<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>MATRIX · OnTrack</title>
  <link>https://derekwang2002.github.io/ai-intelligence/en/projects/ontrack/</link>
  <description>Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations.</description>
  <language>en</language>
  <lastBuildDate>Mon, 02 Nov 2026 16:56:00 GMT</lastBuildDate>
  <atom:link href="https://derekwang2002.github.io/ai-intelligence/en/feeds/projects/ontrack.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>OnTrack flags failing agent trajectories in about one millisecond per step</title>
    <link>https://derekwang2002.github.io/ai-intelligence/en/events/ev-20261008-02/</link>
    <guid isPermaLink="false">ev-20261008-02-published</guid>
    <pubDate>Fri, 09 Oct 2026 15:56:33 GMT</pubDate>
    <category>New</category>
    <category>WATCH</category>
    <description>&lt;p&gt;OnTrack continuously compares an agent trajectory with the structure of successful runs to warn or stop failures during execution. On SWE-bench traces, the authors report about one millisecond per step, a 0.057 AUROC gain over content similarity after eight steps, and about 18% less compute spent on failed runs.&lt;/p&gt;&lt;p&gt;Recommendation: WATCH&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://arxiv.org/abs/2610.12375&quot;&gt;OnTrack&lt;/a&gt;&lt;/p&gt;</description>
  </item>
</channel>
</rss>
