<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Leakage on Programmer.ie</title>
    <link>http://programmer.ie/tags/leakage/</link>
    <description>Recent content in Leakage on Programmer.ie</description>
    <generator>Hugo</generator>
    <language>en-us</language>
    <lastBuildDate>Fri, 28 Aug 2026 11:15:00 +0100</lastBuildDate>
    <atom:link href="http://programmer.ie/tags/leakage/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>DSPy From First Principles 16: Don&#39;t Let the Optimizer Cheat</title>
      <link>http://programmer.ie/books/dspy-from-first-principles/16-chapter/</link>
      <pubDate>Fri, 28 Aug 2026 11:15:00 +0100</pubDate>
      <guid>http://programmer.ie/books/dspy-from-first-principles/16-chapter/</guid>
      <description>&lt;h1 id=&#34;dont-let-the-optimizer-cheat&#34;&gt;Don&amp;rsquo;t Let the Optimizer Cheat&lt;/h1&gt;&#xA;&lt;p&gt;Chapter 15 expanded the program&amp;rsquo;s access. It can search repositories, retrieve memory, use tools, and maybe explore large contexts programmatically.&lt;/p&gt;&#xA;&lt;p&gt;That power threatens the experiment.&lt;/p&gt;&#xA;&lt;p&gt;The central question:&lt;/p&gt;&#xA;&lt;blockquote&gt;&#xA;&lt;p&gt;How can an optimization experiment appear valid while the program or optimizer has quietly gained access to the answer?&lt;/p&gt;&lt;/blockquote&gt;&#xA;&lt;p&gt;Cheating is often accidental. A field named &lt;code&gt;validation_result&lt;/code&gt; slips into generation inputs. A memory tool returns the historical accepted patch. A retriever indexes a later repository revision. A GEPA feedback string includes the exact gold answer. A developer inspects holdout failures, changes instructions, and evaluates on the same holdout again.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
