<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Signal on oceanheart.ai</title>
    <link>https://www.oceanheart.ai/tags/signal/</link>
    <description>Recent content in Signal on oceanheart.ai</description>
    <generator>Hugo</generator>
    <language>en-gb</language>
    <lastBuildDate>Mon, 16 Mar 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://www.oceanheart.ai/tags/signal/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>I invented a DSL for governance. Then I killed it.</title>
      <link>https://www.oceanheart.ai/blog/2026-03-10-kill-signal/</link>
      <pubDate>Mon, 16 Mar 2026 00:00:00 +0000</pubDate>
      <guid>https://www.oceanheart.ai/blog/2026-03-10-kill-signal/</guid>
      <description>&lt;h2 id=&#34;what-signal-was&#34;&gt;What Signal was&lt;/h2&gt;&#xA;&lt;p&gt;I spent a few days developing a notation system for governance rules in an agentic engineering project. It looked like this:&lt;/p&gt;&#xA;&lt;pre tabindex=&#34;0&#34;&gt;&lt;code&gt;FOOTGUN high_on_own_supply :=&#xA;  L12.creativity &amp;amp; L9.sycophancy -&amp;gt; positive_feedback_loop&#xA;  BRAKE: bearing_check(NORTH)                                   [L9, L12]&#xA;&lt;/code&gt;&lt;/pre&gt;&lt;p&gt;The claim was 4.5:1 compression over verbose prose. The operators (&lt;code&gt;:=&lt;/code&gt;, &lt;code&gt;-&amp;gt;&lt;/code&gt;, &lt;code&gt;|&lt;/code&gt;, &lt;code&gt;&amp;amp;&lt;/code&gt;, &lt;code&gt;!&lt;/code&gt;) were meant to communicate logical relationships more efficiently than natural language.&lt;/p&gt;&#xA;&lt;p&gt;It decoded well. Claude scored 6/6 on comprehension. GPT-4 scored 8/8 on reasoning. I had a proof of concept and I was genuinely proud of it. Which is probably why I didn&amp;rsquo;t test the obvious alternative.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
