fix(review): 优化复核逻辑及同步流程

- 调整已审核记录的识别条件，区分不同审核状态
- 移除撤销审核权限中的审核人限制，允许所有审核人撤销
- 修正 SQL 领取语句，避免错误筛选已审核记录的审核人字段
- 更新前端领取逻辑，避免新建/合并/跳过等记录被错误领取
- 增加多人审核首次启用时的数据库迁移说明
- 前端页面增加审核任务领取、数据同步及冲突处理的详细说明
- 添加单元测试对更新代码的覆盖和验证
- 新增 title-norm Python 包相关项目配置及说明文档

@@ -144,11 +144,19 @@ python -m pytest test_dedup.py::TestL2Benchmark::test_l2_recall_topk_efficiency_
 
 
 ### 7. 启动复核前端
 ### 7. 启动复核前端
 
 
+首次启用多人审核时，先让服务补齐领取和乐观锁字段（已有审核结果不会被修改）：
+
+```bash
+python serve_l2_dashboard.py --migrate-review-schema --host 127.0.0.1 --port 8765
+```
+
+后续正常启动：
+
 ```bash
 ```bash
 python serve_l2_dashboard.py --host 127.0.0.1 --port 8765
 python serve_l2_dashboard.py --host 127.0.0.1 --port 8765
 ```
 ```
 
 
-然后访问 http://127.0.0.1:8765 查看测试报告、比对歌词、人工标注。
+然后访问 http://127.0.0.1:8765，填写审核人姓名或工号后开始审核。打开 `pending/unsure` 的人工复核记录时会自动领取；`new/merge/skip/已审核` 记录仅供浏览，不会领取。领取冲突只会将当前记录设为只读，不会刷新列表或跳转。其他审核人的结果约 15 秒内轻量同步，统计信息每分钟刷新一次，页面处于后台时暂停轮询。
 
 
 ## 单元测试
 ## 单元测试
 
 

@@ -840,9 +840,10 @@
       if (!query || state.run?.type !== 'staging' || !query.staging_id) return true;
       if (!query || state.run?.type !== 'staging' || !query.staging_id) return true;
       const status = query.review_status || '';
       const status = query.review_status || '';
       const isPendingReview = rowDecision(query) === 'review' && ['pending', 'unsure', ''].includes(status);
       const isPendingReview = rowDecision(query) === 'review' && ['pending', 'unsure', ''].includes(status);
-      const isOwnReviewedRecord = allowReviewed && query.reviewed_by === state.reviewer;
+      const isReviewedRecord = allowReviewed &&
+        ['approved_import', 'rejected_duplicate', 'deleted'].includes(status);
       const isNonReviewRecord = allowNonReview && status === 'not_required';
       const isNonReviewRecord = allowNonReview && status === 'not_required';
-      const reviewable = isPendingReview || isOwnReviewedRecord || isNonReviewRecord;
+      const reviewable = isPendingReview || isReviewedRecord || isNonReviewRecord;
       // new / merge / skip 等记录只浏览，不应触发领取请求。
       // new / merge / skip 等记录只浏览，不应触发领取请求。
       if (!reviewable) return true;
       if (!reviewable) return true;
       try {
       try {
@@ -1103,8 +1104,7 @@
       const needsManualReview = rowDecision(query) === 'review';
       const needsManualReview = rowDecision(query) === 'review';
       const hasFinalReview = Boolean(query.review_status) &&
       const hasFinalReview = Boolean(query.review_status) &&
         !['pending', 'not_required', 'unsure'].includes(query.review_status);
         !['pending', 'not_required', 'unsure'].includes(query.review_status);
-      const canUndoReview = hasFinalReview && query.staging_status !== 'imported' &&
-        query.reviewed_by === state.reviewer;
+      const canUndoReview = hasFinalReview && query.staging_status !== 'imported';
       const canDeleteRecord = query.staging_status !== 'imported';
       const canDeleteRecord = query.staging_status !== 'imported';
 
 
       // Sticky area: 新入库歌词 + 召回候选（各含音频播放器）
       // Sticky area: 新入库歌词 + 召回候选（各含音频播放器）

+# title-norm
+
+`archive-bridge` 链路B `internal/subject/normalize.go` 中
+`title -> (title_norm, version_type)` 算法的 Python 移植包，供跨语言/离线批处理场景复用。
+
+移植范围（与 Go 源保持处理顺序与规则一致）：
+
+- `normalize_name`：姓名/文本归一化（全角转半角、繁转简、空白折叠、转小写）
+- `normalize_title`：歌名 -> `(title_norm, version)` 主算法
+  - 书名号剥离 → 破折号后缀版本感知剥离 → 多重括号循环剥离 → 无括号版本后缀词剥离
+    → `normalize_name` → 版本枚举归一
+- `strip_clip_markers`：削去片段标题（如抖音切片）的版本/切片标记后缀
+- `is_version_marker`：判定内容是否为版本标记（版本别名/固定词表/速度模式/黑名单）
+- 其余辅助：`split_names` / `normalize_name_set` / `normalize_set` /
+  `singer_identity_key` / `author_identity_key` / `org_identity_key` /
+  `platform_from_singer_table`
+
+## 安装
+
+```bash
+pip install -e .
+```
+
+繁简转换依赖 `opencc-python-reimplemented`（纯查表，非 AI）。若安装失败或加载出错，
+`normalize_name` 会跳过繁简转换，不影响其余归一逻辑（与 Go 源行为一致）。
+
+## 用法
+
+```python
+from title_norm import normalize_title
+
+title_norm, version = normalize_title("七里香（现场版）")
+# title_norm == "七里香", version == "live"
+```
+
+## 测试
+
+```bash
+pip install pytest
+pytest
+```
+
+测试用例（`tests/test_title_norm.py`）与 Go 源 `normalize_test.go` 中的用例逐一对齐。
+
+## 维护
+
+Go 源是权威实现；若 `internal/subject/normalize.go` 的规则（词表、正则、处理顺序）有变更，
+需同步更新本包，保持两侧行为一致。

+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "title-norm"
+version = "0.1.0"
+description = "歌名归一化与版本提取算法(archive-bridge internal/subject/normalize.go 的 Python 移植)"
+readme = "README.md"
+requires-python = ">=3.8"
+dependencies = [
+    "opencc-python-reimplemented>=0.1.4",
+]
+
+[tool.setuptools.packages.find]
+include = ["title_norm*"]

@@ -516,7 +516,7 @@ def _claim_staging_review(staging_id: int, reviewer: str, client_token: str) ->
               AND (
               AND (
                 biz_review_status IN ('pending', 'unsure')
                 biz_review_status IN ('pending', 'unsure')
                 OR biz_review_status = 'not_required'
                 OR biz_review_status = 'not_required'
-                OR (reviewed_by = %s AND biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted'))
+                OR biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted')
               )
               )
               AND (
               AND (
                 review_claim_expires_at IS NULL
                 review_claim_expires_at IS NULL
@@ -524,7 +524,7 @@ def _claim_staging_review(staging_id: int, reviewer: str, client_token: str) ->
                 OR review_claim_token = %s
                 OR review_claim_token = %s
               )
               )
             """,
             """,
-            (reviewer, client_token, REVIEW_CLAIM_TTL_SECONDS, staging_id, reviewer, client_token),
+            (reviewer, client_token, REVIEW_CLAIM_TTL_SECONDS, staging_id, client_token),
         )
         )
         if cursor.rowcount != 1:
         if cursor.rowcount != 1:
             current = _review_snapshot(cursor, staging_id)
             current = _review_snapshot(cursor, staging_id)

@@ -109,6 +109,8 @@ def test_claim_conflict_exposes_current_owner():
         with pytest.raises(dashboard.ReviewConflictError) as raised:
         with pytest.raises(dashboard.ReviewConflictError) as raised:
             dashboard._claim_staging_review(42, "alice", "alice-token")
             dashboard._claim_staging_review(42, "alice", "alice-token")
 
 
+    claim_sql, _params = cursor.executions[0]
+    assert "reviewed_by = %s" not in claim_sql
     assert raised.value.current_record["review_claimed_by"] == "bob"
     assert raised.value.current_record["review_claimed_by"] == "bob"
     assert conn.rolled_back
     assert conn.rolled_back
 
 
@@ -191,3 +193,4 @@ def test_dashboard_distinguishes_reviewed_from_submitted():
     assert row["staging_status"] == "imported"
     assert row["staging_status"] == "imported"
     assert '<option value="reviewed">已审核</option>' in html
     assert '<option value="reviewed">已审核</option>' in html
     assert '<option value="submitted">已提交</option>' in html
     assert '<option value="submitted">已提交</option>' in html
+    assert "hasFinalReview && query.staging_status !== 'imported'" in html