feat(review): 优化审核结果提交和重复数据合并逻辑

沈秋雨
Commit bbb6ce14 ... bbb6ce1447e193340855290021205a90c07891fe authored 2026-07-20 14:56:01 +0800 by 沈秋雨
Showing 3 changed files with 375 additions and 74 deletions
l2_review_dashboard.html
serve_l2_dashboard.py
test_serve_l2_dashboard.py
--- a/l2_review_dashboard.html
View file @bbb6ce1
+++ b/l2_review_dashboard.html
View file @bbb6ce1
@@ -646,6 +646,7 @@
        <option value="unsure">待确认</option>
        <option value="reviewed">已审核</option>
        <option value="submitted">已提交</option>
+        <option value="deleted">已删除</option>
        <option value="">全部</option>
        <option value="l1_hit">L1 命中</option>
        <option value="conflict">L1/L2 冲突</option>
@@ -661,7 +662,7 @@
        <option value="200">200</option>
      </select>
      <button id="exportBtn">导出标注</button>
-      <button id="importReviewedBtn">入库审核通过</button>
+      <button id="importReviewedBtn">提交审核结果</button>
    <div class="toolbar-right">
        <span>审核人：</span><span id="reviewerDisplay"></span>
        <button id="logoutBtn" class="secondary">登出</button>
@@ -1002,7 +1003,7 @@
      group.review_claim_expires_at = record.review_claim_expires_at || '';
      group.claimed_by_me = group.review_claimed_by === state.reviewer;
      group.reviewed = ['approved_import', 'rejected_duplicate', 'deleted'].includes(record.biz_review_status);
-      group.submitted = group.reviewed && record.staging_status === 'imported';
+      group.submitted = group.reviewed && ['imported', 'skipped'].includes(record.staging_status);
    }
    function currentQuery() {
@@ -1059,12 +1060,12 @@
        metric('待审核', row.pending_review_count || '0'),
        metric('已审核', row.reviewed_count || '0'),
        metric('已提交', row.submitted_count || '0'),
-        metric('确认入库', row.approved_count || '0'),
+        metric('确认不重复', row.approved_count || '0'),
-        metric('确认不入库', row.rejected_count || '0'),
+        metric('确认重复', row.rejected_count || '0'),
        metric('已删除', row.deleted_count || '0'),
-        metric('new', row.new_count || '-'),
+        metric('新歌', row.new_count || '-'),
-        metric('merge', row.duplicate_count || '0'),
+        metric('重复', row.duplicate_count || '0'),
-        metric('skip', row.skip_count || '0')
+        metric('已跳过', row.skip_count || '0')
      ].join('');
    }
@@ -1182,20 +1183,21 @@
      </div>`;
    }
-    async function updateStagingReview(query, decision, note) {
+    async function updateStagingReview(query, decision, note, candidateId = '') {
      return reviewPostJSON('/api/staging-review', {
        staging_id: String(query.staging_id),
        expected_version: Number(query.review_version || 0),
        client_token: state.clientToken,
        reviewer: state.reviewer,
        decision,
-        note
+        note,
+        candidate_id: candidateId
      });
    }
    const _DB_REVIEW_LABELS = {
-      approved_import: '确认入库（待入库）',
+      approved_import: '确认不重复（待入库）',
-      rejected_duplicate: '确认不入库',
+      rejected_duplicate: '确认重复（待合并作者）',
      unsure: '待确认',
      deleted: '已删除',
    };
@@ -1293,8 +1295,9 @@
      const needsManualReview = rowDecision(query) === 'review';
      const hasFinalReview = Boolean(query.review_status) &&
        !['pending', 'not_required', 'unsure'].includes(query.review_status);
-      const canUndoReview = hasFinalReview && query.staging_status !== 'imported';
+      const isSubmitted = ['imported', 'skipped'].includes(query.staging_status);
-      const canDeleteRecord = query.staging_status !== 'imported';
+      const canUndoReview = hasFinalReview && query.staging_status === 'staged';
+      const canDeleteRecord = !['imported', 'skipped'].includes(query.staging_status);
      // Sticky area: 新入库歌词 + 召回候选（各含音频播放器）
      els.stickyInfo.innerHTML = `
@@ -1339,7 +1342,7 @@
          <div class="section-head">
            <h2 class="section-title">人工标注</h2>
            ${hasFinalReview
-              ? `<span class="pill ${query.review_status === 'approved_import' ? 'new' : query.review_status === 'rejected_duplicate' ? 'duplicate' : query.review_status === 'deleted' ? 'duplicate' : ''}">${query.review_status === 'deleted' ? '已删除' : query.staging_status === 'imported' ? '已提交' : '已审核'}</span>`
+              ? `<span class="pill ${query.review_status === 'approved_import' ? 'new' : query.review_status === 'rejected_duplicate' ? 'duplicate' : query.review_status === 'deleted' ? 'duplicate' : ''}">${query.review_status === 'deleted' ? '已删除' : isSubmitted ? '已提交' : '已审核'}</span>`
              : ''}
          </div>
          <div class="section-body">
@@ -1347,8 +1350,10 @@
              <div class="review-bar" style="align-items:center">
                <span style="flex:1">
                  <b>${esc(query.review_status === 'approved_import' && query.staging_status === 'imported'
-                    ? '确认入库（已提交）'
+                    ? '确认不重复（已入库）'
-                    : (_DB_REVIEW_LABELS[query.review_status] || query.review_status))}</b>
+                    : query.review_status === 'rejected_duplicate' && query.staging_status === 'skipped'
+                      ? '确认重复（作者已合并）'
+                      : (_DB_REVIEW_LABELS[query.review_status] || query.review_status))}</b>
                  ${query.reviewed_by ? ` · 审核人 ${esc(query.reviewed_by)}` : ''}
                  ${query.reviewed_at ? ` · ${esc(query.reviewed_at)}` : ''}
                  ${query.review_note ? ` · 备注: ${esc(query.review_note)}` : ''}
@@ -1359,7 +1364,7 @@
              <div class="review-bar">
                ${['duplicate', 'not_duplicate', 'unsure'].map(value => `
                  <button class="review-choice ${selectedReview.final_decision === value ? 'active' : ''}" data-review="${value}">
-                    ${value === 'duplicate' ? '确认不入库' : value === 'not_duplicate' ? '确认入库' : '待确认'}
+                    ${value === 'duplicate' ? '确认重复' : value === 'not_duplicate' ? '确认不重复' : '待确认'}
                  </button>`).join('')}
                <input id="reviewNote" value="${esc(selectedReview.note || '')}" placeholder="人工备注">
                <button id="deleteReviewBtn" class="secondary" style="color:#c0392b;border-color:#c0392b">删除</button>
@@ -1411,7 +1416,9 @@
              renderDetail();
              return;
            }
-            const payload = await updateStagingReview(query, dbDecision, note);
+            const payload = await updateStagingReview(
+              query, dbDecision, note, localDecision === 'duplicate' ? (candidate?.candidate_id || '') : ''
+            );
            for (const row of state.queryRows) applyReviewSnapshot(row, payload.record);
            // 刷新统计和加载列表并行
            await Promise.all([reloadSummary(), loadGroups()]);
@@ -1530,12 +1537,12 @@
        alert('历史报表仅供查看，请在 staging-db 中执行入库。');
        return;
      }
-      if (!confirm('确认入库数据库中所有"确认入库（待入库）"样本？')) return;
+      if (!confirm('确认提交所有审核结果？“确认不重复”将入库，“确认重复”将把词/曲作者增量合并到已有记录。')) return;
      els.importReviewedBtn.disabled = true;
      els.importReviewedBtn.textContent = '入库中...';
      try {
        const payload = await reviewPostJSON('/api/import-reviewed');
-        alert(`入库完成：${payload.inserted_count} 条。结果文件：${payload.review_csv}`);
+        alert(`提交完成：不重复入库 ${payload.inserted_count} 条，重复样本 ${payload.duplicate_count || 0} 条，其中作者有增量 ${payload.author_merged_count || 0} 条。结果文件：${payload.review_csv}`);
        // 刷新统计和加载列表并行
        await Promise.all([reloadSummary(), loadGroups()]);
        renderAll();
@@ -1543,14 +1550,17 @@
        if (err.status === 409 && err.payload?.code === 'import_conflict') {
          const conflicts = err.payload.conflicts || [];
          const details = conflicts.slice(0, 20).map((row, index) => {
-            const types = (row.conflict_types || []).map(type =>
+            const typeLabels = {
-              type === 'primary_key' ? '主键 ID 冲突' : '来源键冲突'
+              primary_key: '主键 ID 冲突',
-            ).join(' + ') || '唯一键冲突';
+              source_key: '来源键冲突',
+              duplicate_target_missing: '未找到已入库的重复目标'
+            };
+            const types = (row.conflict_types || []).map(type => typeLabels[type] || type).join(' + ') || '数据冲突';
            const targetIds = [row.id_conflict_target_id, row.source_conflict_target_id]
              .filter(value => value !== null && value !== undefined)
              .filter((value, i, values) => values.indexOf(value) === i)
              .join('/');
-            return `${index + 1}. ${row.name || '(无歌名)'} | source=${row.source_table_name || '-'}:${row.source_song_id || '-'} | staging_id=${row.staging_id} | 待入库ID=${row.incoming_id || '-'} | ${types} | 正式表ID=${targetIds || '-'}`;
+            return `${index + 1}. ${row.name || '(无歌名)'} | source=${row.source_table_name || '-'}:${row.source_song_id || '-'} | staging_id=${row.staging_id} | 待入库ID=${row.incoming_id || '-'} | 重复候选=${row.matched_song_id || '-'} | ${types} | 正式表ID=${targetIds || '-'}`;
          });
          if (conflicts.length > 20) details.push(`……另有 ${conflicts.length - 20} 条未展开`);
          alert([
@@ -1565,7 +1575,7 @@
        }
      } finally {
        els.importReviewedBtn.disabled = false;
-        els.importReviewedBtn.textContent = '入库审核通过';
+        els.importReviewedBtn.textContent = '提交审核结果';
      }
    }
@@ -1659,7 +1669,7 @@
          group.review_claim_expires_at = remote.review_claim_expires_at || '';
          group.claimed_by_me = Boolean(remote.claimed_by_me);
          group.reviewed = ['approved_import', 'rejected_duplicate', 'deleted'].includes(remote.biz_review_status);
-          group.submitted = group.reviewed && remote.staging_status === 'imported';
+          group.submitted = group.reviewed && ['imported', 'skipped'].includes(remote.staging_status);
        }
        const query = currentQuery();
        const selectedRemote = query ? statuses.get(String(query.staging_id)) : null;
--- a/serve_l2_dashboard.py
View file @bbb6ce1
+++ b/serve_l2_dashboard.py
View file @bbb6ce1
@@ -8,17 +8,20 @@ import csv
 import json
 import mimetypes
 import os
+import re
 import secrets
 import subprocess
 import sys
 import threading
 import time
+import unicodedata
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
 from urllib.parse import parse_qs, unquote, urlparse
 import pymysql
 import requests
+import opencc
 from dbutils.pooled_db import PooledDB
 from dotenv import load_dotenv
@@ -79,6 +82,9 @@ REVIEW_SCHEMA_COLUMNS = {
    "review_claim_expires_at": "datetime DEFAULT NULL COMMENT '领取过期时间'",
    "review_version": "bigint(20) NOT NULL DEFAULT '0' COMMENT '审核乐观锁版本'",
 }
+_AUTHOR_SEP_RE = re.compile(r"[,，/;；、]")
+_AUTHOR_T2S = opencc.OpenCC("t2s")
+_AUTHOR_METADATA_PUNCT = set(' \t\n\r，。！？；：、"“”‘’·…—～！￥（）【】《》〈〉「」『』﹏,.:;!?()[]{}<>|/\\_-')
 class ReviewConflictError(Exception):
@@ -250,8 +256,8 @@ def _staging_summary() -> list[dict[str, str]]:
      SUM(CASE WHEN biz_review_status = 'approved_import' THEN 1 ELSE 0 END) AS approved_count,
      SUM(CASE WHEN biz_review_status = 'rejected_duplicate' THEN 1 ELSE 0 END) AS rejected_count,
      SUM(CASE WHEN biz_review_status = 'deleted' THEN 1 ELSE 0 END) AS deleted_count,
-      SUM(CASE WHEN biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted') AND staging_status <> 'imported' THEN 1 ELSE 0 END) AS reviewed_count,
+      SUM(CASE WHEN biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted') AND staging_status NOT IN ('imported', 'skipped') THEN 1 ELSE 0 END) AS reviewed_count,
-      SUM(CASE WHEN biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted') AND staging_status = 'imported' THEN 1 ELSE 0 END) AS submitted_count
+      SUM(CASE WHEN biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted') AND staging_status IN ('imported', 'skipped') THEN 1 ELSE 0 END) AS submitted_count
    FROM (
      SELECT source_song_id, dedup_action, biz_review_status, staging_status
      FROM {TARGET_TABLE_NAME_TMP}
@@ -356,10 +362,13 @@ def _staging_groups_response(
        where_clauses.append("s.dedup_action = 'skip'")
    elif decision == 'reviewed':
        where_clauses.append("s.biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted')")
-        where_clauses.append("s.staging_status <> 'imported'")
+        where_clauses.append("s.staging_status NOT IN ('imported', 'skipped')")
    elif decision == 'submitted':
        where_clauses.append("s.biz_review_status IN ('approved_import', 'rejected_duplicate', 'deleted')")
-        where_clauses.append("s.staging_status = 'imported'")
+        where_clauses.append("s.staging_status IN ('imported', 'skipped')")
+    elif decision == 'deleted':
+        where_clauses.append("s.biz_review_status = 'deleted'")
+        where_clauses.append("s.staging_status = 'deleted'")
    elif decision == 'pending':
        where_clauses.append("(s.biz_review_status IS NULL OR s.biz_review_status = 'pending')")
        where_clauses.append("s.dedup_action = 'review'")
@@ -427,7 +436,7 @@ def _staging_groups_response(
            "reviewed": row.get("biz_review_status") in {"approved_import", "rejected_duplicate", "deleted"},
            "submitted": (
                row.get("biz_review_status") in {"approved_import", "rejected_duplicate", "deleted"}
-                and row.get("staging_status") == "imported"
+                and row.get("staging_status") in {"imported", "skipped"}
            ),
            "staging_id": str(row.get("staging_id") or ""),
            "review_version": int(row.get("review_version") or 0),
@@ -674,6 +683,7 @@ def _update_staging_review(
    reviewer: str,
    expected_version: int,
    client_token: str,
+    candidate_id: str = "",
 ) -> dict[str, object]:
    """Persist one claimed review using optimistic concurrency control."""
    if decision == "pending":
@@ -686,6 +696,11 @@ def _update_staging_review(
            biz_review_status = 'deleted', reviewed_by = %s, reviewed_at = NOW(),
            staging_status = 'deleted'
        """
+    elif decision == "rejected_duplicate":
+        status_fields = """
+            biz_review_status = %s, reviewed_by = %s, reviewed_at = NOW(),
+            matched_song_id = %s
+        """
    else:
        status_fields = "biz_review_status = %s, reviewed_by = %s, reviewed_at = NOW()"
@@ -694,9 +709,31 @@ def _update_staging_review(
        params = [note, staging_id, expected_version, client_token, reviewer]
    elif decision == "deleted":
        params = [reviewer, note, staging_id, expected_version, client_token, reviewer]
+    elif decision == "rejected_duplicate":
+        candidate_id = candidate_id.strip()
+        if not candidate_id:
+            raise ValueError("确认重复前请先选择一条召回候选")
+        params = [decision, reviewer, candidate_id, note, staging_id, expected_version, client_token, reviewer]
    else:
        params = [decision, reviewer, note, staging_id, expected_version, client_token, reviewer]
    with _target_conn() as conn, conn.cursor() as cursor:
+        if decision == "rejected_duplicate":
+            cursor.execute(
+                f"SELECT matched_song_id, recalled_candidates FROM {TARGET_TABLE_NAME_TMP} WHERE staging_id = %s",
+                (staging_id,),
+            )
+            before = cursor.fetchone() or {}
+            allowed_ids = {str(before.get("matched_song_id") or "")}
+            recalled = before.get("recalled_candidates")
+            try:
+                candidates = json.loads(recalled) if isinstance(recalled, str) else (recalled or [])
+                allowed_ids.update(str(item.get("id") or "") for item in candidates if isinstance(item, dict))
+            except (json.JSONDecodeError, TypeError):
+                pass
+            allowed_ids.discard("")
+            if candidate_id not in allowed_ids:
+                conn.rollback()
+                raise ValueError("选中的重复候选不在当前召回结果中，请刷新后重试")
        cursor.execute(
            f"""
            UPDATE {TARGET_TABLE_NAME_TMP}
@@ -724,8 +761,36 @@ def _update_staging_review(
    return current
+def _merge_author_field(existing: str | None, incoming: str | None) -> str:
+    """Match import_hk_songs.py: preserve existing authors and append unique incoming authors."""
+    if not incoming:
+        return existing or ""
+    if not existing:
+        return incoming
+    def split_authors(text: str) -> list[str]:
+        return [author.strip() for author in _AUTHOR_SEP_RE.split(text) if author.strip()]
+    def normalize(author: str) -> str:
+        author = unicodedata.normalize("NFKC", author)
+        author = _AUTHOR_T2S.convert(author.strip().lower())
+        return "".join(char for char in author if char not in _AUTHOR_METADATA_PUNCT)
+    existing_authors = split_authors(existing)
+    incoming_authors = split_authors(incoming)
+    normalized = {normalize(author) for author in existing_authors}
+    result = list(existing_authors)
+    for author in incoming_authors:
+        key = normalize(author)
+        if key not in normalized:
+            result.append(author)
+            normalized.add(key)
+    merged = "、".join(result)
+    return merged if merged != existing else ""
 def _import_approved_staging(source_ids: list[str] | None = None) -> dict[str, object]:
-    """Import approved rows while holding staging-row locks to prevent duplicate imports."""
+    """Finalize reviewed rows: import non-duplicates and merge duplicate authors."""
    columns = ", ".join(f"`{column}`" for column in TARGET_COLUMNS)
    select_columns = ", ".join(f"s.`{column}`" for column in TARGET_COLUMNS)
    with _target_conn() as conn, conn.cursor() as cursor:
@@ -737,7 +802,8 @@ def _import_approved_staging(source_ids: list[str] | None = None) -> dict[str, o
            lock_params.extend(source_ids)
        cursor.execute(
            f"""
-            SELECT s.staging_id, s.source_song_id
+            SELECT s.staging_id, s.id AS incoming_id, s.source_song_id, s.source_table_name, s.name,
+                   s.lyricist, s.composer, s.biz_review_status, s.matched_song_id
            FROM {TARGET_TABLE_NAME_TMP} s
            INNER JOIN (
                SELECT source_song_id, MAX(staging_id) AS max_staging_id
@@ -745,7 +811,7 @@ def _import_approved_staging(source_ids: list[str] | None = None) -> dict[str, o
                GROUP BY source_song_id
            ) latest ON s.staging_id = latest.max_staging_id
            WHERE s.dedup_action = 'review'
-              AND s.biz_review_status = 'approved_import'
+              AND s.biz_review_status IN ('approved_import', 'rejected_duplicate')
              AND s.staging_status = 'staged'
              AND (s.review_claim_expires_at IS NULL OR s.review_claim_expires_at <= NOW())
              {source_filter}
@@ -754,50 +820,80 @@ def _import_approved_staging(source_ids: list[str] | None = None) -> dict[str, o
            """,
            lock_params,
        )
-        approved_rows = cursor.fetchall()
+        reviewed_rows = cursor.fetchall()
-        if not approved_rows:
+        if not reviewed_rows:
            conn.rollback()
-            raise ValueError("没有可入库的审核通过样本")
+            raise ValueError("没有可提交的已审核样本")
+        approved_rows = [row for row in reviewed_rows if row["biz_review_status"] == "approved_import"]
+        duplicate_rows = [row for row in reviewed_rows if row["biz_review_status"] == "rejected_duplicate"]
        staging_ids = [row["staging_id"] for row in approved_rows]
        approved_source_ids = [str(row["source_song_id"]) for row in approved_rows]
-        staging_placeholders = ",".join(["%s"] * len(staging_ids))
+        duplicate_targets: dict[object, int] = {}
-        conflicts = _find_import_conflicts(cursor, staging_ids)
+        unresolved_duplicates = []
-        if conflicts:
+        for row in duplicate_rows:
+            target = _resolve_duplicate_target(cursor, row)
+            if not target:
+                unresolved_duplicates.append({
+                    "staging_id": row["staging_id"],
+                    "source_song_id": row["source_song_id"],
+                    "source_table_name": row.get("source_table_name"),
+                    "name": row.get("name"),
+                    "incoming_id": row.get("incoming_id"),
+                    "id_conflict_target_id": None,
+                    "source_conflict_target_id": None,
+                    "conflict_types": ["duplicate_target_missing"],
+                    "matched_song_id": row.get("matched_song_id"),
+                })
+            else:
+                duplicate_targets[row["staging_id"]] = int(target["id"])
+        if unresolved_duplicates:
            conn.rollback()
            raise ImportConflictError(
-                f"发现 {len(conflicts)} 条待入库数据与正式表已有记录冲突，本次未写入任何数据",
+                f"有 {len(unresolved_duplicates)} 条“确认重复”记录无法找到已入库的重复目标，本次未写入任何数据",
-                conflicts,
+                unresolved_duplicates,
            )
-        try:
+        inserted_count = 0
+        id_map: dict[str, object] = {}
+        if approved_rows:
+            staging_placeholders = ",".join(["%s"] * len(staging_ids))
+            conflicts = _find_import_conflicts(cursor, staging_ids)
+            if conflicts:
+                conn.rollback()
+                raise ImportConflictError(
+                    f"发现 {len(conflicts)} 条待入库数据与正式表已有记录冲突，本次未写入任何数据",
+                    conflicts,
+                )
+            try:
+                cursor.execute(
+                    f"""
+                    INSERT INTO {TARGET_TABLE_NAME} ({columns})
+                    SELECT {select_columns}
+                    FROM {TARGET_TABLE_NAME_TMP} s
+                    WHERE s.staging_id IN ({staging_placeholders})
+                    """,
+                    staging_ids,
+                )
+            except pymysql.err.IntegrityError as exc:
+                conn.rollback()
+                conflicts = _find_import_conflicts(cursor, staging_ids)
+                raise ImportConflictError(
+                    "入库时发生唯一键或主键冲突，本次未写入任何数据",
+                    conflicts,
+                    str(exc),
+                ) from exc
+            inserted_count = cursor.rowcount
+            source_placeholders = ",".join(["%s"] * len(approved_source_ids))
            cursor.execute(
                f"""
-                INSERT INTO {TARGET_TABLE_NAME} ({columns})
+                SELECT source_song_id, id AS target_id
-                SELECT {select_columns}
+                FROM {TARGET_TABLE_NAME}
-                FROM {TARGET_TABLE_NAME_TMP} s
+                WHERE source_song_id IN ({source_placeholders})
-                WHERE s.staging_id IN ({staging_placeholders})
                """,
-                staging_ids,
+                approved_source_ids,
            )
-        except pymysql.err.IntegrityError as exc:
+            id_map = {str(row["source_song_id"]): row["target_id"] for row in cursor.fetchall()}
-            # 预检与 INSERT 之间仍可能有其他程序写入，回滚后再查一次以返回具体行。
-            conn.rollback()
-            conflicts = _find_import_conflicts(cursor, staging_ids)
-            raise ImportConflictError(
-                "入库时发生唯一键或主键冲突，本次未写入任何数据",
-                conflicts,
-                str(exc),
-            ) from exc
-        inserted_count = cursor.rowcount
-        source_placeholders = ",".join(["%s"] * len(approved_source_ids))
-        cursor.execute(
-            f"""
-            SELECT source_song_id, id AS target_id
-            FROM {TARGET_TABLE_NAME}
-            WHERE source_song_id IN ({source_placeholders})
-            """,
-            approved_source_ids,
-        )
-        id_map = {str(row["source_song_id"]): row["target_id"] for row in cursor.fetchall()}
        imported_updated = 0
        for approved_row in approved_rows:
            sid = str(approved_row["source_song_id"])
@@ -814,15 +910,89 @@ def _import_approved_staging(source_ids: list[str] | None = None) -> dict[str, o
                (target_id, approved_row["staging_id"]),
            )
            imported_updated += cursor.rowcount
+        duplicate_updated = 0
+        author_merged = 0
+        for duplicate_row in duplicate_rows:
+            target_id = duplicate_targets[duplicate_row["staging_id"]]
+            cursor.execute(
+                f"SELECT id, lyricist, composer FROM {TARGET_TABLE_NAME} WHERE id = %s AND deleted = '0' FOR UPDATE",
+                (target_id,),
+            )
+            target = cursor.fetchone()
+            if not target:
+                raise RuntimeError(f"重复目标已失效: target_id={target_id}")
+            merged_lyricist = _merge_author_field(target.get("lyricist"), duplicate_row.get("lyricist"))
+            merged_composer = _merge_author_field(target.get("composer"), duplicate_row.get("composer"))
+            changed = False
+            if merged_lyricist:
+                cursor.execute(
+                    f"UPDATE {TARGET_TABLE_NAME} SET lyricist = %s WHERE id = %s",
+                    (merged_lyricist, target_id),
+                )
+                changed = True
+            if merged_composer:
+                cursor.execute(
+                    f"UPDATE {TARGET_TABLE_NAME} SET composer = %s WHERE id = %s",
+                    (merged_composer, target_id),
+                )
+                changed = True
+            if changed:
+                author_merged += 1
+            cursor.execute(
+                f"""
+                UPDATE {TARGET_TABLE_NAME_TMP}
+                SET staging_status = 'skipped', imported_song_id = %s, error_message = NULL,
+                    merge_authors = %s, review_version = review_version + 1
+                WHERE staging_id = %s AND staging_status = 'staged'
+                  AND biz_review_status = 'rejected_duplicate'
+                """,
+                (target_id, 1 if changed else 0, duplicate_row["staging_id"]),
+            )
+            if cursor.rowcount != 1:
+                raise RuntimeError(f"重复样本终态更新失败: staging_id={duplicate_row['staging_id']}")
+            duplicate_updated += 1
        conn.commit()
    return {
        "approved_count": len(approved_rows),
        "inserted_count": inserted_count,
        "imported_song_ids_updated": imported_updated,
+        "duplicate_count": len(duplicate_rows),
+        "duplicate_staging_updated": duplicate_updated,
+        "author_merged_count": author_merged,
        "source_ids": approved_source_ids,
+        "duplicate_source_ids": [str(row["source_song_id"]) for row in duplicate_rows],
    }
+def _resolve_duplicate_target(cursor, row: dict[str, object]) -> dict[str, object] | None:
+    """Resolve a human-confirmed duplicate to an existing, active target record."""
+    matched_id = str(row.get("matched_song_id") or "").strip()
+    if not matched_id:
+        return None
+    try:
+        numeric_id = int(matched_id)
+    except (TypeError, ValueError):
+        numeric_id = None
+    if numeric_id is not None:
+        cursor.execute(
+            f"SELECT id FROM {TARGET_TABLE_NAME} WHERE id = %s AND deleted = '0' FOR UPDATE",
+            (numeric_id,),
+        )
+        target = cursor.fetchone()
+        if target:
+            return target
+    cursor.execute(
+        f"""
+        SELECT id FROM {TARGET_TABLE_NAME}
+        WHERE source_table_name = %s AND source_song_id = %s AND deleted = '0'
+        FOR UPDATE
+        """,
+        (row.get("source_table_name"), matched_id),
+    )
+    return cursor.fetchone()
 def _find_import_conflicts(cursor, staging_ids: list[object]) -> list[dict[str, object]]:
    """Return target-table key conflicts for the selected staging rows."""
    if not staging_ids:
@@ -1276,6 +1446,7 @@ class Handler(BaseHTTPRequestHandler):
                staging_id = int(payload.get("staging_id") or 0)
                decision = str(payload.get("decision", "")).strip()
                note = str(payload.get("note", "")).strip()
+                candidate_id = str(payload.get("candidate_id", "")).strip()
                client_token = str(payload.get("client_token") or "").strip()
                expected_version = int(payload.get("expected_version", -1))
                if not staging_id:
@@ -1285,7 +1456,7 @@ class Handler(BaseHTTPRequestHandler):
                if decision not in ("approved_import", "rejected_duplicate", "unsure", "pending", "deleted"):
                    raise ValueError(f"invalid decision: {decision}")
                result = _update_staging_review(
-                    staging_id, decision, note, reviewer, expected_version, client_token
+                    staging_id, decision, note, reviewer, expected_version, client_token, candidate_id
                )
                _json_response(self, {"record": result})
            except ReviewConflictError as exc:
@@ -1310,11 +1481,19 @@ class Handler(BaseHTTPRequestHandler):
                if source_ids
                else None
            )
-            approved = [
+            finalized = [
                {"source_id": source_id, "review_decision": "import", "review_note": ""}
                for source_id in result["source_ids"]
            ]
-            review_csv = _write_review_import_csv(approved)
+            finalized.extend(
+                {
+                    "source_id": source_id,
+                    "review_decision": "duplicate_merged",
+                    "review_note": "确认重复，词/曲作者已增量合并到已有记录",
+                }
+                for source_id in result["duplicate_source_ids"]
+            )
+            review_csv = _write_review_import_csv(finalized)
            _json_response(self, {"review_csv": str(review_csv), **result})
        except ReviewAuthError as exc:
            _error(self, str(exc), status=401)
--- a/test_serve_l2_dashboard.py
View file @bbb6ce1
+++ b/test_serve_l2_dashboard.py
View file @bbb6ce1
@@ -96,6 +96,45 @@ def test_stale_review_returns_conflict_without_overwrite():
    assert not conn.committed
+def test_rejected_duplicate_persists_the_selected_candidate():
+    snapshot = {
+        "staging_id": 42,
+        "matched_song_id": None,
+        "recalled_candidates": '[{"id":"existing-song"}]',
+        "biz_review_status": "rejected_duplicate",
+        "review_version": 8,
+    }
+    cursor = FakeCursor(update_rowcount=1, snapshot=snapshot)
+    conn = FakeConnection(cursor)
+    with patch.object(dashboard, "_target_conn", return_value=conn):
+        dashboard._update_staging_review(
+            42, "rejected_duplicate", "same", "alice", 7, "browser-token", "existing-song"
+        )
+    update_sql, params = cursor.executions[1]
+    assert "matched_song_id = %s" in update_sql
+    assert params[:3] == ["rejected_duplicate", "alice", "existing-song"]
+    assert conn.committed
+def test_rejected_duplicate_rejects_a_candidate_outside_recall_results():
+    cursor = FakeCursor(
+        update_rowcount=1,
+        snapshot={"staging_id": 42, "matched_song_id": "candidate-a", "recalled_candidates": "[]"},
+    )
+    conn = FakeConnection(cursor)
+    with patch.object(dashboard, "_target_conn", return_value=conn):
+        with pytest.raises(ValueError, match="不在当前召回结果"):
+            dashboard._update_staging_review(
+                42, "rejected_duplicate", "", "alice", 7, "browser-token", "candidate-b"
+            )
+    assert conn.rolled_back
+    assert not any(sql.lstrip().startswith("UPDATE") for sql, _params in cursor.executions)
 def test_pending_review_restores_a_deleted_staging_row():
    snapshot = {
        "staging_id": 42,
@@ -185,7 +224,7 @@ class ImportCursor(FakeCursor):
    def __init__(self) -> None:
        super().__init__(update_rowcount=1, snapshot={})
        self.result_sets = [
-            [{"staging_id": 42, "source_song_id": "song-1"}],
+            [{"staging_id": 42, "source_song_id": "song-1", "biz_review_status": "approved_import"}],
            [],
            [{"source_song_id": "song-1", "target_id": 9001}],
        ]
@@ -217,7 +256,7 @@ def test_import_locks_only_unclaimed_staged_rows():
 def test_import_conflict_rolls_back_and_exposes_the_staging_row():
    cursor = ImportCursor()
    cursor.result_sets = [
-        [{"staging_id": 42, "source_song_id": "song-1"}],
+        [{"staging_id": 42, "source_song_id": "song-1", "biz_review_status": "approved_import"}],
        [{
            "staging_id": 42,
            "source_song_id": "song-1",
@@ -241,6 +280,69 @@ def test_import_conflict_rolls_back_and_exposes_the_staging_row():
    assert not any(sql.lstrip().startswith("INSERT") for sql, _params in cursor.executions)
+class DuplicateImportCursor:
+    def __init__(self) -> None:
+        self.rowcount = 0
+        self.executions: list[tuple[str, object]] = []
+        self.current = None
+    def __enter__(self):
+        return self
+    def __exit__(self, *_args):
+        return False
+    def execute(self, sql, params=None):
+        self.executions.append((sql, params))
+        compact = " ".join(sql.split())
+        self.rowcount = 0
+        if "SELECT s.staging_id" in compact and "FOR UPDATE" in compact:
+            self.current = [{
+                "staging_id": 42,
+                "source_song_id": "new-song",
+                "source_table_name": "hk_song_platform",
+                "name": "same song",
+                "lyricist": "Alice/Bob",
+                "composer": "Composer B",
+                "biz_review_status": "rejected_duplicate",
+                "matched_song_id": "existing-song",
+            }]
+        elif "SELECT id FROM" in compact and "source_table_name = %s" in compact:
+            self.current = {"id": 9001}
+        elif "SELECT id, lyricist, composer" in compact:
+            self.current = {"id": 9001, "lyricist": "Alice", "composer": "Composer A"}
+        elif compact.startswith("UPDATE"):
+            self.current = None
+            self.rowcount = 1
+        else:
+            raise AssertionError(f"unexpected SQL: {compact}")
+    def fetchall(self):
+        return self.current or []
+    def fetchone(self):
+        return self.current
+def test_confirmed_duplicate_merges_authors_without_inserting_a_new_song():
+    cursor = DuplicateImportCursor()
+    conn = FakeConnection(cursor)
+    with patch.object(dashboard, "_target_conn", return_value=conn):
+        result = dashboard._import_approved_staging()
+    sql_params = [(" ".join(sql.split()), params) for sql, params in cursor.executions]
+    assert not any(sql.startswith("INSERT") for sql, _params in sql_params)
+    assert any("SET lyricist = %s" in sql and params == ("Alice、Bob", 9001) for sql, params in sql_params)
+    assert any("SET composer = %s" in sql and params == ("Composer A、Composer B", 9001) for sql, params in sql_params)
+    assert any("SET staging_status = 'skipped'" in sql for sql, _params in sql_params)
+    assert result["duplicate_count"] == 1
+    assert result["author_merged_count"] == 1
+    assert result["inserted_count"] == 0
+    assert result["duplicate_source_ids"] == ["new-song"]
+    assert conn.committed
 def test_browsing_a_list_item_can_auto_claim_without_reloading_groups():
    html = Path("l2_review_dashboard.html").read_text(encoding="utf-8")
    start = html.index("for (const btn of els.queryList.querySelectorAll('.query-item'))")
@@ -274,7 +376,17 @@ def test_dashboard_distinguishes_reviewed_from_submitted():
    assert row["staging_status"] == "imported"
    assert '<option value="reviewed">已审核</option>' in html
    assert '<option value="submitted">已提交</option>' in html
-    assert "hasFinalReview && query.staging_status !== 'imported'" in html
+    assert "hasFinalReview && query.staging_status === 'staged'" in html
+def test_dashboard_has_a_dedicated_deleted_filter():
+    html = Path("l2_review_dashboard.html").read_text(encoding="utf-8")
+    assert '<option value="deleted">已删除</option>' in html
+    source = Path("serve_l2_dashboard.py").read_text(encoding="utf-8")
+    assert "elif decision == 'deleted':" in source
+    assert "s.biz_review_status = 'deleted'" in source
+    assert "s.staging_status = 'deleted'" in source
 def test_review_access_code_issues_and_validates_edit_session():