Coming July 30, 2026

Make Loops,
Not Prompts

A practical, evidence-based model for designing AI agents that check their own work — so reliability comes from the system, not from buying a bigger model.

Preregistered · Open · Free for everyone

67–88 % Lower cost

per correct result, by switching to smaller models that self-verify

~75 % Fewer errors

through structured, rule-based verification — no AI judges

+32 pp More reliability

success rate gain, even making small models outperform flagship AI

Less energy

cheaper models × fewer wasted runs = a structurally smaller footprint

--Days
--Hours
--Minutes
--Seconds
Early access