From 594a762fe0d75840a2cd09c6c9b9898de63b9a80 Mon Sep 17 00:00:00 2001
From: JavaZeroo <71128095+JavaZeroo@users.noreply.github.com>
Date: Thu, 29 Oct 2026 21:34:51 +0800
Subject: [PATCH] research(ui): confirm the 1049 attribute save/restore model
 refs #123 #126

Three crossed cases: entry default plus alternate change exits default, entry 31;44 exits 31;44, and entry 32;45 with an alternate-side reset exits 32;45. Alternate-screen changes never leak back.
---
 .../2026-10-29-session-1049-palette/README.md | 17 +++++++++++++++
 .../validation.json                           | 21 +++++++++++++++++++
 2 files changed, 38 insertions(+)

diff --git a/research/ui/results/2026-10-29-session-1049-palette/README.md b/research/ui/results/2026-10-29-session-1049-palette/README.md
index ff378539..03bc86b8 100644
--- a/research/ui/results/2026-10-29-session-1049-palette/README.md
+++ b/research/ui/results/2026-10-29-session-1049-palette/README.md
@@ -72,6 +72,23 @@
 
 **归因边界**：仍是 conhost + ConPTY + WindTerm 整条管线的可观察渲染，无法判定哪一层执行保存/恢复；与 #126 要求的应用内采集不同。
 
+
+## 附加观察三：三组交叉验证"进入保存 / 退出恢复"
+
+把两组观察再补一组反向用例，形成三组交叉验证：
+
+| 组 | 进入时的属性 | 备用屏内的动作 | 退出后（不重设）的当前属性 | 结论 |
+|---|---|---|---|---|
+| 1 | 默认（脚本先 reset） | 设为 `32;45` | 默认 `#1C1C1C/#E3E3DD` | 备用屏改动**不泄漏**，恢复进入值 |
+| 2 | `31;44`（进入前设置） | 不变 | `#3465A4/#C4265E`（=31;44） | 恢复进入值 |
+| 3 | `32;45`（进入前设置） | reset 成默认 | `#8C6BC8/#86B42B`（=32;45） | 恢复进入值，备用屏的 reset 不带走主屏样式 |
+
+第 3 组的三阶段像素：进入前 `#8C6BC8/#86B42B` → 备用屏内 reset 后 `#1C1C1C/#E3E3DD` → 退出后回到 `#8C6BC8/#86B42B`。
+
+三组一致支持同一模型：**`1049h` 保存进入时刻的属性，`1049l` 恢复它**；备用屏内的任何 SGR 变更（设置或 reset）都不会影响主屏退出后的样式。
+
+**归因边界**：conhost + ConPTY + WindTerm 整条管线的可观察渲染；不判定哪一层执行保存/恢复，也不等同于 #126 要求的应用内采集。
+
 ## 解释与边界
 
 - 在这套固定配置里，`1049` 的进入/重复进入/退出**不改变**默认前景/背景，也不改变所测的三个调色板颜色；备用屏重进渲染与首次进入一致。
diff --git a/research/ui/results/2026-10-29-session-1049-palette/validation.json b/research/ui/results/2026-10-29-session-1049-palette/validation.json
index b67981cf..4370fe16 100644
--- a/research/ui/results/2026-10-29-session-1049-palette/validation.json
+++ b/research/ui/results/2026-10-29-session-1049-palette/validation.json
@@ -96,5 +96,26 @@
     },
     "interpretation": "1049h saves the attribute state at entry, alternate-screen changes do not leak back, and 1049l restores the entry-time attributes.",
     "attribution_limit": "pipeline-level observation; layer attribution unknown"
+  },
+  "attribute_save_restore": {
+    "model": "1049h saves the attribute state at entry; 1049l restores it; alternate-screen changes never leak to the main screen",
+    "cases": [
+      {
+        "entry": "default (reset by script)",
+        "alternate": "set 32;45",
+        "exit": "default 1C1C1C/E3E3DD"
+      },
+      {
+        "entry": "31;44",
+        "alternate": "unchanged",
+        "exit": "31;44 (#3465A4/#C4265E)"
+      },
+      {
+        "entry": "32;45",
+        "alternate": "reset to default",
+        "exit": "32;45 (#8C6BC8/#86B42B)"
+      }
+    ],
+    "attribution_limit": "pipeline-level observation; layer attribution unknown"
   }
 }