Run a classifier in shadow mode before it can gate
Let a new AI judge watch silently until there's evidence it's right.
how it works
Log scores only; a hand rubric backtested at chance (AUC 0.514) and a tiny-sample refit had a wide CI, so thresholds wait for labels; contrast with fail-closed gates like agent-merge.
Technique: A pattern or hard-won lesson, written down so it isn't learned twice.
The code for this one lives in a private repo. This page explains the idea so you can build your own.