Skip to content

ALL-DEPL1001 · progressive_rollout

Judge whether a risky deployment verifies behavior before broad exposure.

This is a contextual rule for all languages. Read its implementation.

Ask the selected judgment backend for five independently cited rollout facts and reduce them through a fixed decision table. The model never chooses the final category. Compare change exposure, stage representativeness, attributable outcomes, decision thresholds, and recovery.

The frozen bundle cites rollout configuration, populations, signals, comparisons, thresholds, observation windows, and recovery actions. Missing, duplicate, conflicting, or uncited model answers remain unknown and reduce to uncertain.

Low-risk offline artifacts may not need progressive exposure when equivalent verification covers the complete risk.

A representative canary compared with a control and protected by rollback is verified. Sending traffic to an unrepresentative stage without recovery is unsafe.

  • Cites “The Site Reliability Workbook”, Canarying Releases
  • Cites “Accelerate”, continuous delivery
  • Cites “Release It”, stability patterns