Skip to main content

Handle deviation detection

A deviation is a recorded non-blocking event. The orchestrator increments current_cycle_deviation_count when a deviation is acknowledged.

Steps​

  1. Identify the deviation: read the latest findings row or the orchestrator's diagnostic message.
  2. Classify the deviation:
    • Bounded: known, expected, non-blocking. Acknowledge and continue.
    • Unbounded: unexpected, blocking. Open a repair goal.
  3. Acknowledge the deviation in the current goal's evidence.
  4. If a repair goal is needed, copy a REPAIR_GOAL template and dispatch.
  5. The orchestrator increments the deviation count automatically.

Examples of bounded deviations​

  • PR15 historical reconciliation (pre-existing).
  • A non-blocking audit finding acknowledged in evidence.
  • A bounded repair against a frozen surface.

Examples of unbounded deviations​

  • A direct push to main.
  • A force push.
  • A retroactive authorization.
  • A skipped predecessor FROZEN_PASS.

What this playbook is NOT​

  • It does not authorize the deviation retroactively.
  • It does not decrement the deviation count.
  • It does not edit a past findings row.