Chaos Engineering Platform  ·  View 04 of 21  ·  People and journeys

Journey — Proving a Dependency Is Really Optional

A playback-api owner turns a post-incident action item into a recorded verdict, in six phases.

Editable source SVG draw.io All views
Service owner playback-api, tier 1 Goal — Show that playback still starts when the recommender is slow Trigger — A post-incident action item: "confirm the recommender is non-blocking" Done when — A recorded verdict, and either confidence or a tracked finding 1 · Write in the repo 2 · Rehearse dry run + staging 3 · Ask approval 4 · Inject ◆ moment of truth 5 · Read verdict 6 · Act fix or file What they do Copies a template States the hypothesis Runs dry run Runs it in staging Requests approval Watches the run Opens the verdict Files the finding Where Repo + CLI chaosctl CLI dry run Console approval card Live run view Evidence bundle Findings tracker How it feels Confident Fine Anxious Behind the glass Validator Selector resolver Radius calculator Approval workflow Notification Lease issuer Injection agent Verdict engine Findings tracker Where it hurts Hypothesis is guesswork Waiting on a human Fear of a real outage Fix competes with roadmap What would fix it Suggest SLIs from the SLO store Standing consent for proven definitions Visible abort countdown Finding linked to a tracked item How we know Rejected definitions 22% Dry run p95 5 s Approval wait 40 min Aborts 1 in 7 runs Inconclusive 6% Findings closed 61% User Journey — A Service Owner Proves a Dependency Is Really Optional The trough is the injection moment: the platform's whole job is to make that fear unjustified and still real. v 1.0 · owner Reliability Architecture · date 2026-09

The trough is the point

  • Confidence drops at approval and stays low through injection. That is the moment the whole architecture exists to make safe — and the platform's job is to make the fear unjustified without making it disappear.
  • The recovery only happens at the verdict, which is why an INCONCLUSIVE outcome is so expensive: the owner paid the fear and got nothing back.

What the set owes this journey

  • Phase 2 is answered by the dry-run resolver and radius calculator on views 07 and 14.
  • Phase 4's fear is answered by the abort paths on view 13 and the lease mechanics on views 12 and 21.
  • Phase 6 is answered by the findings tracker on views 11 and 19, which is the only thing that converts a verdict into a fix.

Assumptions

  • 22% of authored definitions are rejected at validation, 6% of runs land INCONCLUSIVE, and about one run in seven aborts. Stated assumptions, used to size the queue and the support load.
  • A 40-minute median approval wait is the number that justifies standing consent for proven definitions.
◀  Actors and Their Core Journeys All views Journey — Running the Quarterly Region Game Day  ▶
Solution Architecture v1.0 · Google Cloud · Reliability Architecture · 2026-09 · 21 views · 16 architecture decision records04-journey-prove-a-dependency-timeout