Skip to main content

Research

Research runs alongside the ladder, not as a step within it: what we learn building harnesses, writing evals, and running the first dark factories and Company OS instances, published where we can. Proof, in public, that the methodology holds up.

Why we gate graduation on evals, not intuition

A harness that feels ready and a harness that's measured ready are different claims. Notes from building the first custom benchmarks: what makes a task worth testing, and why a passing score should be as boring as a build passing CI.

Tell us what you'd want proven first.

We reply within a day, from someone who will do the work.

hello@bakenlabs.com