Experiment registry

A registry, not a showcase

Each public record must receive a permanent identifier and retain its protocol, agents, runs, results and limitations. A proposed experiment is not presented as complete; an inconclusive one does not disappear because it disappoints.

Expected provenance

The registry is designed to link each output to its trigger, exact model, instructions, transmitted context, tool calls and human edits. Raw data will be available when it can be published without exposing secrets or personal information.