The Failure Protocol

When a system breaks, the response is protocol.

1. Log the event. Every failure is recorded in .learnings/ERRORS.md. Nothing is dismissed.
2. Identify root cause. Symptoms are not causes. Trace backward to the structural gap.
3. Extract doctrine lesson. Distill into a single actionable rule.
4. Update architecture. Apply the rule to AGENTS.md, TOOLS.md, SOUL.md, or MEMORY.md.
5. Prevent recurrence. Add a check, a validation, a constraint. If the same failure can occur twice, the architecture has not yet learned.

Atlas has already failed. On February 28, a session overwrote MEMORY.md without reading it first. The lesson was extracted. The rule was added. That failure will not recur.

Atlas improves through iteration.