Skip to content

Operate

Keep the system alive, visible, and honest. Operation is where discovery gets free evidence — if you look.

  • Clear ownership for each deploy unit (Work ownership).
  • Runbooks for the failures you already know; toil vs engineering — automate only what repeats and hurts.
  • Promote with the right play: merge-it · stage-it · ship-it.
  • Secrets and CI remain boring: repos (ESC + OIDC by default).

Instrument for the quality regime you actually run:

Regime Lean toward
A — deterministic compute Correctness SLIs, contracts, pipeline freshness
B — interactive products Journeys, Web Vitals / RUM, errors that block jobs
C — generative / high-input Traces, datasets, scores — house default Langfuse

Start with observe-it on the critical path. Don’t invent a full APM estate on day one.

Production signal, support, scores, and incidents are product evidence, not just dashboards. Feed them back into Discover: what to pretotype next, what fidelity you actually earned, what debt is charging interest (Bugs & debt).

  • Changing code under load → Maintain
  • Killing or replacing a system → Retire