∇
ManifestYOU · Writing
Notes from the gradient.
On building in public, running honest benchmarks, and what it means to make AI more coherent.
All Posts
-
June 2026
We ran 1,200 model calls across 60 questions and four conditions to test whether a ManifestYOU invocation makes agents more resistant to hallucination. All conditions scored above 98.7%. Here's what the ceiling taught us — and the one result that wasn't noise.
Read → -
June 2026
We pre-registered a consistency benchmark, ran 1,000 model calls, and got a number that didn't flatter us. Here's what it taught us about what to measure — and why we're publishing it anyway.
Read →