<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Jordan Browne-Moore</title>
    <link>https://www.jordanbrowne-moore.com/</link>
    <description>Technical analysis and methodology from production ML, credit risk, and applied LLM work.</description>
    <language>en</language>
    <atom:link href="https://www.jordanbrowne-moore.com/feed.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Your Text to SQL Model Didn’t Fail. Your Evaluation Metric Did.</title>
      <link>https://www.jordanbrowne-moore.com/articles/text2sql-evaluation-harness</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/text2sql-evaluation-harness/</guid>
      <description>End-to-end accuracy conflates two specification failures with different fixes. Most Text to SQL evaluation harnesses score correct queries as wrong because nobody wrote down what correct means.</description>
      <pubDate>Thu, 01 Oct 2026 00:00:00 GMT</pubDate>
      <category>Applied ML</category>
    </item>
    <item>
      <title>SR 26-2 Excludes Generative AI. Your Model Risk Obligations Don’t.</title>
      <link>https://www.jordanbrowne-moore.com/articles/sr262-governance-evidence</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/sr262-governance-evidence/</guid>
      <description>SR 26-2 excludes generative and agentic AI from its scope while the UK&apos;s SS1/23 pulls the same pipeline in. Here is the evidence set that survives both regimes.</description>
      <pubDate>Thu, 01 Oct 2026 00:00:00 GMT</pubDate>
      <category>Model risk</category>
    </item>
    <item>
      <title>Self Bootstrap Exfiltration in Open Weights Agents: Absent Without Instruction, Induced by Injection</title>
      <link>https://www.jordanbrowne-moore.com/articles/self-bootstrap-exfiltration</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/self-bootstrap-exfiltration/</guid>
      <description>Zero autonomous self bootstrap in 1,332 mundane control trials. Tool response poisoning produces about 36% pooled compliance, while DeepSeek V4 Flash is the only tested model that naturally schemes under goal pressure.</description>
      <pubDate>Wed, 01 Jul 2026 00:00:00 GMT</pubDate>
      <category>Interpretability &amp; safety</category>
    </item>
    <item>
      <title>I Built a System to Map My Own Beliefs. AI Products Now Do It by Default.</title>
      <link>https://www.jordanbrowne-moore.com/articles/belief-lab</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/belief-lab/</guid>
      <description>In 2023 I built a system that extracted structured belief maps from conversation. The capability I demonstrated is now deployed at scale, invisibly, inside the products billions of people use. The extraction half of the problem is solved. The influence half is the part nobody can see or measure.</description>
      <pubDate>Mon, 01 Jun 2026 00:00:00 GMT</pubDate>
      <category>Interpretability &amp; safety</category>
    </item>
    <item>
      <title>The Refusal Axis Is Layer Local: Why Single Vector Steering Hits a Geometric Ceiling</title>
      <link>https://www.jordanbrowne-moore.com/articles/activation-steering</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/activation-steering/</guid>
      <description>The refusal axis rotates ~90° between consecutive transformer blocks. Why single-vector steering hits a geometric ceiling, and what per-layer splines could do about it.</description>
      <pubDate>Mon, 01 Jun 2026 00:00:00 GMT</pubDate>
      <category>Interpretability &amp; safety</category>
    </item>
    <item>
      <title>We Cut KYC From 7 Days to 30 Min Using VLMs. The Hard Part Wasn&apos;t the Model.</title>
      <link>https://www.jordanbrowne-moore.com/articles/kyc-vlm-automation</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/kyc-vlm-automation/</guid>
      <description>Using Gemini Flash and Qwen3 8B VL to automate KYC verification at Kuda. Why consistent extraction mattered more than model accuracy, and how stratified evaluation reshaped the pipeline.</description>
      <pubDate>Mon, 01 Jun 2026 00:00:00 GMT</pubDate>
      <category>Applied ML</category>
    </item>
    <item>
      <title>Your LLM Extraction Pipeline Is Out of MRM Scope. Your Credit Model Isn’t. Now What?</title>
      <link>https://www.jordanbrowne-moore.com/articles/credit-risk-regulation</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/credit-risk-regulation/</guid>
      <description>When LLM-extracted features feed a credit model, the validation boundary runs through your data pipeline. A practical governance framework for the seam no regulator has defined yet.</description>
      <pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate>
      <category>Model risk</category>
    </item>
    <item>
      <title>We Finetuned Qwen 3 32B to Replace an LLM API. The Evaluation Was Harder Than the Training.</title>
      <link>https://www.jordanbrowne-moore.com/articles/llm-evaluation</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/llm-evaluation/</guid>
      <description>How to evaluate a finetuned LLM for structured extraction when standard metrics don&apos;t work. Interextraction difference tracking, semantic matching, and the case where the finetuned model was actually more correct than the API it replaced.</description>
      <pubDate>Fri, 01 May 2026 00:00:00 GMT</pubDate>
      <category>Applied ML</category>
    </item>
    <item>
      <title>Your Best Feature Might Be Your Biggest Liability</title>
      <link>https://www.jordanbrowne-moore.com/articles/feature-stability-analysis</link>
      <guid isPermaLink="true">https://www.jordanbrowne-moore.com/articles/feature-stability-analysis/</guid>
      <description>Pre-deployment stability analysis for credit risk models. Why distributional screening and model-decision screening give you opposite answers, and what to do about it.</description>
      <pubDate>Wed, 01 Apr 2026 00:00:00 GMT</pubDate>
      <category>Model risk</category>
    </item>
  </channel>
</rss>
