Crumb Axiom: a toaster-alignment lab

Field notes

The site runs a deterministic preference test rather than a trained model. Three hidden-level comparisons narrow a 1–10 range, then a midpoint becomes the visitor’s inferred target; the displayed match uses 100 minus ten times the distance from that target. Runs stay in the browser, up to 20 at a time, and exports include the choices and score formula. Proposed token governance and on-chain proof submission are described as future work, not functioning features.