feat: organize news packages into verified event timelines

Add optional events, evidence-backed timelines, versioned package corrections and bounded historical reprocessing. Keep independent packages readable and isolate retries by failed stage. Configure reasoning effort per model stage and refresh default news Skill drafts without publishing.

backend/src/alphapulse/news_risk/research.py

@@ -78,6 +78,8 @@ TOOLS = {
     'read_news': (Read, '读取站内新闻原文'),
     'search_packages': (Search, '按关键词和标签检索其他包，AI总结仅作线索'),
     'read_package': (Read, '读取包及关联原文ID，核验应读取原文'),
+    'search_events': (Search, '检索历史事件及时间线，仅作线索'),
+    'read_event': (Read, '读取事件、成员包和时间线，原文仍须另行读取'),
     'search_calendar': (Search, '检索已有经济日历'),
     'search_web': (Search, '通过SearXNG搜索外部来源，摘要仅作线索'),
     'read_web': (ReadWeb, '读取搜索结果中的公开网页正文'),
@@ -92,6 +94,8 @@ class Research:
         self.cache = {}
         self.urls = set()
         self.package_ids = set()
+        self.read_package_ids = set()
+        self.read_event_ids = set()
 
     @staticmethod
     def tool_specs():
@@ -113,8 +117,10 @@ class Research:
                 self.sources[result['article_id']] = result
             if name == 'search_web': self.urls.update(r['url'] for r in result['results'] if r.get('url'))
             if name == 'search_calendar': self.urls.update(r['release_url'] for r in result if r.get('release_url'))
+            if name == 'read_event': self.read_event_ids.add(result['id'])
             if name == 'search_packages': self.package_ids.update(r['id'] for r in result)
-            if name == 'read_package': self.package_ids.add(result['id'])
+            if name == 'read_package':
+                self.package_ids.add(result['id']); self.read_package_ids.add(result['id'])
 
     async def tool(self, name, arguments):
         if name not in TOOLS: raise ValueError('未知研究工具')
@@ -140,11 +146,23 @@ class Research:
                 identifier = -(len(self.sources)+1)
                 self.sources[identifier] = result
                 result = {**result, 'article_id': identifier}
+            elif name in ('search_events', 'read_event'):
+                from alphapulse.models.risk import NewsEvent, EventMember
+                if name == 'search_events':
+                    rows = (await session.scalars(select(NewsEvent).where(NewsEvent.title.contains(args.query, autoescape=True), NewsEvent.state != 'retired').limit(20))).all()
+                    result = [{'id': e.id, 'title': e.title} for e in rows]
+                else:
+                    row = await session.get(NewsEvent, args.id)
+                    if row is None: raise ValueError('事件不存在')
+                    self.read_event_ids.add(row.id)
+                    members = list((await session.scalars(select(EventMember.package_id).where(EventMember.event_id == row.id))).all())
+                    result = {'id':row.id, 'title':row.title, 'result':row.result, 'package_ids':members, 'scope':'ai_summary_not_evidence'}
             elif name in ('search_news', 'search_packages'):
                 model = NewsItem if name == 'search_news' else NewsSummary
                 conditions = or_(model.title.contains(args.query, autoescape=True),
                     model.content_text.contains(args.query, autoescape=True)) if model is NewsItem else or_(
                     model.title.contains(args.query, autoescape=True), model.tags.contains([args.query]), model.fact.contains(args.query, autoescape=True))
+                if model is NewsSummary: conditions = conditions & (NewsSummary.replacement_ids == []) & (NewsSummary.state != 'withdrawn')
                 rows = (await session.scalars(select(model).where(conditions).order_by(model.published_at.desc()).limit(20))).all()
                 result = [{'id': r.id, 'title': r.title} for r in rows]
                 if model is NewsSummary: self.package_ids.update(r.id for r in rows)
@@ -156,12 +174,14 @@ class Research:
                     'summary': row.summary, 'url': row.url, 'published_at': row.published_at.isoformat()}
                 self.sources[identifier] = result
             elif name == 'read_package':
-                from alphapulse.models.risk import SummaryArticle, ArticleAnalysis
+                from alphapulse.models.risk import SummaryArticle, ArticleAnalysis, EventMember
                 row = await session.get(NewsSummary, args.id)
                 if row is None: raise ValueError('包不存在')
-                self.package_ids.add(row.id)
-                ids = (await session.scalars(select(ArticleAnalysis.news_id).join(SummaryArticle).where(SummaryArticle.summary_id == row.id))).all()
-                result = {'id': row.id, 'result': row.result, 'news_ids': ids, 'scope': 'ai_summary_not_evidence'}
+                self.package_ids.add(row.id); self.read_package_ids.add(row.id)
+                ids = (await session.scalars(select(ArticleAnalysis.news_id).join(SummaryArticle).where(SummaryArticle.summary_id == row.id, SummaryArticle.active.is_(True)))).all()
+                assignments = (await session.scalars(select(ArticleAnalysis).join(SummaryArticle).join(NewsItem).where(SummaryArticle.summary_id == row.id, SummaryArticle.active.is_(True), ArticleAnalysis.content_hash == NewsItem.content_hash))).all()
+                membership = await session.get(EventMember, row.id)
+                result = {'main_event_id': membership.event_id if membership else None, 'replacement_ids': row.replacement_ids, 'state': row.state, 'id': row.id, 'revision': row.revision, 'fact': row.fact, 'result': row.result, 'articles': [{'analysis_id': a.id, **a.input} for a in assignments], 'news_ids': ids, 'scope': 'ai_summary_not_evidence'}
             else:
                 from alphapulse.models.calendar import EconomicRelease
                 rows = (await session.scalars(select(EconomicRelease).where(EconomicRelease.title.contains(args.query, autoescape=True)).limit(20))).all()