Refresh papers and define Agent evaluation

papers/paper-insights.md

@@ -6,13 +6,16 @@ status: seed
 
 ## Current Snapshot
 
-- date: 2026-07-08
-- scope: arXiv Agent-related papers from 2025-07-08 to 2026-07-08
-- candidates seen: 1506
-- high-relevance candidates selected into paper items: 970
-- reserve candidates retained in manifest: 536
-- total paper items: 974
-- detailed summary: [Paper Corpus Summary](corpus-summary-2026-07-08.md)
+- date: 2026-07-27
+- baseline scope: arXiv Agent-related papers from 2025-07-08 to 2026-07-08
+- incremental scope: 2026-07-09 to 2026-07-27
+- incremental candidates seen: 249
+- incremental high-relevance candidates: 157
+- incremental new paper items: 151
+- incremental reserve candidates: 92
+- total paper items after refresh: 1125
+- baseline summary: [2026-07-08 corpus](corpus-summary-2026-07-08.md)
+- incremental summary: [2026-07-27 corpus increment](corpus-summary-2026-07-27.md)
 - caveat: high-recall automated sweep; most items are `queued` and need human skim.
 
 ## Strong Signals
@@ -23,6 +26,9 @@ status: seed
 - JD 高频词和论文趋势高度重合：memory、eval、planning/tool use、coding agent。
 - 最近一年 Agent 论文的重心明显落在 evaluation、tool use、memory、safety、coding agent、computer-use/GUI agent。
 - 2026-05 到 2026-07 的论文密度明显上升，说明 Agent research 正在快速扩张。
+- 最新增量继续把评测推向 protocol validity、动态工具环境、trajectory 诊断、确定性
+  verification、成本和部署条件。
+- 组件评测需要同时报告 gains 与 regressions；平均提升可能掩盖原本成功任务被新组件破坏。
 
 ## Weak or Unproven Claims