Project Relay Keeps a Record of AI Disagreement

Field notes

Relay organizes multi-model work around a canonical evidence record rather than a conversational transcript. Its stated sequence fixes the question and evidence boundary first, compares attributed work independently, then preserves the reasoning behind a human decision. The public status board separates local implementation from gated foundations, including R19 authenticator and mobile approval work, while hosted tenancy and external operations remain disabled. The project also describes projections into Notion, Slack, and GitHub as rebuildable views rather than competing sources of authority.