fix: 去重合并多来源后同步到 events/Qdrant/MySQL，新增 sync-sources 回填命令

- dedup/pipeline: 合并 source_ids 后自动同步已生成的 events JSON 和 Qdrant payload
- 新增 sync_sources_from_deduped / uv run en-news sync-sources，用于历史数据回填
- sync-sources 会扫描 deduped 多来源唯一篇，更新 events、Qdrant、MySQL news_event.sources
- docs: 补充 sync-sources 使用说明

dedup/pipeline.py

@@ -172,10 +172,160 @@ def _merge_duplicate_source(article: ProcessedArticle, result: DedupResult) -> N
         )
         logger.debug("来源合并: %s → %s (sources=%s)",
                      article.source_id, result.matched_url_hash, merged)
+        # 同步到已生成的 events / Qdrant，避免“deduped 有多来源但 DB 没有”
+        _sync_merged_sources_downstream(result.matched_url_hash, merged)
     except Exception as e:
         logger.exception("来源合并失败 %s: %s", target, e)
 
 
+def _sync_event_source_ids(url_hash: str, source_ids: list[str]) -> int:
+    """把合并后的 source_ids 同步到已生成的 events JSON。"""
+    updated = 0
+    for ev_path in sorted(Path("data/events").glob(f"*/{url_hash}.json")):
+        try:
+            data = json.loads(ev_path.read_text(encoding="utf-8"))
+            if data.get("source_ids") != source_ids:
+                data["source_ids"] = list(source_ids)
+                ev_path.write_text(
+                    json.dumps(data, indent=2, ensure_ascii=False),
+                    encoding="utf-8",
+                )
+                logger.info("来源合并同步 events: %s -> %s", ev_path, source_ids)
+                updated += 1
+        except Exception as e:
+            logger.exception("同步 events 来源失败 %s: %s", ev_path, e)
+    return updated
+
+
+def _sync_qdrant_source_ids(url_hash: str, source_ids: list[str]) -> int:
+    """把合并后的 source_ids 同步到 Qdrant 中已存在的向量点。"""
+    emb_paths = sorted(Path("data/embeddings").glob(f"*/{url_hash}.json"))
+    if not emb_paths:
+        return 0
+
+    event_path = next(Path("data/events").glob(f"*/{url_hash}.json"), None)
+    if event_path is None:
+        return 0
+
+    try:
+        article_data = json.loads(event_path.read_text(encoding="utf-8"))
+    except Exception as e:
+        logger.warning("读取 events 失败，跳过 Qdrant 来源同步 %s: %s", event_path, e)
+        return 0
+
+    try:
+        from vectorstore.client import VectorStore, make_qdrant_client
+        from vectorstore.pipeline import _build_payload
+
+        client = make_qdrant_client()
+        store = VectorStore(client)
+        try:
+            payload = _build_payload(article_data)
+            for emb_path in emb_paths:
+                emb_data = json.loads(emb_path.read_text(encoding="utf-8"))
+                store.upsert([{
+                    "id": url_hash,
+                    "vector": emb_data["vector"],
+                    "payload": payload,
+                }])
+        finally:
+            store.close()
+        logger.info("来源合并同步 Qdrant: %s -> %s", url_hash, source_ids)
+        return len(emb_paths)
+    except Exception as e:
+        logger.warning("同步 Qdrant 来源失败 %s: %s", url_hash, e)
+        return 0
+
+
+def _sync_merged_sources_downstream(url_hash: str, source_ids: list[str]) -> None:
+    """去重合并来源后，同步更新 events 与 Qdrant 中已存在的下游数据。"""
+    _sync_event_source_ids(url_hash, source_ids)
+    _sync_qdrant_source_ids(url_hash, source_ids)
+
+
+def sync_sources_from_deduped() -> dict:
+    """历史数据回填：扫描所有 deduped 唯一篇，把多来源同步到 events/Qdrant/MySQL。
+
+    该函数用于修复“deduped 已合并多来源，但旧 events/数据库未更新”的历史数据。
+    """
+    stats = {
+        "deduped_scanned": 0,
+        "multi_source": 0,
+        "events_updated": 0,
+        "qdrant_updated": 0,
+        "db_updated": 0,
+    }
+
+    targets = sorted(Path("data/deduped").glob("*/uniques/*.json"))
+    for target in targets:
+        try:
+            data = json.loads(target.read_text(encoding="utf-8"))
+        except Exception:
+            continue
+        stats["deduped_scanned"] += 1
+
+        raw_ids = data.get("source_ids") or [data.get("source_id")]
+        source_ids = list(dict.fromkeys(x for x in raw_ids if x))
+        if len(source_ids) <= 1:
+            continue
+
+        stats["multi_source"] += 1
+        url_hash = data.get("url_hash", target.stem)
+        stats["events_updated"] += _sync_event_source_ids(url_hash, source_ids)
+        stats["qdrant_updated"] += _sync_qdrant_source_ids(url_hash, source_ids)
+        stats["db_updated"] += _update_db_sources(url_hash, source_ids)
+
+    logger.info(
+        "来源回填完成: scanned=%d multi=%d events=%d qdrant=%d db=%d",
+        stats["deduped_scanned"], stats["multi_source"],
+        stats["events_updated"], stats["qdrant_updated"], stats["db_updated"],
+    )
+    return stats
+
+
+def _update_db_sources(url_hash: str, source_ids: list[str]) -> int:
+    """根据最新 events 来源，更新 MySQL news_event 中该文章的 source/sources。"""
+    event_path = next(Path("data/events").glob(f"*/{url_hash}.json"), None)
+    if event_path is None:
+        return 0
+    try:
+        article_data = json.loads(event_path.read_text(encoding="utf-8"))
+    except Exception:
+        return 0
+
+    try:
+        from scheduler.reporter import _article_source_label, _article_sources_full
+        source_label = _article_source_label(article_data)
+        source_list = _article_sources_full(article_data)
+        if not source_label and not source_list:
+            return 0
+
+        from report_db import connect
+        conn = connect()
+        try:
+            with conn.cursor() as cur:
+                cur.execute(
+                    """
+                    UPDATE news_event
+                    SET source = %s, sources = %s
+                    WHERE section = 'intl' AND url = %s
+                    """,
+                    (
+                        source_label,
+                        json.dumps(source_list, ensure_ascii=False) if source_list else None,
+                        article_data.get("url"),
+                    ),
+                )
+                updated = cur.rowcount
+            conn.commit()
+            return updated or 0
+        finally:
+            conn.close()
+    except Exception as e:
+        logger.warning("同步 MySQL 来源失败 %s: %s", url_hash, e)
+        return 0
+
+
 def _layer_num(result: DedupResult) -> int:
     """DedupResult → 命中层编号。"""
     if result.matched_layer is None: