From bdc6a53f81d97da4521101efe23374f8330ae6cd Mon Sep 17 00:00:00 2001
From: LuminousRuoxi <2494735589@qq.com>
Date: Fri, 11 Sep 2026 10:26:04 +0800
Subject: [PATCH] =?UTF-8?q?refactor:=20=E5=88=A0=E6=8E=89=E9=9A=8F=20agent?=
 =?UTF-8?q?=20=E5=8C=96=E5=A4=B1=E6=95=88=E7=9A=84=20applyUnderstandingCor?=
 =?UTF-8?q?rection=EF=BC=8C=E5=8E=9F=E4=BD=8D=E7=95=99=E5=B7=B2=E7=9F=A5?=
 =?UTF-8?q?=E5=81=8F=E5=B7=AE=E8=AF=B4=E6=98=8E?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

顾问纠正某段理解时，改造前它会同时改写该段的 semantic_anchor；现在走
record_advisor_note 只落成分段约束，锚点不跟着改。纠正内容仍会进诊断/复检的
paragraph_constraints，所以不是静默丢失，但与原行为有差——写在原函数位置而不是
只记在方案文档里，读代码的人能看见。
---
 index.html | 17 +++++------------
 1 file changed, 5 insertions(+), 12 deletions(-)

diff --git a/index.html b/index.html
index ebc156a..3dc40b9 100644
--- a/index.html
+++ b/index.html
@@ -770,18 +770,11 @@ function handleAction(action){
     renderEntryEssay();
   }
 }
-function applyUnderstandingCorrection(text){
-  const idx = parseParagraphIndex(text);
-  if(idx===null || !analysis || !analysis.paragraphs[idx]){
-    globalConstraints.push(text);
-    return {scope:'global', label:'文章理解的补充'};
-  }
-  analysis.paragraphs[idx].natural_meaning_zh = text;
-  analysis.paragraphs[idx].semantic_anchor = text;
-  const pid = pidOf(idx);
-  (paragraphConstraints[pid]=paragraphConstraints[pid]||[]).push(text);
-  return {scope:pid, label:`${paragraphLabel(idx)} 的理解`};
-}
+// 原先这里是 applyUnderstandingCorrection：顾问在理解态说「第三段重点不是……」时，
+// 它把这句话同时写进 analysis.paragraphs[i] 的 natural_meaning_zh / semantic_anchor 和
+// paragraphConstraints。Agent 化之后这条输入走 record_advisor_note，只落成分段约束，
+// **锚点不再跟着改**（纠正内容仍会进诊断/复检的 paragraph_constraints）。
+// 这是已知偏差，不是遗漏：要不要补一个带 anchor 更新的「纠正理解」工具待定。
 function applyDiagnosisConstraint(text){
   const raw = text.trim();
   const idx = parseParagraphIndex(raw);