Sabitlenmiş Tweet

I built RatchetLab: an eval-locked ratchet for making agents measurably better.
Give an agent one editable artifact.
Protect the evaluator.
Run a fixed score.
Keep only measured wins.
Restore everything else.
v0.1 is a local Python CLI with doctor checks, proof receipts, offline demos, and CI.
github.com/stoicpickle/ra…
Feedback welcome!

English



















