跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.20540 · v1

Changing a View Is Not Amending a Record: Timing and Visibility in a Model of Public Correction

首次提交 2026-11-16 13:18
版本更新 2026-11-16 13:18
雷达观测 2026-11-17 13:07
arXiv ↗PDF ↗

摘要

An accusation may remain on the public record after its author has changed their mind. Observers can respond either to what a person says now or to an earlier statement that has never been explicitly amended. We examine the consequences of these two signals with a generative agent-based model that separates private evidence updates, current expression, and revisions linked to a specific declaration. In paired worlds, evidence, prehistory, and opportunities to revisit that declaration remain fixed. When expression and amendment are synchronized and fully observed, the feedback channels produce identical trajectories. With sparse review, record-based feedback leaves 19.44 percentage points more eligible accusations unamended after 24 rounds (paired-world 95 percent bootstrap interval 13.89 to 25.31). The comparison averages 50 eligible worlds, containing 556 agents, from 128 newly generated worlds. Event traces show how a persistent public cue changes subsequent amendment propensity; an exact small-state calculation independently establishes the timing boundary. Reducing record salience can erase or reverse the increment, and a generic held-state process reproduces the dynamics. The findings show when current expression ceases to track correction of a particular public statement.

作者备注:21 pages, 6 figures, including supplementary material

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-16 13:18 · 最新观测版本

Changing a View Is Not Amending a Record: Timing and Visibility in a Model of Public Correction

An accusation may remain on the public record after its author has changed their mind. Observers can respond either to what a person says now or to an earlier statement that has never been explicitly amended. We examine the consequences of these two signals with a generative agent-based model that separates private evidence updates, current expression, and revisions linked to a specific declaration. In paired worlds, evidence, prehistory, and opportunities to revisit that declaration remain fixed. When expression and amendment are synchronized and fully observed, the feedback channels produce identical trajectories. With sparse review, record-based feedback leaves 19.44 percentage points more eligible accusations unamended after 24 rounds (paired-world 95 percent bootstrap interval 13.89 to 25.31). The comparison averages 50 eligible worlds, containing 556 agents, from 128 newly generated worlds. Event traces show how a persistent public cue changes subsequent amendment propensity; an exact small-state calculation independently establishes the timing boundary. Reducing record salience can erase or reverse the increment, and a generic held-state process reproduces the dynamics. The findings show when current expression ceases to track correction of a particular public statement.

查看此版本解读 →查看此版本原文 ↗