logger.warning(
+                    "push target skipped vendor not configured user_id=%s type=%s "
+                    "notification_id=%s device_id=%s vendor=%s missing_settings=%s",
+                    row.user_id, row.type, row.id, dev.device_id, vendor, missing,
+                )
+                continue
+            logger.info(
+                "push send started user_id=%s type=%s notification_id=%s "
+                "device_id=%s vendor=%s",
+                row.user_id, row.type, row.id, dev.device_id, vendor,
+            )
             try:
-                vendor_push.send_notification(
+                response = vendor_push.send_notification(
                     vendor, dev.push_token, title=title, body=body, extras=extras
                 )
+                if vendor == "huawei" and row.type in {"feedback_reply", "feedback_reward"}:
+                    # 华为反馈推送联调日志：保留厂商返回码/requestId 等排障信息，
+                    # 请求本身的 token、Authorization 和应用密钥不会进入 response。
+                    logger.info(
+                        "huawei feedback push response user_id=%s type=%s "
+                        "notification_id=%s response=%s",
+                        row.user_id, row.type, row.id, response,
+                    )
                 logger.info(
-                    "push sent user_id=%s type=%s vendor=%s notification_id=%s",
-                    row.user_id, row.type, vendor, row.id,
+                    "push sent user_id=%s type=%s vendor=%s notification_id=%s device_id=%s",
+                    row.user_id, row.type, vendor, row.id, dev.device_id,
+                )
+                sent += 1
+            except vendor_push.VendorPushError as e:
+                failed += 1
+                logger.warning(
+                    "push failed user_id=%s type=%s vendor=%s notification_id=%s "
+                    "device_id=%s error=%s",
+                    row.user_id, row.type, vendor, row.id, dev.device_id, e,
+                )
+            try:
+                data_response = vendor_push.send_data_event(
+                    vendor,
+                    dev.push_token,
+                    event=vendor_push.DATA_EVENT_NOTIFICATION_CREATED,
+                    notification_id=str(row.id),
+                )
+                data_sent += 1
+                logger.info(
+                    "push data event completed user_id=%s type=%s vendor=%s "
+                    "notification_id=%s device_id=%s response=%s",
+                    row.user_id, row.type, vendor, row.id, dev.device_id, data_response,
                 )
             except vendor_push.VendorPushError as e:
+                data_failed += 1
+                # 透传只负责前台铃铛实时刷新，失败不能影响通知栏消息或站内消息。
                 logger.warning(
-                    "push failed user_id=%s type=%s vendor=%s: %s", row.user_id, row.type, vendor, e
+                    "push data event failed user_id=%s type=%s vendor=%s notification_id=%s "
+                    "device_id=%s error=%s",
+                    row.user_id, row.type, vendor, row.id, dev.device_id, e,
                 )
+        logger.info(
+            "push dispatch completed user_id=%s type=%s notification_id=%s targets=%s "
+            "sent=%s failed=%s skipped=%s data_sent=%s data_failed=%s",
+            row.user_id, row.type, row.id, len(targets),
+            sent, failed, skipped, data_sent, data_failed,
+        )
     except Exception:  # noqa: BLE001 — 渲染/查设备等意外失败同样不外抛
         logger.exception("push notification failed user_id=%s type=%s", row.user_id, row.type)
 
@@ -156,7 +244,12 @@ def _push_to_user_devices(db: Session, row: Notification, push_vars: dict[str, s
 # ---------------------------------------------------------------------------
 
 
-def notify_withdraw_success(db: Session, order: WithdrawOrder) -> None:
+def notify_withdraw_success(
+    db: Session,
+    order: WithdrawOrder,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """#3 提现成功:款项已存入微信零钱。点击无跳转仅消红点(extra 空)。"""
     _dispatch(
         db,
@@ -170,10 +263,16 @@ def notify_withdraw_success(db: Session, order: WithdrawOrder) -> None:
         extra={},
         dedup_key=order.out_bill_no,
         push_vars={"amount": notif_repo.cash_yuan(order.amount_cents)},
+        push_token_contains=push_token_contains,
     )
 
 
-def notify_withdraw_failed(db: Session, order: WithdrawOrder) -> None:
+def notify_withdraw_failed(
+    db: Session,
+    order: WithdrawOrder,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """#4 提现失败/退回:含微信侧失败、审核拒绝、解绑退回。点击跳提现页重新提现。
 
     失败原因用 order.fail_reason(与 /withdraw/status 下发的用户可读原因同源)。
@@ -191,10 +290,16 @@ def notify_withdraw_failed(db: Session, order: WithdrawOrder) -> None:
         extra={"withdrawId": order.out_bill_no},
         dedup_key=order.out_bill_no,
         push_vars={"amount": notif_repo.cash_yuan(order.amount_cents), "reason": reason},
+        push_token_contains=push_token_contains,
     )
 
 
-def notify_feedback_reply(db: Session, feedback: Feedback) -> None:
+def notify_feedback_reply(
+    db: Session,
+    feedback: Feedback,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """#9 官方回复:运营审核了反馈且未采纳(用户可见原因/留言落在反馈记录上)。
     点击跳反馈历史页滚动高亮该条(extra.feedbackId)。"""
     _dispatch(
@@ -204,10 +309,16 @@ def notify_feedback_reply(db: Session, feedback: Feedback) -> None:
         info_rows=[{"label": "说明文案", "value": "快去看看官方给您的回复吧～"}],
         extra={"feedbackId": str(feedback.id)},
         dedup_key=str(feedback.id),
+        push_token_contains=push_token_contains,
     )
 
 
-def notify_feedback_reward(db: Session, feedback: Feedback) -> None:
+def notify_feedback_reward(
+    db: Session,
+    feedback: Feedback,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """#10 反馈奖励:反馈被采纳,金币已到账。PRD 约定发奖必带官方留言(admin_reply);
     运营漏填时省略该信息行,不硬造文案。"""
     coins = int(feedback.reward_coins or 0)
@@ -225,10 +336,16 @@ def notify_feedback_reward(db: Session, feedback: Feedback) -> None:
         extra={"feedbackId": str(feedback.id)},
         dedup_key=str(feedback.id),
         push_vars={"coins": str(coins)},
+        push_token_contains=push_token_contains,
     )
 
 
-def notify_report_approved(db: Session, report: PriceReport) -> None:
+def notify_report_approved(
+    db: Session,
+    report: PriceReport,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """#11 爆料审核通过:上报的更低价过审,金币已到账。点击跳爆料记录页高亮该条。"""
     coins = int(report.reward_coins or 0)
     store = (report.store_name or "").strip() or "该店铺"
@@ -244,11 +361,17 @@ def notify_report_approved(db: Session, report: PriceReport) -> None:
         extra={"reportId": str(report.id)},
         dedup_key=str(report.id),
         push_vars={"store": store, "coins": str(coins)},
+        push_token_contains=push_token_contains,
     )
 
 
 def notify_invite_order_reward(
-    db: Session, *, inviter_user_id: int, invitee_user_id: int, cash_cents: int
+    db: Session,
+    *,
+    inviter_user_id: int,
+    invitee_user_id: int,
+    cash_cents: int,
+    push_token_contains: str | None = None,
 ) -> None:
     """#12 好友下单到账:被邀请好友完成首次下单(比价),现金奖励已入邀请人账户。
     通知发给【邀请人】;每个好友只发一次奖 → dedup 按被邀请人。"""
@@ -266,4 +389,5 @@ def notify_invite_order_reward(
         extra={"inviteeNickname": nickname},
         dedup_key=str(invitee_user_id),
         push_vars={"nickname": nickname, "amount": _yuan_trim(cash_cents)},
+        push_token_contains=push_token_contains,
     )

docs/api/push-vendor-test.md

@@ -87,6 +87,7 @@ PRD §5 的 13 条 push 文案(标题固定 ≤11 字不带变量;正文 `{var}`
 
 **真发注意**:
 - 目标手机必须先装 App 且客户端已集成对应厂商 SDK、`/device/register` 上报过 token;
+- 荣耀正式环境必须使用正式消息(`HONOR_PUSH_TARGET_USER_TYPE=0`);开发联调测试设备时才设为 `1`;
 - vivo 未上架前走测试推送(`VIVO_PUSH_MODE=1`),目标手机需在 vivo 开放平台加入测试设备;
 - 小米新设备需在开放平台把签名/包名配好,token 才有效。
 

docs/superpowers/plans/2026-07-30-sms-jiguang-chuanglan-fallback.md

@@ -0,0 +1,639 @@
+# 短信 极光→创蓝 容灾 fallback 实现计划
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** 极光为主、创蓝为备的短信验证码容灾——极光供应商不可用(503)时自动转创蓝补发，发/校验全程后台可区分实际渠道，由配置开关控制、可秒回退。
+
+**Architecture:** 全部逻辑收在分派层 `app/integrations/sms/__init__.py`（provider 文件 jiguang/chuanglan/aliyun 一行不动）。`send_code` 返回携带渠道的 `SendResult`；`verify_code` 用 try-both 遍历「启用的 fallback 链」。开关 `SMS_FALLBACK_PROVIDER` 默认空=关。
+
+**Tech Stack:** FastAPI + pydantic-settings + pytest(monkeypatch)。SMS provider 均 Mode B(本地内存自管码)。
+
+设计依据：`docs/superpowers/specs/2026-07-30-sms-jiguang-chuanglan-fallback-design.md`
+
+---
+
+## 文件结构
+
+| 文件 | 职责 | 动作 |
+|---|---|---|
+| `app/integrations/sms/base.py` | 加 `SendResult` 结果类型（provider 无关，与 `SmsError` 同层） | 修改 |
+| `app/core/config.py` | 加 `SMS_FALLBACK_PROVIDER` 开关 | 修改 |
+| `app/integrations/sms/__init__.py` | 分派层：主/备解析 + `send_code` fallback + `verify_code` try-both | 修改（核心） |
+| `app/api/v1/auth.py` | `sms_send` 读 `SendResult.cooldown_sec` + `details` 记 `provider`/`fallback` | 修改（一处） |
+| `.env.example` | 加 `SMS_FALLBACK_PROVIDER=` 注释 | 修改 |
+| `tests/test_sms_dispatch.py` | 现有 3 处 `== 60` 断言改读 `.cooldown_sec` | 修改 |
+| `tests/test_sms_fallback.py` | fallback + try-both 全量新测试 | 新建 |
+| `tests/test_auth.py` | `sms_send` 成功 `details` 含 provider/fallback | 修改（加 1 用例） |
+
+`jiguang.py` / `chuanglan.py` / `aliyun.py` **不动**。
+
+---
+
+## Task 1: `SendResult` 结果类型（base.py）
+
+**Files:**
+- Modify: `app/integrations/sms/base.py`
+- Test: `tests/test_sms_base.py`（新建）
+
+- [ ] **Step 1: 写失败测试**
+
+Create `tests/test_sms_base.py`:
+
+```python
+"""SMS 共享基座类型 SendResult。"""
+from __future__ import annotations
+
+import dataclasses
+
+import pytest
+
+from app.integrations.sms.base import SendResult
+
+
+def test_send_result_fields():
+    r = SendResult(cooldown_sec=60, provider="jiguang", fallback=True)
+    assert r.cooldown_sec == 60
+    assert r.provider == "jiguang"
+    assert r.fallback is True
+
+
+def test_send_result_fallback_defaults_false():
+    r = SendResult(cooldown_sec=60, provider="jiguang")
+    assert r.fallback is False
+
+
+def test_send_result_is_frozen():
+    r = SendResult(cooldown_sec=60, provider="jiguang")
+    with pytest.raises(dataclasses.FrozenInstanceError):
+        r.provider = "chuanglan"
+```
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_sms_base.py -q`
+Expected: FAIL — `ImportError: cannot import name 'SendResult'`
+
+- [ ] **Step 3: 最小实现**
+
+在 `app/integrations/sms/base.py` 顶部 import 区加 `from dataclasses import dataclass`，并在 `SmsError` 类后新增：
+
+```python
+@dataclass(frozen=True)
+class SendResult:
+    """发码结果：距下次可发秒数 + 实际发码的 provider 名 + 是否走了备(fallback)。
+
+    分派层 send_code 的返回类型（provider 各自的 send_code 仍返回 int cooldown，
+    由分派层包装）。auth 层据此把渠道记入风控流水 details。
+    """
+
+    cooldown_sec: int
+    provider: str
+    fallback: bool = False
+```
+
+- [ ] **Step 4: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_sms_base.py -q`
+Expected: PASS (3 passed)
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add app/integrations/sms/base.py tests/test_sms_base.py
+git commit -m "feat(sms): 加 SendResult 结果类型(cooldown+provider+fallback)"
+```
+
+---
+
+## Task 2: `SMS_FALLBACK_PROVIDER` 配置开关（config.py）
+
+**Files:**
+- Modify: `app/core/config.py`（`SMS_PROVIDER` 定义附近，约 line 150）
+- Test: `tests/test_sms_fallback.py`（新建，本 Task 起逐步补全）
+
+- [ ] **Step 1: 写失败测试**
+
+Create `tests/test_sms_fallback.py`:
+
+```python
+"""极光→创蓝 容灾 fallback（分派层 send_code / verify_code）。
+
+monkeypatch 各 provider 的 send_code/verify_code 为标记函数，不发真网络；
+断言：仅主 503 才转备；429/400 不转；verify try-both 只遍历启用的链。
+"""
+from __future__ import annotations
+
+import pytest
+
+from app.core.config import settings
+from app.integrations import sms
+from app.integrations.sms import aliyun, chuanglan, jiguang
+from app.integrations.sms.base import SendResult, SmsError
+
+PHONE = "13800138000"
+
+
+def _raiser(exc: Exception):
+    """返回一个调用即 raise exc 的 send_code 替身（lambda 不能 raise）。"""
+    def _f(phone):
+        raise exc
+    return _f
+
+
+def test_fallback_provider_defaults_empty():
+    assert settings.SMS_FALLBACK_PROVIDER == ""
+```
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_sms_fallback.py::test_fallback_provider_defaults_empty -q`
+Expected: FAIL — `AttributeError: 'Settings' object has no attribute 'SMS_FALLBACK_PROVIDER'`
+
+- [ ] **Step 3: 最小实现**
+
+在 `app/core/config.py` 的 `SMS_PROVIDER` 定义行下方新增：
+
+```python
+    # 主 provider 供应商不可用(SmsError 503)时转投的备 provider；空=关闭 fallback(默认，
+    # 保持现状零风险)。仅 Mode B(jiguang/chuanglan)互为主备有意义；生产建议设 "chuanglan"。
+    SMS_FALLBACK_PROVIDER: Literal["jiguang", "chuanglan", ""] = ""
+```
+
+- [ ] **Step 4: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_sms_fallback.py::test_fallback_provider_defaults_empty -q`
+Expected: PASS
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add app/core/config.py tests/test_sms_fallback.py
+git commit -m "feat(sms): 加 SMS_FALLBACK_PROVIDER 开关(默认空=关)"
+```
+
+---
+
+## Task 3: 分派层 `send_code` 返回 SendResult（主路径）
+
+先让主路径返回 `SendResult`（不含 fallback），并把现有 `test_sms_dispatch.py` 的 `== 60` 断言迁到 `.cooldown_sec`。fallback 分支在 Task 4 加。
+
+**Files:**
+- Modify: `app/integrations/sms/__init__.py`
+- Modify: `tests/test_sms_dispatch.py`（line 20/22/24、line 45-53）
+- Test: `tests/test_sms_fallback.py`
+
+- [ ] **Step 1: 写失败测试**
+
+在 `tests/test_sms_fallback.py` 追加：
+
+```python
+def test_send_code_primary_success_returns_sendresult(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")
+    monkeypatch.setattr(jiguang, "send_code", lambda phone: 60)
+
+    r = sms.send_code(PHONE)
+    assert isinstance(r, SendResult)
+    assert r.cooldown_sec == 60
+    assert r.provider == "jiguang"
+    assert r.fallback is False
+```
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_sms_fallback.py::test_send_code_primary_success_returns_sendresult -q`
+Expected: FAIL — `assert isinstance(r, SendResult)`（现 send_code 返回 int 60）
+
+- [ ] **Step 3: 最小实现**
+
+把 `app/integrations/sms/__init__.py` 整体替换为（本 Task 只到主路径；Task 4/5 再扩 fallback 与 try-both）：
+
+```python
+"""短信验证码服务 —— provider 分派入口（极光主 + 可选创蓝备）。
+
+对外暴露 `send_code`（返回 SendResult） / `verify_code` / `SmsError` / `SendResult`。
+主 provider = settings.SMS_PROVIDER（默认 jiguang）；备 = settings.SMS_FALLBACK_PROVIDER
+（默认空=无备）。**每次调用读 settings**，支持运行时切换 / 灰度回退。
+"""
+from __future__ import annotations
+
+import logging
+
+from app.core.config import settings
+
+from . import aliyun, chuanglan, jiguang
+from .base import SendResult, SmsError
+
+__all__ = ["SmsError", "SendResult", "send_code", "verify_code"]
+
+logger = logging.getLogger("shagua.sms")
+
+# provider 名 <-> 模块。未知/缺省主 provider 回退 jiguang（防误配把登录打挂）。
+_ALL = {"jiguang": jiguang, "aliyun": aliyun, "chuanglan": chuanglan}
+_NAME = {jiguang: "jiguang", aliyun: "aliyun", chuanglan: "chuanglan"}
+
+
+def _primary():
+    """当前主 provider 模块（未知/缺省回退 jiguang）。"""
+    return _ALL.get(settings.SMS_PROVIDER, jiguang)
+
+
+def _fallback():
+    """启用的备 provider 模块；未配置 / 同主 / 未知名 → None（= 不启用 fallback）。"""
+    name = (settings.SMS_FALLBACK_PROVIDER or "").strip()
+    if not name or name == settings.SMS_PROVIDER:
+        return None
+    return _ALL.get(name)
+
+
+def send_code(phone: str) -> SendResult:
+    """发码：主成功即返回；失败抛 SmsError。返回 SendResult（cooldown + 实际渠道）。"""
+    primary = _primary()
+    cooldown = primary.send_code(phone)
+    return SendResult(cooldown_sec=cooldown, provider=_NAME[primary], fallback=False)
+
+
+def verify_code(phone: str, code: str) -> bool:
+    """校验：委托当前主 provider（try-both 在 Task 5 加）。"""
+    return _primary().verify_code(phone, code)
+```
+
+- [ ] **Step 4: 迁移现有 dispatch 测试断言**
+
+在 `tests/test_sms_dispatch.py` 中，把 3 处 `assert sms.send_code("13800138000") == 60`（line 20/22/24）改为：
+
+```python
+    assert sms.send_code("13800138000").cooldown_sec == 60
+```
+
+（line 45-53 的 `test_unknown_provider_falls_back_to_jiguang` 用 `sms.send_code(...)` 不比较返回值，无需改；`test_verify_code_routes_by_provider` 不涉及返回值，无需改。）
+
+- [ ] **Step 5: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_sms_fallback.py tests/test_sms_dispatch.py -q`
+Expected: PASS（新用例 + 原 dispatch 3 用例全绿）
+
+- [ ] **Step 6: 提交**
+
+```bash
+git add app/integrations/sms/__init__.py tests/test_sms_dispatch.py tests/test_sms_fallback.py
+git commit -m "feat(sms): send_code 返回 SendResult(主路径),迁移 dispatch 断言"
+```
+
+---
+
+## Task 4: 分派层 `send_code` fallback 分支（仅 503 转备）
+
+**Files:**
+- Modify: `app/integrations/sms/__init__.py`（`send_code`）
+- Test: `tests/test_sms_fallback.py`
+
+- [ ] **Step 1: 写失败测试**
+
+在 `tests/test_sms_fallback.py` 追加：
+
+```python
+def test_send_code_503_falls_back_to_configured_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        lambda phone: (calls.append("chuanglan"), 60)[1])
+
+    r = sms.send_code(PHONE)
+    assert r.provider == "chuanglan"
+    assert r.fallback is True
+    assert r.cooldown_sec == 60
+    assert calls == ["chuanglan"]
+
+
+@pytest.mark.parametrize("status", [429, 400])
+def test_send_code_non_503_does_not_fall_back(monkeypatch, status):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("x", status_code=status)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        lambda phone: (calls.append("chuanglan"), 60)[1])
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert ei.value.status_code == status
+    assert calls == []                      # 备未被调用
+
+
+def test_send_code_503_without_backup_reraises(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")   # 无备
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert ei.value.status_code == 503
+
+
+def test_send_code_backup_also_fails_raises_backup_error(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        _raiser(SmsError("创蓝也不可用", status_code=503)))
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert "创蓝" in str(ei.value)
+```
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_sms_fallback.py -k "fall_back or 503 or non_503 or backup" -q`
+Expected: FAIL — 主抛 503 时当前 `send_code` 未捕获，直接透出（`test_...falls_back...` 期望走到创蓝但没走）
+
+- [ ] **Step 3: 实现 fallback 分支**
+
+把 `app/integrations/sms/__init__.py` 的 `send_code` 替换为：
+
+```python
+def send_code(phone: str) -> SendResult:
+    """发码：主成功即返回；仅主「供应商不可用(503)」且配置了备时转备补发。
+
+    429（本地冷却/超频）、400（手机号无效）不转——不绕过防刷、不为无效号白烧。
+    备也失败则抛备的 SmsError。返回 SendResult（cooldown + 实际渠道 + 是否 fallback）。
+    """
+    primary = _primary()
+    fb = _fallback()
+    try:
+        cooldown = primary.send_code(phone)
+        return SendResult(cooldown_sec=cooldown, provider=_NAME[primary], fallback=False)
+    except SmsError as e:
+        if fb is not None and e.status_code == 503:
+            logger.warning("[SMS] primary=%s 不可用(%s)，fallback→%s",
+                           _NAME[primary], e, _NAME[fb])
+            cooldown = fb.send_code(phone)   # 备的冷却/错误码原样透出
+            return SendResult(cooldown_sec=cooldown, provider=_NAME[fb], fallback=True)
+        raise
+```
+
+- [ ] **Step 4: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_sms_fallback.py -q`
+Expected: PASS（含 parametrize 的 429/400 两条）
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add app/integrations/sms/__init__.py tests/test_sms_fallback.py
+git commit -m "feat(sms): send_code 仅主 503 时转创蓝补发(429/400 不转)"
+```
+
+---
+
+## Task 5: 分派层 `verify_code` try-both（只遍历启用的链）
+
+**Files:**
+- Modify: `app/integrations/sms/__init__.py`（`verify_code`）
+- Test: `tests/test_sms_fallback.py`
+
+- [ ] **Step 1: 写失败测试**
+
+在 `tests/test_sms_fallback.py` 追加：
+
+```python
+def test_verify_hits_primary_without_touching_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), True)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is True
+    assert calls == ["jiguang"]             # 主命中即停，不查备
+
+
+def test_verify_falls_through_to_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), False)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is True
+    assert calls == ["jiguang", "chuanglan"]  # 主未命中 → 查备（码在备）
+
+
+def test_verify_all_miss_returns_false(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code", lambda p, c: False)
+    monkeypatch.setattr(chuanglan, "verify_code", lambda p, c: False)
+
+    assert sms.verify_code(PHONE, "123456") is False
+
+
+def test_verify_backup_not_touched_when_fallback_off(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")   # 关闭 fallback
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), False)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is False
+    assert calls == ["jiguang"]             # 关闭时创蓝零参与
+```
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_sms_fallback.py -k verify -q`
+Expected: FAIL — `test_verify_falls_through_to_backup`（当前 verify 只问主，主 False 即返 False，不查备）
+
+- [ ] **Step 3: 实现 try-both**
+
+把 `app/integrations/sms/__init__.py` 的 `verify_code` 替换为：
+
+```python
+def verify_code(phone: str, code: str) -> bool:
+    """校验：try-both，遍历「启用的 fallback 链」（主→备），任一命中即 True。
+
+    码只存在实际发码那家（fallback 前主已 pop 掉自己的码），另一家 rec is None 即 False、
+    不误判、不累加其防爆破计数。关闭 fallback 时链中只有主，备完全不参与。
+    """
+    chain = [_primary()]
+    fb = _fallback()
+    if fb is not None:
+        chain.append(fb)
+    for prov in chain:
+        if prov.verify_code(phone, code):   # Mode B：纯本地内存比对，不联网
+            logger.info("[SMS] verify hit provider=%s", _NAME[prov])
+            return True
+    return False
+```
+
+- [ ] **Step 4: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_sms_fallback.py tests/test_sms_dispatch.py -q`
+Expected: PASS（全部 fallback + 原 dispatch 用例）
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add app/integrations/sms/__init__.py tests/test_sms_fallback.py
+git commit -m "feat(sms): verify_code try-both 遍历启用的 fallback 链"
+```
+
+---
+
+## Task 6: auth `sms_send` 接入 SendResult + 记 provider/fallback
+
+**Files:**
+- Modify: `app/api/v1/auth.py`（line 210 调用、line 231-243 成功侧 details）
+- Test: `tests/test_auth.py`（加 1 用例）
+
+- [ ] **Step 1: 写失败测试**
+
+先确认现有 `test_auth.py` 里 `sms_send` 的调用/断言与 fixture 写法（找一个已有的 sms_send 成功用例作参照），再在 `tests/test_auth.py` 末尾追加：
+
+```python
+def test_sms_send_records_provider_in_details(client, db_session, monkeypatch):
+    """sms_send 成功后风控流水 details 带 provider/fallback（mock 下 jiguang/False）。"""
+    from app.repositories import risk as risk_repo
+
+    captured = {}
+    orig = risk_repo.record_behavior_event
+
+    def _spy(db, **kw):
+        if kw.get("event_type") == risk_repo.EVENT_SMS_SEND and kw.get("outcome") == "success":
+            captured.update(kw.get("details") or {})
+        return orig(db, **kw)
+
+    monkeypatch.setattr(risk_repo, "record_behavior_event", _spy)
+
+    resp = client.post("/api/v1/auth/sms/send",
+                       json={"phone": "13800009999", "device_id": "dev-provider-test"})
+    assert resp.status_code == 200, resp.text
+    assert captured.get("provider") == "jiguang"
+    assert captured.get("fallback") is False
+```
+
+> 注：`client` / `db_session` fixture 名、请求路径前缀（`/api/v1/auth/...`）、`record_behavior_event` 的 import 路径以 `tests/test_auth.py` 现有用例为准，若不同则对齐后再写。
+
+- [ ] **Step 2: 跑测试确认失败**
+
+Run: `python -m pytest tests/test_auth.py::test_sms_send_records_provider_in_details -q`
+Expected: FAIL — `captured.get("provider")` 为 None（当前 details 只有 `{"mock": ...}`）
+
+- [ ] **Step 3: 实现**
+
+在 `app/api/v1/auth.py` 的 `sms_send` 中：
+
+(a) line 209-210 的
+```python
+    try:
+        cooldown = send_code(req.phone)
+```
+改为
+```python
+    try:
+        send_result = send_code(req.phone)
+        cooldown = send_result.cooldown_sec
+```
+
+(b) line 231-243 成功侧 `record_behavior_event` 的 `details` 由
+```python
+        details={"mock": settings.SMS_MOCK},
+```
+改为
+```python
+        details={"mock": settings.SMS_MOCK,
+                 "provider": send_result.provider,
+                 "fallback": send_result.fallback},
+```
+
+- [ ] **Step 4: 跑测试确认通过**
+
+Run: `python -m pytest tests/test_auth.py::test_sms_send_records_provider_in_details -q`
+Expected: PASS
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add app/api/v1/auth.py tests/test_auth.py
+git commit -m "feat(sms): sms_send 记 provider/fallback 到风控流水 details"
+```
+
+---
+
+## Task 7: `.env.example` + 全量回归 + ruff
+
+**Files:**
+- Modify: `.env.example`
+
+- [ ] **Step 1: 加配置注释**
+
+在 `.env.example` 的 `SMS_PROVIDER` 附近加：
+
+```bash
+# 极光→创蓝 容灾：主 provider(SMS_PROVIDER)供应商不可用(503)时转投的备。
+# 空=关闭 fallback(默认)；生产设 chuanglan 开启极光主、创蓝备；置空即秒回退。
+SMS_FALLBACK_PROVIDER=
+```
+
+- [ ] **Step 2: 全量 SMS 相关回归**
+
+Run: `python -m pytest tests/test_sms_fallback.py tests/test_sms_dispatch.py tests/test_sms_chuanglan.py tests/test_sms_aliyun.py tests/test_auth.py -q`
+Expected: PASS（新增全绿；原 SMS/auth 用例不回归）
+
+- [ ] **Step 3: lint**
+
+Run: `python -m ruff check app/integrations/sms/ app/core/config.py app/api/v1/auth.py tests/test_sms_fallback.py tests/test_sms_base.py`
+Expected: `All checks passed!`
+
+- [ ] **Step 4: 基线对比（确认无新增失败）**
+
+Run: `python -m pytest -q`
+Expected: 失败数 ≤ 基线（已知先前债 ~8 个）；本改动不新增失败。若有新增，定位并修。
+
+- [ ] **Step 5: 提交**
+
+```bash
+git add .env.example
+git commit -m "docs(sms): .env.example 加 SMS_FALLBACK_PROVIDER 说明"
+```
+
+---
+
+## Self-Review
+
+**Spec 覆盖核对**（对 `2026-07-30-sms-jiguang-chuanglan-fallback-design.md`）：
+
+- §2-A 仅 503 转 / 429·400 不转 → Task 4（`test_send_code_503_falls_back` + `test_send_code_non_503_does_not_fall_back`）✅
+- §2-B try-both 只遍历启用链 → Task 5（4 个 verify 用例，含关闭时备零参与）✅
+- §2-C details 记 provider/fallback + 日志 → Task 6（details）+ Task 4/5（logger.warning/info）✅
+- §2-D 配置开关默认空、生产设开 → Task 2 + Task 7 ✅
+- §4.1 主/备解析（未知/同主→None）→ Task 3（`_primary`/`_fallback`）✅
+- §4.2 SendResult + 只认 503 + 备失败抛备错 → Task 1 + Task 4（`test_send_code_backup_also_fails`）✅
+- §4.4 码只在一家/防爆破不被绕过 → provider 内部行为（不改），try-both 逻辑 Task 5 覆盖；防爆破由 provider 各自既有测试保证 ✅
+- §4.5 auth 一处改、verify 调用点不变 → Task 6 ✅
+- §5 配置项 → Task 2 ✅
+- §6 mock 永不 fallback → mock 下主 send_code 成功不抛异常，天然不进 fallback 分支（现有 mock 测试在 Task 7 回归保证）✅
+- §10 send_code int→SendResult 仅 auth 一处 + dispatch 测试 → Task 3/6 ✅
+
+**占位扫描**：无 TBD/TODO；每个 code step 均含完整代码与命令。Task 6 Step 1 标注了「以现有 fixture 为准」——这是**对齐现有约定**的指示，非占位（fixture 名在同文件已存在，执行时照抄）。
+
+**类型/命名一致性**：`SendResult(cooldown_sec, provider, fallback)` 三字段在 Task 1/3/4/6 全程一致；`_primary()`/`_fallback()`/`_ALL`/`_NAME` 命名跨 Task 3/4/5 一致；`SMS_FALLBACK_PROVIDER` 跨 Task 2/3/4/5/7 一致。
+
+---
+
+## 执行顺序说明
+
+Task 1→2→3 有依赖（后者用前者的类型/配置/分派骨架），必须顺序执行。Task 4/5 都改 `send_code`/`verify_code`，也应在 Task 3 之后顺序做。Task 6 依赖 Task 3 的 `SendResult` 返回。Task 7 收尾。全程一条功能分支 `feat/sms-jiguang-chuanglan-fallback`，每 Task 一提交。

docs/superpowers/specs/2026-07-30-sms-jiguang-chuanglan-fallback-design.md

@@ -0,0 +1,208 @@
+# 短信验证码 极光→创蓝 容灾 fallback — 设计方案
+
+- 日期：2026-07-30
+- 状态：待评审
+- 范围：短信发送验证码服务，默认极光；**极光供应商不可用时自动转创蓝补发**；发/校验全程后台可区分本次走的是极光还是创蓝。
+- 关联：`docs/superpowers/specs/2026-07-26-chuanglan-sms-verify-design.md`（创蓝 provider 接入）
+
+## 1. 背景与目标
+
+短信验证码已是**可切换 provider** 架构（`app/integrations/sms/`：`__init__` 分派 + `jiguang` / `aliyun` / `chuanglan` + `base`）。极光（默认）与创蓝都是 **Mode B（自管码）**：本服务 `secrets` 生成码 → 存**各自模块的进程内存** → provider 只负责发送；`verify_code` 是**纯本地内存比对**，不联网。
+
+现状是"单选一个 provider"：`send_code` / `verify_code` 都只走 `SMS_PROVIDER` 指定的那一家。极光一旦供应商侧故障（欠费 / 网络 / 服务异常），`/sms/send` 直接 503，用户收不到码、登录中断。
+
+**目标**：把极光设为主、创蓝设为备，在**极光供应商不可用**时自动转创蓝补发，最大化验证码送达率；同时让后台能查到每次发码/校验实际走的是哪家、是否发生了 fallback。
+
+**非目标**：不改极光/创蓝各自的 Mode B 机器（发码/存码/校验/防刷），不引入 DB/Redis 共享存码，不做多 worker 共享（沿用现有内存态技术债）。
+
+## 2. 关键决策（已与需求方确认）
+
+| # | 决策 | 结论 |
+|---|---|---|
+| A | **fallback 触发范围** | **仅主 provider 返回"供应商不可用"（`SmsError.status_code == 503`：网络错误 / 余额不足 / 服务故障）时转备。** 本地单号 60s 冷却与极光侧超频（均 429）、手机号无效（400）**不转**——不绕过防刷、不为无效号白烧一条。 |
+| B | **校验路由** | **try-both，但只遍历"当前启用的 fallback 链"。** verify 依次问链上 provider（主→备），任一命中即通过；关闭 fallback 时链中只有主，创蓝完全不参与。 |
+| C | **后台可见性** | **风控行为流水 + 结构化日志。** 复用 `risk_repo.record_behavior_event(EVENT_SMS_SEND)` 的 `details` 加 `provider` / `fallback` 字段；dispatch 层对 fallback 触发、校验命中打点。不新增表、不改客户端响应。 |
+| D | **默认开关** | **配置开关 opt-in，生产设开。** 新增 `SMS_FALLBACK_PROVIDER`，默认空=关（保持现状零风险）；生产 `.env` 设 `chuanglan` 开启。出问题置空即秒回退、不改代码。 |
+
+补充约束：**fallback 仅在 Mode B ↔ Mode B（jiguang / chuanglan）之间有意义**。aliyun 是 Mode A（阿里云托管码、校验走接口），不参与 fallback 链。
+
+## 3. 现状回顾（设计依赖的事实）
+
+- **分派层** [`app/integrations/sms/__init__.py`](../../../app/integrations/sms/__init__.py)：`_provider()` 每次按 `settings.SMS_PROVIDER` 选一个模块（默认 jiguang），`send_code` / `verify_code` 委托给它。
+- **极光 / 创蓝均 Mode B**，各持一份**独立**的 `_codes` / `_last_sent` / `_lock`（两个互不相通的进程内存 dict）。`verify_code` 纯本地比对（`_codes.get(phone)` + `secrets.compare_digest`），**不发网络**；provider 的网络调用只在**发送**一步。
+- **`SmsError(message, status_code)`**（[`base.py`](../../../app/integrations/sms/base.py)）：`429`=过频/超频（客户端稍后再来）、`503`=供应商不可用、`400`=手机号无效。这是 A 决策用来判定"是否 fallback"的唯一信号。
+- **极光/创蓝 `send_code` 失败时"保留冷却、清掉未发出的码"**：`_last_sent[phone]=now` 保留、`_codes.pop(phone)`。此行为不改（见 §4.3 冷却交互）。
+- **auth 层落点**：`sms_send` 发码成功 [auth.py:231](../../../app/api/v1/auth.py) 记 `EVENT_SMS_SEND` `outcome="success" details={"mock":...}`；失败 [auth.py:212](../../../app/api/v1/auth.py) 记 `outcome="failed" reason=...`。`send_code` 是内部函数，全仓仅此一处调用。`verify_code` 在 `sms_login` 与换绑两处调用，签名 `(phone, code)->bool`。
+
+## 4. 设计
+
+### 4.1 fallback 链模型（分派层，`__init__.py`）
+
+引入"主 + 备"两级链，全部在分派层组装，**provider 文件一行不动**：
+
+```python
+_ALL = {"jiguang": jiguang, "aliyun": aliyun, "chuanglan": chuanglan}
+_NAME = {jiguang: "jiguang", aliyun: "aliyun", chuanglan: "chuanglan"}
+
+def _primary():
+    # 未知/缺省回退 jiguang（保持 test_unknown_provider_falls_back_to_jiguang 语义）
+    return _ALL.get(settings.SMS_PROVIDER, jiguang)
+
+def _fallback():
+    """启用的备 provider；未配置/同主/未知名 → None（= 不启用 fallback）。"""
+    name = (settings.SMS_FALLBACK_PROVIDER or "").strip()
+    if not name or name == settings.SMS_PROVIDER:
+        return None
+    return _ALL.get(name)  # 未知名 → None
+```
+
+### 4.2 发码 `send_code(phone) -> SendResult`（透出实际渠道）
+
+现返回 `int`（cooldown），改为返回携带渠道信息的小结果对象（放 `base.py`，provider 无关）：
+
+```python
+@dataclass(frozen=True)
+class SendResult:
+    cooldown_sec: int   # 距下次可发秒数（= 实际发码 provider 的 SMS_SEND_INTERVAL_SEC）
+    provider: str       # 实际发码的 provider 名："jiguang" / "chuanglan"
+    fallback: bool      # 是否走了备（True=主不可用转备）
+```
+
+分派逻辑：
+
+```python
+def send_code(phone) -> SendResult:
+    primary, fb = _primary(), _fallback()
+    try:
+        cooldown = primary.send_code(phone)
+        return SendResult(cooldown, _NAME[primary], fallback=False)
+    except SmsError as e:
+        # A 决策：仅"供应商不可用(503)"且配置了备时转备；429/400 原样抛
+        if fb is not None and e.status_code == 503:
+            logger.warning("[SMS] primary=%s 不可用(%s)，fallback→%s",
+                           _NAME[primary], e, _NAME[fb])
+            cooldown = fb.send_code(phone)   # 备也可能抛 SmsError（其冷却/错误码）→ 原样透出
+            return SendResult(cooldown, _NAME[fb], fallback=True)
+        raise
+```
+
+要点：
+- **只认 `status_code == 503`**：极光的 429（本地冷却 / 超频 50009）、400（手机号无效）直接 raise，不 fallback。
+- **备自带完整 Mode B 机器**：创蓝重新走自己的单号冷却 / 存码 / 错误码映射。备也失败则抛备的 `SmsError`，`auth` 按其 `status_code` 返回。
+- **主/备文件不动**：`jiguang.send_code` / `chuanglan.send_code` 仍返回 `int`；由分派层包装成 `SendResult`。
+
+### 4.3 冷却交互（已知行为，保守安全，不改 provider）
+
+极光 `send_code` 失败（503）时**保留自己的冷却**（现有行为，防"欠费/签名失效时前端狂打极光"）。在 fallback 下的连锁：
+
+1. 首发：极光 503（记极光冷却）→ 转创蓝发成功（记创蓝冷却）。用户收到创蓝的码。
+2. 60s 内重发：主=极光**冷却中 → 429**，`429 ≠ 503` → **不 fallback** → 用户收到 429"过于频繁"。**这是期望行为**：挡住 60s 内重复补发烧钱；且此时创蓝也在自己冷却期。
+3. 极光**持续故障**：每过 60s，极光冷却过期→再试极光（仍 503）→转创蓝补发。稳态下每 60s 白试一次极光、成功走创蓝，功能可用。
+
+结论：现有"失败保留冷却"语义在 fallback 下自洽且保守。**不改极光/创蓝任何冷却逻辑。**
+
+### 4.4 校验 `verify_code(phone, code) -> bool`（try-both over 启用链）
+
+```python
+def verify_code(phone, code) -> bool:
+    chain = [_primary()]
+    fb = _fallback()
+    if fb is not None:
+        chain.append(fb)
+    for prov in chain:
+        if prov.verify_code(phone, code):   # Mode B：纯本地内存比对，不联网、不抛异常
+            logger.info("[SMS] verify hit provider=%s", _NAME[prov])
+            return True
+    return False
+```
+
+要点：
+- **关闭 fallback 时 `chain=[极光]`**，与现状字节级等价，**创蓝完全不参与**。
+- **码只存在实际发码那家**（fallback 发码前极光已 `_codes.pop` 清掉自己的码）→ 同一手机号同一时刻只有一家有码；另一家 `rec is None` 立即 False，不误判、不累加其 `attempts`。**防爆破仍在真实那家正常生效，不被绕过**。
+- **命中即知渠道**（日志打点），天然满足 C 的校验侧可见性，无需显式记忆 phone→provider。
+- 绝大多数登录：极光发成功→码在极光→第一家命中，创蓝那次内存查根本不发生。
+- Mode B 的 `verify_code` 不抛 `SmsError`；若链中混入 Mode A（aliyun，本设计约束外）其 verify 降级抛 503，应原样透出中断——但按 §2 约束 aliyun 不入链，此路不触发。
+
+### 4.5 auth.py 改动（最小）
+
+- **`sms_send`**（唯一改动点）：
+  ```python
+  result = send_code(req.phone)              # 现 SendResult
+  cooldown = result.cooldown_sec
+  # 成功侧 details 加 provider/fallback（C）
+  details={"mock": settings.SMS_MOCK, "provider": result.provider, "fallback": result.fallback}
+  ```
+  `SmsSendResponse(cooldown_sec=cooldown, ...)` 不变——**不把 provider 返回客户端**（C 选风控流水，非接口返回）。
+- **`verify_code` 两处调用点不变**：签名 `(phone,code)->bool` 未改；校验命中渠道由分派层 §4.4 日志打点，auth 无需感知。
+- **失败可观测**：极光 503 且创蓝也失败时，抛创蓝的 `SmsError` → 现有失败侧 `record_behavior_event(reason=str(e))` 记录；fallback 触发/失败另有 §4.2 分派层 `logger.warning`。
+
+### 4.6 可观测性汇总（满足 C）
+
+| 事件 | 落点 | 内容 |
+|---|---|---|
+| 发码成功 | `EVENT_SMS_SEND` `details` | `provider`（jiguang/chuanglan）、`fallback`（bool） |
+| fallback 触发 | 分派层 `logger.warning` | 主不可用原因 + 转向的备 |
+| 校验命中 | 分派层 `logger.info` | 命中的 `provider` |
+| 发码彻底失败 | `EVENT_SMS_SEND` `outcome=failed` | `reason`（末端 SmsError 文案） |
+
+## 5. 配置项（`app/core/config.py` 新增）
+
+```python
+# 主 provider 供应商不可用(503)时转投的备 provider；空=关闭 fallback(默认，保持现状)。
+# 仅 Mode B(jiguang/chuanglan)互为主备有意义；生产建议设 "chuanglan"。
+SMS_FALLBACK_PROVIDER: Literal["jiguang", "chuanglan", ""] = ""  # Mode B only(aliyun 不入链)
+```
+
+- 复用现有 `SMS_PROVIDER`（主，默认 jiguang）与两家的凭证/旋钮（`CHUANGLAN_SMS_*`、`SMS_CODE_LENGTH` 等），**无其它新增**。
+- `.env.example` 加 `SMS_FALLBACK_PROVIDER=`（注释：生产设 `chuanglan` 开启极光→创蓝容灾）。
+- 生产前置：创蓝凭证齐全（`chuanglan_sms_configured`）+ 创蓝控制台 IP 白名单，否则 fallback 到创蓝仍 503。
+
+## 6. 错误处理与边界
+
+- **主 503 + 未配备**：`_fallback()` 返 None → 原样抛主的 503（= 现状）。
+- **主 503 + 备也失败**：抛备的 `SmsError`（按备错误码 429/400/503），auth 照常返回。
+- **主 429/400**：不 fallback，原样抛。
+- **mock 模式（`SMS_MOCK=true`）**：主 `send_code` mock 分支直接成功、不抛异常 → **永不触发 fallback**；`verify` mock 分支放行任意 N 位 → try-both 第一家（主）即 True。**现有 mock 测试全绿不变。**
+- **未知 `SMS_FALLBACK_PROVIDER` 名**：`_ALL.get` 返 None → 视为未配置、不 fallback（不崩）。
+- **`SMS_FALLBACK_PROVIDER == SMS_PROVIDER`**：`_fallback()` 返 None（不自我 fallback）。
+
+## 7. 测试计划
+
+新增 `tests/test_sms_fallback.py`（monkeypatch 各 provider 的 `send_code`/`verify_code`，不真发）：
+
+1. 主发成功 → `SendResult(provider=主, fallback=False)`，不调备。
+2. 主 503 + 配备 → 转备，备成功 → `SendResult(provider=备, fallback=True)`。
+3. 主 429 / 400 → 不 fallback，原样抛（断言备未被调用）。
+4. 主 503 + 未配备 → 抛 503。
+5. 主 503 + 备也 503 → 抛备的 SmsError。
+6. verify：码在主 → 命中主；码在备 → 命中备；两家都无 → False。
+7. **关闭 fallback**（`SMS_FALLBACK_PROVIDER=""`）→ verify 只问主，**断言备 `verify_code` 未被调用**；send 主失败原样抛、不转备。
+8. 防爆破：码在主，连续错到 `SMS_MAX_VERIFY_ATTEMPTS` 作废；全程备 `rec is None` 不受影响。
+
+auth 层扩 `tests/test_auth.py`：`sms_send` 成功后 `EVENT_SMS_SEND` 的 `details` 含 `provider`/`fallback`（mock 下 provider=jiguang、fallback=False）。
+
+回归：`test_sms_dispatch.py`、现有 `test_auth.py` 中依赖 `send_code` 返回 `int` 的断言改为读 `SendResult.cooldown_sec`。
+
+## 8. YAGNI（明确不做）
+
+- ❌ 不改极光/创蓝的 Mode B 机器（发/存/校验/冷却/防爆破）。
+- ❌ 不做 DB/Redis 共享存码或 phone→provider 记忆表（多 worker 债维持现状、两家一起将来迁）。
+- ❌ 不做 aliyun（Mode A）参与 fallback。
+- ❌ 不做多级（>2）fallback 链、不做主备自动健康探测/熔断。
+- ❌ 不把 provider 渠道透传给客户端（仅后台可见）。
+- ❌ 不做发送量/成功率统计报表（C 只取"风控流水+日志"，报表另议）。
+
+## 9. 验收标准
+
+- `SMS_FALLBACK_PROVIDER=chuanglan` 且创蓝配置齐全时：极光注入 503（欠费/网络）→ `/sms/send` 实际由创蓝下发、真机可收码并登录；`EVENT_SMS_SEND.details.provider=chuanglan`、`fallback=true`。
+- 极光正常时：走极光，`provider=jiguang`、`fallback=false`；创蓝零调用。
+- `SMS_FALLBACK_PROVIDER=""`（默认）：行为与当前完全一致，创蓝不参与发/校验。
+- 极光 429/400：不转创蓝，原样返回。
+- `SMS_MOCK=true`：任意 N 位通过、不真发、不触发 fallback。
+- `ruff check .` 通过；新增/现有 `pytest` 全绿。
+
+## 10. 影响面 / 回滚
+
+- **改动文件**：`sms/__init__.py`（分派层，主要）、`sms/base.py`（加 `SendResult`）、`core/config.py`（加开关）、`api/v1/auth.py`（`sms_send` 一处记 details）、`.env.example`、测试。**`jiguang.py` / `chuanglan.py` / `aliyun.py` 不动。**
+- **破坏性接口变更**：`send_code` 返回 `int → SendResult`，仅 `auth.py:210` 一处调用 + 相关测试需同步。
+- **回滚**：置 `SMS_FALLBACK_PROVIDER=""` 即运行时秒回退到"纯极光"，无需改代码/迁移。

scripts/fire_push_events.py

@@ -16,6 +16,10 @@ xiaomi/oppo/vivo)到该用户 device_liveness 里已注册的 push token。
   .venv\\Scripts\\python.exe scripts\\fire_push_events.py --count 1          # 各 1 条(先小量验证通道)
   .venv\\Scripts\\python.exe scripts\\fire_push_events.py --types withdraw_failed --count 3
   .venv\\Scripts\\python.exe scripts\\fire_push_events.py --types all --count 2
+  .venv\\Scripts\\python.exe scripts\\fire_push_events.py --phone 22222222222 --token ABC123 --count 1
+
+`--token` 是可选、不区分大小写的 token 子串过滤器;会推送给该账号下所有 token 包含该
+子串的设备。没有匹配设备时脚本直接停止,不会生成站内消息。
 
 推送成败看输出里的 `shagua.vendor_push` 日志(push sent / push failed);2 台设备则每条各推 2 次。
 凭据缺失或 token 失效时 notification_events 只记日志、不抛错(站内消息仍会落库)。
@@ -58,33 +62,52 @@ _FAIL_REASONS = [
 ]
 
 
-def _fire_one(db, uid: int, type_key: str, i: int) -> None:
+def _fire_one(
+    db,
+    uid: int,
+    type_key: str,
+    i: int,
+    *,
+    push_token_contains: str | None = None,
+) -> None:
     """构造一条该类型的瞬态业务对象(不落业务表,只为给 notify 函数读字段),触发真实推送。"""
     if type_key == "withdraw_success":
         order = WithdrawOrder(user_id=uid, out_bill_no=uuid.uuid4().hex, amount_cents=50, source="coin_cash")
-        notification_events.notify_withdraw_success(db, order)
+        notification_events.notify_withdraw_success(
+            db, order, push_token_contains=push_token_contains
+        )
     elif type_key == "withdraw_failed":
         order = WithdrawOrder(
             user_id=uid, out_bill_no=uuid.uuid4().hex, amount_cents=350, source="coin_cash",
             fail_reason=random.choice(_FAIL_REASONS),
         )
-        notification_events.notify_withdraw_failed(db, order)
+        notification_events.notify_withdraw_failed(
+            db, order, push_token_contains=push_token_contains
+        )
     elif type_key == "invite_order_reward":
         # 假被邀请人 id(> 真实用户范围,避重):昵称回退「好友」。真实昵称请走 API 流程(见文末说明)。
         fake_invitee = random.randint(900000, 999999)
         notification_events.notify_invite_order_reward(
-            db, inviter_user_id=uid, invitee_user_id=fake_invitee, cash_cents=INVITE_COMPARE_REWARD_CENTS
+            db,
+            inviter_user_id=uid,
+            invitee_user_id=fake_invitee,
+            cash_cents=INVITE_COMPARE_REWARD_CENTS,
+            push_token_contains=push_token_contains,
         )
     elif type_key == "feedback_reward":
         fb = Feedback(user_id=uid, content="(直发)", contact="", status="adopted",
                       reward_coins=300, admin_reply="感谢反馈,您说的问题已修复上线,金币请查收～")
         fb.id = random.randint(900000, 999999)
-        notification_events.notify_feedback_reward(db, fb)
+        notification_events.notify_feedback_reward(
+            db, fb, push_token_contains=push_token_contains
+        )
     elif type_key == "feedback_reply":
         fb = Feedback(user_id=uid, content="(直发)", contact="", status="rejected",
                       admin_reply="您的建议我们记录啦,会在后续版本评估~")
         fb.id = random.randint(900000, 999999)
-        notification_events.notify_feedback_reply(db, fb)
+        notification_events.notify_feedback_reply(
+            db, fb, push_token_contains=push_token_contains
+        )
     elif type_key == "report_approved":
         rep = PriceReport(
             user_id=uid, reported_platform_id="jd", reported_platform_name="京东外卖",
@@ -92,7 +115,9 @@ def _fire_one(db, uid: int, type_key: str, i: int) -> None:
             reward_coins=PRICE_REPORT_REWARD_COINS, store_name=f"测试火锅店{i:02d}",
         )
         rep.id = random.randint(900000, 999999)
-        notification_events.notify_report_approved(db, rep)
+        notification_events.notify_report_approved(
+            db, rep, push_token_contains=push_token_contains
+        )
     else:
         raise SystemExit(f"未知类型: {type_key}(可选: {', '.join(ALL_TYPES)})")
 
@@ -100,6 +125,11 @@ def _fire_one(db, uid: int, type_key: str, i: int) -> None:
 def main() -> None:
     parser = argparse.ArgumentParser(description="直接触发消息通知中心真实推送(后台驱动不了的事件用)")
     parser.add_argument("--phone", default=DEFAULT_PHONE, help=f"目标用户手机号(默认 {DEFAULT_PHONE})")
+    parser.add_argument(
+        "--token",
+        default="",
+        help="可选的厂商 token 子串(不区分大小写);仅推送到该账号下所有匹配设备",
+    )
     parser.add_argument("--count", type=int, default=10, help="每类发多少条(默认 10)")
     parser.add_argument(
         "--types", default=",".join(DEFAULT_TYPES),
@@ -122,14 +152,40 @@ def main() -> None:
         uid = user.id
 
         targets = device_repo.list_push_targets(db, user_id=uid)
-        print(f"目标用户 {args.phone}(id={uid});已注册推送设备 {len(targets)} 台:"
-              f"{[t.push_vendor for t in targets] or '无(手机收不到!先在 App 上报 push token)'}")
+        token_filter = args.token.strip()
+        normalized_filter = token_filter.casefold()
+        matched_targets = [
+            target
+            for target in targets
+            if not token_filter
+            or normalized_filter in (target.push_token or "").casefold()
+        ]
+        if token_filter and not matched_targets:
+            print(
+                f"❌ 用户 {args.phone} 的 {len(targets)} 台有效推送设备中，"
+                f"没有 token 包含 {token_filter!r}；未生成站内消息，也未发送推送。"
+            )
+            return
+
+        print(
+            f"目标用户 {args.phone}(id={uid});有效推送设备 {len(targets)} 台;"
+            f"本次匹配 {len(matched_targets)} 台:"
+            f"{[target.push_vendor for target in matched_targets] or '无(手机收不到!先在 App 上报 push token)'}"
+        )
+        if token_filter:
+            print(f"token 子串过滤(不区分大小写): {token_filter!r}")
         print(f"即将触发:{types},每类 {args.count} 条 → 共 {len(types) * args.count} 条\n")
 
         for t in types:
             print(f"── {t} ×{args.count} " + "─" * 30)
             for i in range(1, args.count + 1):
-                _fire_one(db, uid, t, i)
+                _fire_one(
+                    db,
+                    uid,
+                    t,
+                    i,
+                    push_token_contains=token_filter or None,
+                )
 
         print(f"\n✅ 已触发完。站内消息已落 notification 表(用 {args.phone} 登录 App 可在消息中心看到);"
               "\n   手机推送成败见上方 `shagua.vendor_push` 日志(push sent=成功 / push failed=失败)。")

scripts/test_push_invite_order_reward.py

@@ -15,6 +15,10 @@ services/notification_events.notify_invite_order_reward —— 与生产同一
   .venv\\Scripts\\python.exe scripts\\test_push_invite_order_reward.py                        # 随机金额发 1 条
   .venv\\Scripts\\python.exe scripts\\test_push_invite_order_reward.py --cents 200            # 固定 2.00 元
   .venv\\Scripts\\python.exe scripts\\test_push_invite_order_reward.py --invitee-phone 12000000001  # 真实昵称
+  .venv\\Scripts\\python.exe scripts\\test_push_invite_order_reward.py --phone 22222222222 --token gkha
+
+`--token` 是可选、不区分大小写的 token 子串过滤器；会推送给该账号下所有 token 包含该
+子串的设备。没有匹配设备时脚本直接停止，不会生成站内消息。
 
 结果判读(看输出日志):
   push sent   = 厂商接口受理成功,手机应弹「好友下单奖励到账」通知
@@ -62,6 +66,11 @@ def _notif_count(db, uid: int) -> int:
 def main() -> None:
     parser = argparse.ArgumentParser(description="#12 好友下单到账 推送联调(每次 1 条,金额默认随机)")
     parser.add_argument("--phone", default="11111111111", help="邀请人(收通知方)手机号,默认 11111111111")
+    parser.add_argument(
+        "--token",
+        default="",
+        help="可选的厂商 token 子串(不区分大小写);仅推送到该账号下所有匹配设备",
+    )
     parser.add_argument("--cents", type=int, default=None,
                         help="奖励金额,单位分(默认随机 1~9999;线上真实值 200)")
     parser.add_argument("--invitee-phone", default="",
@@ -87,13 +96,37 @@ def main() -> None:
             invitee_id = random.randint(900000, 999999)  # 假 id,昵称兜底「好友」,永不去重
 
         targets = device_repo.list_push_targets(db, user_id=user.id)
-        vendors = [t.push_vendor for t in targets]
-        print(f"邀请人 {args.phone}(id={user.id});推送设备 {len(targets)} 台:{vendors or '无 ← 手机收不到!先在 App 上报 push token'}")
+        token_filter = args.token.strip()
+        normalized_filter = token_filter.casefold()
+        matched_targets = [
+            target
+            for target in targets
+            if not token_filter
+            or normalized_filter in (target.push_token or "").casefold()
+        ]
+        if token_filter and not matched_targets:
+            print(
+                f"❌ 用户 {args.phone} 的 {len(targets)} 台有效推送设备中，"
+                f"没有 token 包含 {token_filter!r}；未生成站内消息，也未发送推送。"
+            )
+            return
+
+        print(
+            f"邀请人 {args.phone}(id={user.id});有效推送设备 {len(targets)} 台;"
+            f"本次匹配 {len(matched_targets)} 台:"
+            f"{[target.push_vendor for target in matched_targets] or '无 ← 手机收不到!先在 App 上报 push token'}"
+        )
+        if token_filter:
+            print(f"token 子串过滤(不区分大小写): {token_filter!r}")
 
         before = _notif_count(db, user.id)
         print(f"→ 本次奖励 【{cents / 100:.2f} 元】(invitee_user_id={invitee_id}),手机上按金额认领这条通知")
         notification_events.notify_invite_order_reward(
-            db, inviter_user_id=user.id, invitee_user_id=invitee_id, cash_cents=cents
+            db,
+            inviter_user_id=user.id,
+            invitee_user_id=invitee_id,
+            cash_cents=cents,
+            push_token_contains=token_filter or None,
         )
 
         created = _notif_count(db, user.id) - before

tests/test_ad_reward_coin_floor.py

@@ -0,0 +1,59 @@
+"""发奖公式下限:有真实正 eCPM 时最低发 1 金币,eCPM 缺失/为 0 仍发 0。
+
+针对「eCPM 过低时公式四舍五入成 0 金币 → 被记 too_short 不发」的问题:只要广告有真实
+正 eCPM,单份金币至少 1(不再 0);但 eCPM 缺失/为 0/非法(没有真实广告价值)仍发 0,
+不凭空铸币、不破坏 ecpm_missing 语义。公式是发奖与后台审计对账的唯一口径,改这一处两边同源。
+"""
+from __future__ import annotations
+
+from app.core.rewards import calculate_ad_reward_coin
+
+
+def test_low_positive_ecpm_floors_to_one_coin() -> None:
+    """真实但极低的 eCPM 原本四舍五入成 0 金币,现在兜底为 1(线上 record 4667 的 43 分场景)。"""
+    # 43 分 = ¥0.43 CPM,因子1=0.1,重度用户 LT 第 69 条=1.0 → 0.43/1000×0.1×1.0×10000=0.43 → 旧口径 round=0
+    assert calculate_ad_reward_coin("43", 69) == 1
+    # 更低的 5 分同理:算出来 <0.5,旧口径也是 0
+    assert calculate_ad_reward_coin("5", 11) == 1
+
+
+def test_zero_or_missing_ecpm_stays_zero() -> None:
+    """eCPM 缺失 / 为 0 / 非法(没有真实广告价值)不兜底,仍发 0。"""
+    assert calculate_ad_reward_coin("0", 1) == 0
+    assert calculate_ad_reward_coin(None, 1) == 0
+    assert calculate_ad_reward_coin("", 1) == 0
+    assert calculate_ad_reward_coin("abc", 1) == 0
+
+
+def test_normal_ecpm_value_unchanged() -> None:
+    """正常量级 eCPM 本就 ≥1,下限不改变其取值(回归保护)。"""
+    # 200 分 = ¥2 CPM,因子1=0.1,LT 第 1 条=2.0 → 2/1000×0.1×2.0×10000 = 4
+    assert calculate_ad_reward_coin("200", 1) == 4
+
+
+def test_feed_reward_low_ecpm_grants_one_coin_instead_of_too_short() -> None:
+    """端到端:低 eCPM 的信息流广告看满一份(≥10s),原本单份金币算成 0 被记 too_short,
+    现在兜底发 1 金币、状态 granted(线上 record 4667 那类「看了却零发」的修复)。"""
+    from app.db.session import SessionLocal
+    from app.models.user import User
+    from app.repositories.ad_feed_reward import grant_feed_reward
+
+    db = SessionLocal()
+    try:
+        user = User(phone="19900000043", username="feedfloor43", register_channel="sms")
+        db.add(user)
+        db.commit()
+        db.refresh(user)
+
+        rec = grant_feed_reward(
+            db, user.id,
+            client_event_id="feed-floor-low-ecpm-0001",
+            ecpm="2",                 # ¥0.02 CPM,单份收益四舍五入 <0.5 → 旧口径算成 0
+            duration_seconds=15,      # ≥10s → 看满一份,不走「时长不足」的 else 分支
+            ad_type="draw",
+            feed_scene="comparison",
+        )
+        assert rec.status == "granted"
+        assert rec.coin == 1
+    finally:
+        db.close()

tests/test_admin_config.py

@@ -68,12 +68,12 @@ def test_list_config(admin_client: TestClient, token: str) -> None:
     r = admin_client.get("/admin/api/config", headers=_auth(token))
     assert r.status_code == 200, r.text
     items = {i["key"]: i for i in r.json()}
-    # 非 hidden 项照常返回;看广告组保留可见的:每日上限 / 单次金币上限 / 关闭后冷却。
-    assert "signin_rewards" in items and "ad_daily_limit" in items and "ad_cooldown_sec" in items
-    # hidden 项(任务/里程碑、首页轮播数据源、看广告组的单次金币/每轮次数/信息流广告开关)不在配置页返回。
+    # 非 hidden 项照常返回；广告次数上限迁到「白名单」统一配置。
+    assert "signin_rewards" in items and "ad_cooldown_sec" in items
+    # hidden 项(任务/里程碑、首页轮播数据源、广告次数/单次金币/每轮次数/信息流广告开关)不在配置页返回。
     for hidden_key in (
         "task_rewards", "record_milestones", "marquee_feed_mode",
-        "ad_reward_coin", "ad_round_count", "comparing_ad_enabled",
+        "ad_daily_limit", "ad_reward_coin", "ad_round_count", "comparing_ad_enabled",
     ):
         assert hidden_key not in items, f"{hidden_key} 应被 hidden 过滤"
     assert items["signin_rewards"]["value"] == [
@@ -124,6 +124,50 @@ def test_update_ad_limit_takes_effect(admin_client: TestClient, token: str) -> N
         db.close()
 
 
+def test_list_config_reads_limit_values_from_global_bundle(
+    admin_client: TestClient,
+    token: str,
+) -> None:
+    changed = admin_client.patch(
+        "/admin/api/config/ad_cooldown_sec",
+        json={"value": 17},
+        headers=_auth(token),
+    )
+    assert changed.status_code == 200, changed.text
+
+    items = {
+        item["key"]: item
+        for item in admin_client.get(
+            "/admin/api/config",
+            headers=_auth(token),
+        ).json()
+    }
+    assert items["ad_cooldown_sec"]["value"] == 17
+    assert items["ad_cooldown_sec"]["overridden"] is True
+
+
+@pytest.mark.parametrize(
+    ("key", "value"),
+    [
+        ("ad_daily_limit", 0),
+        ("ad_daily_limit", 100_001),
+        ("ad_cooldown_sec", 86_401),
+    ],
+)
+def test_update_limit_config_rejects_out_of_range_values(
+    admin_client: TestClient,
+    token: str,
+    key: str,
+    value: int,
+) -> None:
+    response = admin_client.patch(
+        f"/admin/api/config/{key}",
+        json={"value": value},
+        headers=_auth(token),
+    )
+    assert response.status_code == 400, response.text
+
+
 def test_update_bool_config(admin_client: TestClient, token: str) -> None:
     # 提现自动对账开关默认 True
     items = {

tests/test_admin_read.py

@@ -219,6 +219,115 @@ def test_user_reward_stats_can_scope_withdrawals_by_account(
     assert invite.json()["cash_balance_cents"] == 456
 
 
+def test_user_reward_stats_draw_ecpm_uses_all_filtered_impressions(
+    admin_client: TestClient, admin_token: str
+) -> None:
+    """Draw 平均 eCPM 应与收益报表一致，不能只平均成功发奖记录。"""
+    from app.models.ad_ecpm import AdEcpmRecord
+    from app.models.ad_feed_reward import AdFeedRewardRecord
+
+    uid = _seed_user_with_data("13800000024")
+    created_at = datetime(2038, 1, 15, 4, tzinfo=UTC)
+    db = SessionLocal()
+    try:
+        db.add_all(
+            [
+                AdFeedRewardRecord(
+                    client_event_id="reward-stats-granted-high",
+                    ad_session_id="reward-stats-granted-high",
+                    user_id=uid,
+                    reward_date="2038-01-15",
+                    duration_seconds=10,
+                    unit_count=1,
+                    ecpm_raw="9000",
+                    ad_type="draw",
+                    feed_scene="coupon",
+                    app_env="prod",
+                    our_code_id="104098712",
+                    coin=9,
+                    status="granted",
+                    created_at=created_at,
+                ),
+                AdEcpmRecord(
+                    user_id=uid,
+                    ad_type="draw",
+                    feed_scene="coupon",
+                    ad_session_id="reward-stats-impression-low",
+                    app_env="prod",
+                    our_code_id="104098712",
+                    ecpm_raw="1000",
+                    report_date="2038-01-15",
+                    created_at=created_at,
+                ),
+                AdEcpmRecord(
+                    user_id=uid,
+                    ad_type="feed",
+                    feed_scene="coupon",
+                    ad_session_id="reward-stats-impression-mid",
+                    app_env="prod",
+                    our_code_id="104098712",
+                    ecpm_raw="3000",
+                    report_date="2038-01-15",
+                    created_at=created_at,
+                ),
+                # 同用户但不同场景/环境/非业务代码位，均不应进入本次详情筛选。
+                AdEcpmRecord(
+                    user_id=uid,
+                    ad_type="draw",
+                    feed_scene="comparison",
+                    ad_session_id="reward-stats-other-scene",
+                    app_env="prod",
+                    our_code_id="104098712",
+                    ecpm_raw="7000",
+                    report_date="2038-01-15",
+                    created_at=created_at,
+                ),
+                AdEcpmRecord(
+                    user_id=uid,
+                    ad_type="draw",
+                    feed_scene="coupon",
+                    ad_session_id="reward-stats-test-env",
+                    app_env="test",
+                    our_code_id="104127529",
+                    ecpm_raw="8000",
+                    report_date="2038-01-15",
+                    created_at=created_at,
+                ),
+                AdEcpmRecord(
+                    user_id=uid,
+                    ad_type="draw",
+                    feed_scene="coupon",
+                    ad_session_id="reward-stats-non-business",
+                    app_env="prod",
+                    our_code_id="demo-slot",
+                    ecpm_raw="9000",
+                    report_date="2038-01-15",
+                    created_at=created_at,
+                ),
+            ]
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    response = admin_client.get(
+        f"/admin/api/users/{uid}/reward-stats",
+        params={
+            "date_from": "2038-01-15T00:00:00Z",
+            "date_to": "2038-01-15T23:59:59Z",
+            "app_env": "prod",
+            "revenue_scope": "business",
+            "feed_scene": "coupon",
+        },
+        headers=_auth(admin_token),
+    )
+    assert response.status_code == 200, response.text
+    data = response.json()
+    assert data["feed_count"] == 1
+    # 全部真实展示 (1000 + 3000) / 2；不能返回成功发奖记录的 9000。
+    assert data["feed_avg_ecpm"] == 2000.0
+
+
 def test_user_coin_record_sort_accepts_mixed_timezone_datetimes() -> None:
     """线上 PostgreSQL 返回 aware，SQLite/历史转换可能返回 naive，二者必须可混排。"""
     naive = datetime(2038, 1, 1, 8, 0)

tests/test_admin_roles.py

@@ -73,6 +73,7 @@ def test_monitoring_audit_catalog_and_api_permissions(
         "analytics-health",
         "event-logs",
         "audit-logs",
+        "limit-whitelist",
     ]
 
     # 运营默认可查风控和设备存活，不能绕过导航直调技术/审计接口。
@@ -82,6 +83,9 @@ def test_monitoring_audit_catalog_and_api_permissions(
     assert admin_client.get(
         "/admin/api/device-liveness/stats", headers=_auth(operator_token)
     ).status_code == 200
+    assert admin_client.get(
+        "/admin/api/limit-whitelist/rules", headers=_auth(operator_token)
+    ).status_code == 200
     for path in (
         "/admin/api/analytics-health/overview?date_from=2026-07-01T00:00:00Z&date_to=2026-07-02T00:00:00Z",
         "/admin/api/event-logs",
@@ -89,13 +93,14 @@ def test_monitoring_audit_catalog_and_api_permissions(
     ):
         assert admin_client.get(path, headers=_auth(operator_token)).status_code == 403
 
-    # 技术角色默认拥有监控审计组全部五项权限。
+    # 技术角色默认拥有监控审计组全部六项权限。
     for path in (
         "/admin/api/risk-monitor/summary",
         "/admin/api/device-liveness/stats",
         "/admin/api/analytics-health/overview?date_from=2026-07-01T00:00:00Z&date_to=2026-07-02T00:00:00Z",
         "/admin/api/event-logs",
         "/admin/api/audit-logs",
+        "/admin/api/limit-whitelist/rules",
     ):
         assert admin_client.get(path, headers=_auth(tech_token)).status_code == 200
 
@@ -199,7 +204,7 @@ def test_builtin_roles_labels_and_pages(admin_client, super_token) -> None:
     }
     assert set(roles["tech"]["pages"]) == {
         "dashboard", "risk-monitor", "device-liveness", "analytics-health", "config",
-        "ad-revenue", "huawei-review", "event-logs", "audit-logs",
+        "ad-revenue", "huawei-review", "event-logs", "audit-logs", "limit-whitelist",
     }
 
 

tests/test_auth.py

@@ -68,6 +68,29 @@ def test_sms_login_and_me_flow(client) -> None:
     assert r.json()["ok"] is True
 
 
+def test_sms_send_records_provider_and_fallback_in_details(client, monkeypatch) -> None:
+    """批次3(Task 6):sms_send 成功后风控流水 details 带 provider/fallback。
+    mock 默认走极光主(SMS_FALLBACK_PROVIDER 空)→ provider=jiguang、fallback=False。
+    (回归:修复前 send_code 返回 SendResult 被塞进 cooldown_sec:int → 500,此处 200 断言即挡住。)"""
+    from app.repositories import risk as risk_repo
+
+    captured: dict = {}
+    orig = risk_repo.record_behavior_event
+
+    def _spy(db, **kw):
+        if kw.get("event_type") == risk_repo.EVENT_SMS_SEND and kw.get("outcome") == "success":
+            captured.update(kw.get("details") or {})
+        return orig(db, **kw)
+
+    monkeypatch.setattr(risk_repo, "record_behavior_event", _spy)
+
+    r = client.post("/api/v1/auth/sms/send",
+                    json={"phone": "13800009999", "device_id": "dev-provider-test"})
+    assert r.status_code == 200, r.text
+    assert captured.get("provider") == "jiguang"
+    assert captured.get("fallback") is False
+
+
 def test_sms_send_too_frequent(client) -> None:
     phone = "13900139000"
     assert client.post("/api/v1/auth/sms/send", json={"phone": phone}).status_code == 200

tests/test_compare_harvest.py

@@ -111,6 +111,116 @@ def test_harvest_done_derives_and_newly_success_once(client) -> None:
         assert newly2 is False
 
 
+def test_harvest_done_below_minimum_counts_as_completed_success(client) -> None:
+    """未达起送是可信业务结论:主状态/完成奖励归 success,细分结局仍留在 raw_payload。"""
+    tid = _tid()
+    done_below_minimum = {
+        "record_status": "below_minimum",
+        "comparison_results": [
+            {
+                "platform_id": "meituan",
+                "platform_name": "美团",
+                "package": "com.sankuai.meituan",
+                "price": 59.0,
+                "is_source": True,
+                "rank": 1,
+                "store_name": "测试店",
+                "items": [{"name": "红乌苏", "qty": 1}],
+            },
+        ],
+        "platform_results": {
+            "meituan": {"is_source": True, "status": "source", "price": 59.0},
+            "taobao_flash": {
+                "is_source": False,
+                "status": "below_minimum",
+                "reason": "购物车未达起送门槛（差 ¥25.2）",
+            },
+        },
+        "information": "淘宝未达起送门槛，可加菜凑单后下单",
+    }
+    with SessionLocal() as db:
+        crud.harvest_running(db, trace_id=tid, user_id=None)
+        rec, newly = crud.harvest_done(
+            db, trace_id=tid, user_id=None, done_params=done_below_minimum
+        )
+
+        assert rec.status == "success"
+        assert newly is True
+        assert rec.fail_reason is None
+        assert rec.raw_payload["record_status"] == "below_minimum"
+        assert (
+            rec.raw_payload["platform_results"]["taobao_flash"]["status"]
+            == "below_minimum"
+        )
+
+
+def test_harvest_done_platforms_below_minimum_counts_as_success(client) -> None:
+    """新 platforms 单源派生路径也必须执行同一 below_minimum → success 归一化。"""
+    tid = _tid()
+    with SessionLocal() as db:
+        rec, newly = crud.harvest_done(
+            db,
+            trace_id=tid,
+            user_id=None,
+            done_params={
+                "record_status": "below_minimum",
+                "platforms": [
+                    {
+                        "role": "source",
+                        "platform_id": "meituan",
+                        "platform_name": "美团",
+                        "price": 59.0,
+                        "store_name": "测试店",
+                        "items": [{"name": "红乌苏", "qty": 1}],
+                    },
+                    {
+                        "role": "target",
+                        "platform_id": "taobao_flash",
+                        "platform_name": "淘宝",
+                        "status": "below_minimum",
+                        "price": None,
+                    },
+                ],
+                "platform_results": {
+                    "taobao_flash": {
+                        "is_source": False,
+                        "status": "below_minimum",
+                        "reason": "购物车未达起送门槛",
+                    }
+                },
+            },
+        )
+
+        assert rec.status == "success"
+        assert newly is True
+        assert rec.raw_payload["record_status"] == "below_minimum"
+
+
+def test_legacy_upsert_below_minimum_counts_as_success(client) -> None:
+    """灰度期客户端直报路径无论走 status 还是 record_status 都不能落第四种主状态。"""
+    tid = _tid()
+    payload = ComparisonRecordIn(
+        trace_id=tid,
+        business_type="food",
+        status="below_minimum",
+        comparison_results=[],
+        platform_results={
+            "taobao_flash": {
+                "is_source": False,
+                "status": "below_minimum",
+                "reason": "购物车未达起送门槛",
+            }
+        },
+        information="淘宝未达起送门槛，可加菜凑单后下单",
+    )
+    with SessionLocal() as db:
+        rec = crud.upsert_record(db, user_id=987654, payload=payload)
+
+        assert rec.status == "success"
+        assert rec.fail_reason is None
+        assert rec.raw_payload["status"] == "below_minimum"
+
+
 def test_harvest_done_failed_derives_fail_reason(client) -> None:
     """failed 记录:记录级 information 笼统,但 fail_reason 从 platform_results 救出具体原因
     (id 3030 型:美团系统失败 + 京东 items_not_found → 展示京东那条)。"""
@@ -140,6 +250,75 @@ def test_harvest_done_failed_derives_fail_reason(client) -> None:
         assert rec.information == "比价过程出错，请稍后重试"  # 原文案仍留存
 
 
+def _done_params_platforms_no_isbest() -> dict:
+    """id 3304 型:done 帧 platforms 全平台「相似替换/仅供参考」(has_dish_diff),pricebot
+    一个 is_best 都没标,但有有价目标(美团 57.8 < 源淘宝闪购 60.8)。"""
+    return {
+        "platforms": [
+            {"role": "source", "platform_id": "eleme", "platform_name": "淘宝闪购",
+             "package": "me.ele", "price": 60.8, "is_best": False, "has_dish_diff": False,
+             "store_name": "窑鸡王", "items": [{"name": "招牌窑鸡 整只-香辣", "qty": 1}]},
+            {"role": "target", "platform_id": "meituan_waimai", "platform_name": "美团外卖",
+             "package": "com.sankuai.meituan.takeoutnew", "price": 57.8, "is_best": False,
+             "has_dish_diff": True},
+            {"role": "target", "platform_id": "jd_waimai_standalone", "platform_name": "京东外卖",
+             "package": "com.jd.waimai", "price": 74.9, "is_best": False, "has_dish_diff": True},
+        ],
+        "information": "美团更便宜（含相似商品替换）",
+    }
+
+
+def test_harvest_done_platforms_no_isbest_falls_back_to_cheapest_target(client) -> None:
+    """回归(id 3304):platforms 全无 is_best(全平台相似替换)但有有价目标 → best 应兜底取
+    有价目标里最低那家,不能让 best_*/saved 整条落 NULL(否则首页价 0.00 / 记录页无最低红框 /
+    省额丢失)。"""
+    tid = _tid()
+    with SessionLocal() as db:
+        crud.harvest_running(db, trace_id=tid, user_id=None)
+        rec, _ = crud.harvest_done(db, trace_id=tid, user_id=None,
+                                   done_params=_done_params_platforms_no_isbest())
+        assert rec.status == "success"
+        assert rec.best_platform_id == "meituan_waimai"   # 有价目标里最低
+        assert rec.best_price_cents == 5780               # 57.8 元
+        assert rec.source_price_cents == 6080             # 源淘宝闪购 60.8
+        assert rec.saved_amount_cents == 300              # 60.8 - 57.8
+        assert rec.is_source_best is False                # 兜底选的是目标,非源
+
+
+def _done_params_no_isbest_source_cheapest() -> dict:
+    """无 is_best 且源本身最便宜:源淘宝闪购 50.0 < 全部 dish-diff 目标(美团 57.8 / 京东 74.9)。
+    此时不能强选更贵的目标当 best(否则 saved 变负、污染「累计发现可省」),应回落源。"""
+    return {
+        "platforms": [
+            {"role": "source", "platform_id": "eleme", "platform_name": "淘宝闪购",
+             "package": "me.ele", "price": 50.0, "is_best": False, "has_dish_diff": False,
+             "store_name": "窑鸡王", "items": [{"name": "招牌窑鸡 整只-香辣", "qty": 1}]},
+            {"role": "target", "platform_id": "meituan_waimai", "platform_name": "美团外卖",
+             "package": "com.sankuai.meituan.takeoutnew", "price": 57.8, "is_best": False,
+             "has_dish_diff": True},
+            {"role": "target", "platform_id": "jd_waimai_standalone", "platform_name": "京东外卖",
+             "package": "com.jd.waimai", "price": 74.9, "is_best": False, "has_dish_diff": True},
+        ],
+        "information": "源平台已是最低（其余为相似替换）",
+    }
+
+
+def test_harvest_done_no_isbest_source_cheapest_falls_back_to_source(client) -> None:
+    """回归:无 is_best 且源最便宜 → best 回落源(与 _derive「全目标缺菜回落源、不虚报省」同语义),
+    saved=0、is_source_best=True,绝不因强选更贵目标而让 saved 变负、倒扣「累计发现可省」。"""
+    tid = _tid()
+    with SessionLocal() as db:
+        crud.harvest_running(db, trace_id=tid, user_id=None)
+        rec, _ = crud.harvest_done(db, trace_id=tid, user_id=None,
+                                   done_params=_done_params_no_isbest_source_cheapest())
+        assert rec.status == "success"
+        assert rec.best_platform_id == "eleme"            # 回落到源(源最便宜)
+        assert rec.best_price_cents == 5000               # 源 50.0
+        assert rec.source_price_cents == 5000
+        assert rec.saved_amount_cents == 0                # 没省到,绝不为负
+        assert rec.is_source_best is True                 # 源就是最便宜
+
+
 def test_harvest_abort_cancels_running(client) -> None:
     tid = _tid()
     with SessionLocal() as db:

tests/test_comparison_below_minimum_migration.py

@@ -0,0 +1,81 @@
+"""Regression coverage for the below-minimum status data migration."""
+from __future__ import annotations
+
+import importlib.util
+from pathlib import Path
+
+import sqlalchemy as sa
+from alembic.migration import MigrationContext
+from alembic.operations import Operations
+
+
+def _migration_module():
+    path = (
+        Path(__file__).parents[1]
+        / "alembic"
+        / "versions"
+        / "comparison_below_minimum_as_success.py"
+    )
+    spec = importlib.util.spec_from_file_location(path.stem, path)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_below_minimum_migration_upgrade_and_downgrade() -> None:
+    engine = sa.create_engine("sqlite:///:memory:")
+    metadata = sa.MetaData()
+    records = sa.Table(
+        "comparison_record",
+        metadata,
+        sa.Column("id", sa.Integer, primary_key=True),
+        sa.Column("status", sa.String(16), nullable=False),
+        sa.Column("fail_reason", sa.String(256)),
+        sa.Column("raw_payload", sa.JSON, nullable=False),
+    )
+    metadata.create_all(engine)
+    module = _migration_module()
+
+    with engine.begin() as connection:
+        connection.execute(
+            records.insert(),
+            [
+                {
+                    "status": "below_minimum",
+                    "fail_reason": "旧失败原因",
+                    "raw_payload": {"record_status": "below_minimum"},
+                },
+                {
+                    "status": "success",
+                    "fail_reason": None,
+                    "raw_payload": {"record_status": "success"},
+                },
+                {
+                    "status": "failed",
+                    "fail_reason": "技术异常",
+                    "raw_payload": {"status": "failed"},
+                },
+            ],
+        )
+        module.op = Operations(MigrationContext.configure(connection))
+
+        module.upgrade()
+        upgraded = connection.execute(
+            sa.select(records.c.status, records.c.fail_reason).order_by(records.c.id)
+        ).all()
+        assert upgraded == [
+            ("success", None),
+            ("success", None),
+            ("failed", "技术异常"),
+        ]
+
+        module.downgrade()
+        downgraded = connection.execute(
+            sa.select(records.c.status, records.c.fail_reason).order_by(records.c.id)
+        ).all()
+        assert downgraded == [
+            ("below_minimum", None),
+            ("success", None),
+            ("failed", "技术异常"),
+        ]

tests/test_guide_video.py

@@ -201,7 +201,7 @@ def test_start_loses_seq_race_degrades_to_ad(client, guide_configured, monkeypat
     assert first["should_play"] is True and first["seq"] == 1
 
     # 本次请求读到的是过期计数 → 仍会算出 seq=1
-    monkeypatch.setattr(crud_guide, "used_plays", lambda db, user_id: 0)
+    monkeypatch.setattr(crud_guide, "used_plays", lambda db, user_id, scene="coupon": 0)
     with SessionLocal() as db:
         result = crud_guide.start_play(db, uid)
 

tests/test_push_center.py

@@ -7,7 +7,9 @@ mock/真发/变量覆盖/站内联动/设备反查/参数校验。厂商 HTTP
 from __future__ import annotations
 
 import json
+import logging
 
+import httpx
 import pytest
 from fastapi.testclient import TestClient
 
@@ -110,6 +112,99 @@ def test_huawei_non_success_code_raises(monkeypatch) -> None:
         vendor_push.send_notification("huawei", "bad-token", title="t", body="b")
 
 
+def test_vendor_http_logs_device_token_raw_response_and_elapsed(monkeypatch, caplog) -> None:
+    vendor_push._token_cache.clear()
+
+    def _fake_request(method, url, **kwargs):  # noqa: ANN001
+        if url == vendor_push.settings.HUAWEI_PUSH_TOKEN_ENDPOINT:
+            return _Resp({"access_token": "raw-access-token", "expires_in": 3600})
+        return _Resp({"code": "80000000", "msg": "Success", "requestId": "vendor-request-1"})
+
+    monkeypatch.setattr(vendor_push.settings, "HUAWEI_PUSH_APP_ID", "10086001")
+    monkeypatch.setattr(vendor_push.settings, "HUAWEI_PUSH_APP_SECRET", "raw-app-secret")
+    monkeypatch.setattr(vendor_push.httpx, "request", _fake_request)
+
+    with caplog.at_level(logging.INFO, logger=vendor_push.logger.name):
+        vendor_push.send_notification(
+            "huawei",
+            "raw-device-token",
+            title="需要进入日志的标题",
+            body="需要进入日志的正文",
+            extras={"type": "withdraw_success", "notificationId": "90001"},
+        )
+
+    messages = "\n".join(record.getMessage() for record in caplog.records)
+    assert "operation=authenticate" in messages
+    assert "operation=send_notification" in messages
+    assert "request=" in messages
+    assert "response=" in messages
+    assert "elapsed_ms=" in messages
+    assert "vendor-request-1" in messages
+    assert "withdraw_success" in messages
+    assert "raw-access-token" in messages
+    assert "raw-device-token" in messages
+    assert "需要进入日志的标题" in messages
+    assert "需要进入日志的正文" in messages
+    assert "<redacted" in messages
+    assert "raw-app-secret" not in messages
+    send_http_log = next(
+        record.getMessage()
+        for record in caplog.records
+        if "operation=send_notification" in record.getMessage()
+        and "vendor push http completed" in record.getMessage()
+    )
+    assert "raw-device-token" in send_http_log
+    assert "raw-access-token" not in send_http_log  # 请求头中的厂商鉴权 token 仍脱敏
+
+
+def test_vendor_http_network_error_logs_request_context(monkeypatch, caplog) -> None:
+    def _fake_request(*args, **kwargs):  # noqa: ANN002, ANN003
+        raise httpx.ConnectError("connection refused")
+
+    monkeypatch.setattr(vendor_push.settings, "XIAOMI_PUSH_APP_SECRET", "raw-xiaomi-secret")
+    monkeypatch.setattr(vendor_push.httpx, "request", _fake_request)
+
+    with (
+        caplog.at_level(logging.ERROR, logger=vendor_push.logger.name),
+        pytest.raises(vendor_push.VendorPushError, match="push http error"),
+    ):
+        vendor_push.send_notification(
+            "xiaomi",
+            "raw-xiaomi-token",
+            title="title",
+            body="body",
+            extras={"type": "push_test"},
+        )
+
+    messages = "\n".join(record.getMessage() for record in caplog.records)
+    assert "outcome=network_error" in messages
+    assert "response=<no_response>" in messages
+    assert "operation=send_notification" in messages
+    assert "elapsed_ms=" in messages
+    assert "raw-xiaomi-secret" not in messages
+    assert "raw-xiaomi-token" in messages
+
+
+@pytest.mark.parametrize(
+    ("request_kwargs", "device_token"),
+    [
+        ({"json": {"token": ["honor-device-token"]}}, "honor-device-token"),
+        ({"json": {"message": {"token": ["huawei-device-token"]}}}, "huawei-device-token"),
+        ({"json": {"regId": "vivo-device-token"}}, "vivo-device-token"),
+        ({"data": {"registration_id": "xiaomi-device-token"}}, "xiaomi-device-token"),
+        (
+            {"data": {"message": '{"target_value":"oppo-device-token"}'}},
+            "oppo-device-token",
+        ),
+    ],
+)
+def test_all_vendor_device_token_fields_remain_visible_in_request_summary(
+    request_kwargs, device_token
+) -> None:
+    summary = vendor_push._request_summary(request_kwargs)
+    assert device_token in summary
+
+
 def test_vendor_aliases_normalize() -> None:
     assert vendor_push.normalize_vendor("华为") == "huawei"
     assert vendor_push.normalize_vendor("HMS") == "huawei"

tests/test_risk_monitor.py

@@ -10,11 +10,7 @@ from fastapi.testclient import TestClient
 from app.admin.main import admin_app
 from app.admin.repositories import admin_user as admin_repo
 from app.api.v1 import compare as compare_api
-from app.core.config_schema import (
-    RISK_COMPARE_DAILY_THRESHOLD_KEY,
-    RISK_ONECLICK_DAILY_THRESHOLD_KEY,
-    RISK_SMS_HOURLY_THRESHOLD_KEY,
-)
+from app.core.config_schema import LIMIT_POLICY_GLOBAL_KEY
 from app.core.security import issue_token_pair
 from app.db.session import SessionLocal
 from app.models.app_config import AppConfig
@@ -28,9 +24,7 @@ from app.repositories import user as user_repo
 @pytest.fixture(autouse=True)
 def _reset_risk_rule_config():
     keys = (
-        RISK_SMS_HOURLY_THRESHOLD_KEY,
-        RISK_ONECLICK_DAILY_THRESHOLD_KEY,
-        RISK_COMPARE_DAILY_THRESHOLD_KEY,
+        LIMIT_POLICY_GLOBAL_KEY,
         risk_repo.RISK_RESET_BASELINES_KEY,
     )
     with SessionLocal() as db:
@@ -533,7 +527,7 @@ def test_admin_can_edit_rules_and_current_window_is_reconciled() -> None:
             "/admin/api/risk-monitor/rules",
             headers=headers,
             json={
-                "sms_hourly_threshold": 21,
+                "sms_hourly_threshold": 100_001,
                 "oneclick_daily_threshold": 20,
                 "compare_daily_threshold": 100,
             },

tests/test_sms_base.py

@@ -0,0 +1,26 @@
+"""SMS 共享基座类型 SendResult。"""
+from __future__ import annotations
+
+import dataclasses
+
+import pytest
+
+from app.integrations.sms.base import SendResult
+
+
+def test_send_result_fields():
+    r = SendResult(cooldown_sec=60, provider="jiguang", fallback=True)
+    assert r.cooldown_sec == 60
+    assert r.provider == "jiguang"
+    assert r.fallback is True
+
+
+def test_send_result_fallback_defaults_false():
+    r = SendResult(cooldown_sec=60, provider="jiguang")
+    assert r.fallback is False
+
+
+def test_send_result_is_frozen():
+    r = SendResult(cooldown_sec=60, provider="jiguang")
+    with pytest.raises(dataclasses.FrozenInstanceError):
+        r.provider = "chuanglan"

tests/test_sms_dispatch.py

@@ -17,11 +17,11 @@ def test_send_code_routes_by_provider_and_switches_per_call(monkeypatch) -> None
     monkeypatch.setattr(chuanglan, "send_code", lambda phone: (calls.append("chuanglan"), 60)[1])
 
     monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
-    assert sms.send_code("13800138000") == 60
+    assert sms.send_code("13800138000").cooldown_sec == 60
     monkeypatch.setattr(settings, "SMS_PROVIDER", "aliyun")
-    assert sms.send_code("13800138000") == 60
+    assert sms.send_code("13800138000").cooldown_sec == 60
     monkeypatch.setattr(settings, "SMS_PROVIDER", "chuanglan")
-    assert sms.send_code("13800138000") == 60
+    assert sms.send_code("13800138000").cooldown_sec == 60
 
     assert calls == ["jiguang", "aliyun", "chuanglan"]  # 每次按当前 provider 路由,运行时可切
 

tests/test_sms_fallback.py

@@ -0,0 +1,185 @@
+"""极光→创蓝 容灾 fallback（分派层 send_code / verify_code）。
+
+monkeypatch 各 provider 的 send_code/verify_code 为标记函数，不发真网络；
+断言：仅主 503 才转备；429/400 不转；verify try-both 只遍历启用的链。
+"""
+from __future__ import annotations
+
+import pytest
+
+from app.core.config import settings
+from app.integrations import sms
+from app.integrations.sms import chuanglan, jiguang
+from app.integrations.sms.base import SendResult, SmsError
+
+PHONE = "13800138000"
+
+
+def _raiser(exc: Exception):
+    """返回一个调用即 raise exc 的 send_code 替身（lambda 不能 raise）。"""
+    def _f(phone):
+        raise exc
+    return _f
+
+
+def test_fallback_provider_defaults_empty():
+    assert settings.SMS_FALLBACK_PROVIDER == ""
+
+
+def test_send_code_primary_success_returns_sendresult(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")
+    monkeypatch.setattr(jiguang, "send_code", lambda phone: 60)
+
+    r = sms.send_code(PHONE)
+    assert isinstance(r, SendResult)
+    assert r.cooldown_sec == 60
+    assert r.provider == "jiguang"
+    assert r.fallback is False
+
+
+def test_send_code_503_falls_back_to_configured_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        lambda phone: (calls.append("chuanglan"), 60)[1])
+
+    r = sms.send_code(PHONE)
+    assert r.provider == "chuanglan"
+    assert r.fallback is True
+    assert r.cooldown_sec == 60
+    assert calls == ["chuanglan"]
+
+
+@pytest.mark.parametrize("status", [429, 400])
+def test_send_code_non_503_does_not_fall_back(monkeypatch, status):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("x", status_code=status)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        lambda phone: (calls.append("chuanglan"), 60)[1])
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert ei.value.status_code == status
+    assert calls == []
+
+
+def test_send_code_503_without_backup_reraises(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert ei.value.status_code == 503
+
+
+def test_send_code_backup_also_fails_raises_backup_error(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code",
+                        _raiser(SmsError("极光不可用", status_code=503)))
+    monkeypatch.setattr(chuanglan, "send_code",
+                        _raiser(SmsError("创蓝也不可用", status_code=503)))
+
+    with pytest.raises(SmsError) as ei:
+        sms.send_code(PHONE)
+    assert "创蓝" in str(ei.value)
+
+
+def test_dynamic_cooldown_is_forwarded_to_primary_and_backup(monkeypatch):
+    calls = []
+
+    def primary(phone, *, cooldown_sec=None):
+        calls.append(("jiguang", cooldown_sec))
+        raise SmsError("极光不可用", status_code=503)
+
+    def backup(phone, *, cooldown_sec=None):
+        calls.append(("chuanglan", cooldown_sec))
+        return cooldown_sec
+
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "send_code", primary)
+    monkeypatch.setattr(chuanglan, "send_code", backup)
+
+    result = sms.send_code(PHONE, cooldown_sec=15)
+
+    assert result.cooldown_sec == 15
+    assert result.provider == "chuanglan"
+    assert result.fallback is True
+    assert calls == [("jiguang", 15), ("chuanglan", 15)]
+
+
+def test_verify_hits_primary_without_touching_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), True)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is True
+    assert calls == ["jiguang"]
+
+
+def test_verify_falls_through_to_backup(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), False)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is True
+    assert calls == ["jiguang", "chuanglan"]
+
+
+def test_verify_all_miss_returns_false(monkeypatch):
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code", lambda p, c: False)
+    monkeypatch.setattr(chuanglan, "verify_code", lambda p, c: False)
+
+    assert sms.verify_code(PHONE, "123456") is False
+
+
+def test_verify_backup_not_touched_when_fallback_off(monkeypatch):
+    calls = []
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "")
+    monkeypatch.setattr(jiguang, "verify_code",
+                        lambda p, c: (calls.append("jiguang"), False)[1])
+    monkeypatch.setattr(chuanglan, "verify_code",
+                        lambda p, c: (calls.append("chuanglan"), True)[1])
+
+    assert sms.verify_code(PHONE, "123456") is False
+    assert calls == ["jiguang"]
+
+
+def test_dynamic_verify_attempt_limit_is_forwarded_to_fallback_chain(monkeypatch):
+    calls = []
+
+    def verify(provider, result):
+        def _verify(phone, code, *, max_failed_attempts=None):
+            calls.append((provider, max_failed_attempts))
+            return result
+
+        return _verify
+
+    monkeypatch.setattr(settings, "SMS_PROVIDER", "jiguang")
+    monkeypatch.setattr(settings, "SMS_FALLBACK_PROVIDER", "chuanglan")
+    monkeypatch.setattr(jiguang, "verify_code", verify("jiguang", False))
+    monkeypatch.setattr(chuanglan, "verify_code", verify("chuanglan", True))
+
+    assert sms.verify_code(PHONE, "123456", max_failed_attempts=2) is True
+    assert calls == [("jiguang", 2), ("chuanglan", 2)]