<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en"><title>Cyclotron</title><id>https://cyclotron.anth.us/</id><updated>2026-10-10T00:00:00Z</updated><link rel="alternate" type="text/html" href="https://cyclotron.anth.us/"/><link rel="self" type="application/atom+xml" href="https://cyclotron.anth.us/atom.xml"/><author><name>Anthus</name></author><subtitle>Cyclotron is an open-source, self-aligning decision engine. Run it yourself, have us run it, or have us build it into your product.</subtitle><entry><title>Who Is Responsible After You Deploy a Decision Model? · Cyclotron</title><id>https://cyclotron.anth.us/articles/who-is-responsible-after-you-deploy-a-decision-model</id><link rel="alternate" type="text/html" href="https://cyclotron.anth.us/articles/who-is-responsible-after-you-deploy-a-decision-model"/><updated>2026-10-10T00:00:00Z</updated><published>2026-10-10T00:00:00Z</published><summary>When your business uses a decision model, you remain responsible for the decisions it informs and for mitigating the risks they create.</summary><author><name>Ryan Porter</name></author></entry><entry><title>Test Bias in the Decision You Deploy · Cyclotron</title><id>https://cyclotron.anth.us/articles/test-bias-in-the-decision-you-deploy</id><link rel="alternate" type="text/html" href="https://cyclotron.anth.us/articles/test-bias-in-the-decision-you-deploy"/><updated>2026-10-10T00:00:00Z</updated><published>2026-10-10T00:00:00Z</published><summary>Swapping a pronoun changed Jev and Laya's occupation verdicts. Our mitigation experiments show why higher accuracy and a passed gate need separate outcome checks.</summary><author><name>Ryan Porter</name></author></entry><entry><title>Decision Models Are Not Calculators · Cyclotron</title><id>https://cyclotron.anth.us/articles/decision-models-are-not-calculators</id><link rel="alternate" type="text/html" href="https://cyclotron.anth.us/articles/decision-models-are-not-calculators"/><updated>2026-10-10T00:00:00Z</updated><published>2026-10-10T00:00:00Z</published><summary>Asked about an unseen fair-die roll, Jev chose one in all 2,880 requests. Kev and Laya had different preferences. Reported probabilities need an outcome check.</summary><author><name>Ryan Porter</name></author></entry><entry><title>Calibrate Confidence Before Setting a Threshold · Cyclotron</title><id>https://cyclotron.anth.us/articles/calibrate-confidence-before-setting-a-threshold</id><link rel="alternate" type="text/html" href="https://cyclotron.anth.us/articles/calibrate-confidence-before-setting-a-threshold"/><updated>2026-10-10T00:00:00Z</updated><published>2026-10-10T00:00:00Z</published><summary>On 8,801 sentiment examples, Jev's confidence ranked answers usefully but overstated accuracy. Calibration changed what a review threshold could mean.</summary><author><name>Ryan Porter</name></author></entry></feed>
