CI/CD Platform  ·  View 19 of 22  ·  Operations

The Run Lifecycle Loop

Seven steps, and the loop closes through observation rather than through a dashboard.

Editable source SVG draw.io All views
Trigger event, schedule, API Admit compile + classify Schedule fair queue Execute single-use sandbox Seal sign + store Gate allow or deny Observe flake, cost, ratio Run lifecycle run record job graph job + lease digest evidence outcome next definition CI/CD Platform — The Run Lifecycle Loop The loop closes through observation: flake rate and infra-failure ratio change the next definition and the next entitlement, not a dashboard. v 1.0 · owner Platform Engineering · date 2026-09

Why this is a loop and not a line

  • Observe feeds the next definition and the next entitlement. Flake rate changes a pipeline; infra-failure ratio changes capacity; cost per job-minute changes which tier a job belongs in.
  • A platform whose telemetry only produces dashboards has a line, not a loop, and will keep re-learning the same lesson about the same pipeline.

What is not on the ring

  • Retry is not a step. It re-enters at Schedule with the cause recorded, because a retry that looks like a new run makes the infra-failure ratio unmeasurable.
  • Rollback is not on this ring either; it belongs to the environment's lifecycle, drawn in view 17.