Versioned methods
Each working specification exposes its inputs, steps, outputs, publication gate, limitations, sources, version, and change history.
A seven-stage record for moving from a real job and baseline to a measured follow-up outcome without hiding failures or human decisions.
A privacy boundary and evidence manifest for turning a private project record into a deliberately sanitized, reviewable public submission.
A suppression-first method for aggregating comparable, reviewed AI workflow outcomes across multiple independent organizations.
Start with a local project, freeze a baseline and evaluation set, and export the resulting record without sending private inputs to AI Amigos.
Start in the workbench →Use the published schema, document deviations, and send a non-sensitive correction or independent verification link.
Propose a review or replication →