<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Fieldframe Labs</title><description>Independent AI behavior research and reasoning-governance infrastructure. Home of FF-STACK, Cade, Crucible, Foundry, and Cortex.</description><link>https://fieldframelabs.ai/</link><item><title>The Research Behind the HLE Score: A Year of AI Behavior Research</title><link>https://fieldframelabs.ai/posts/research/</link><guid isPermaLink="true">https://fieldframelabs.ai/posts/research/</guid><description>The methodology behind the agent, the failure modes it catches, the products that came out of the same research moat, and where the program goes next.</description><pubDate>Wed, 13 May 2026 12:00:00 GMT</pubDate></item><item><title>HLE Submission Methodology Paper — FF-STACK v8</title><link>https://fieldframelabs.ai/posts/hle-methodology/</link><guid isPermaLink="true">https://fieldframelabs.ai/posts/hle-methodology/</guid><description>Full methodology paper for the FF-STACK v8 HLE submission: architecture, filtering policy, calibration, cost, and disclosure.</description><pubDate>Wed, 13 May 2026 12:00:00 GMT</pubDate></item><item><title>51.85% on Humanity&apos;s Last Exam: How a Solo Researcher Built a Multi-Agent HLE Submission</title><link>https://fieldframelabs.ai/posts/hle/</link><guid isPermaLink="true">https://fieldframelabs.ai/posts/hle/</guid><description>1,119 out of 2,158 on canonical HLE. Single workstation, no GPU cluster, no fine-tuning. The architecture, the numbers, and what&apos;s next.</description><pubDate>Wed, 13 May 2026 12:00:00 GMT</pubDate></item></channel></rss>