[JavaZeroo]

内容

新增只读分析脚本 research/ui/analyze_output_trace.py：把场景产物里的三份诊断文件拼成一张时间线。

• 读 output-trace.jsonl（输出块/输入块、终端号、网格尺寸、repaint 标记）；
• 读 actions-progress.jsonl 或 *-progress.jsonl（阶段与 t_ms）；
• 读 selection-discard.jsonl（丢弃原因、行号、字节/列数）；
• 对每条丢弃打印前后阶段，并列出 ±2 秒内的重绘批次及其消费时刻网格。

实际输出示例（本地全绿运行）

== run-el4xnl28/local-0
   discard row-text-changed row=0 between live-selected and invalidated-copy-rejected saved_bytes=23 current_bytes=17 saved_columns=74 current_columns=74
      repaint t=471562 terminal=1 columns=111 rows=7 bytes=210
      repaint t=471640 terminal=1 columns=149 rows=10 bytes=231
      repaint t=471687 terminal=1 columns=111 rows=7 bytes=210

即同一轮里重绘批次分别以 111×7 与 149×10 被消费——正是 #194 需要的"消费时刻几何"证据。

边界

• 纯只读分析，不运行 shell、不改期限；
• 不做启发式归因，只呈现时间线与尺寸；
• 不关闭 [R2][调查] 核对 resize 期间 ConPTY 重绘与终端视口/缓冲的对齐 #194/[R2][回归] 定位 hosted 选择场景首记录前超时 #73。

Refs #194 #73

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T19:43:00.698985Z	42f4b39	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.