Change Data Capture Pipeline  ·  View 03 of 21  ·  People and journeys

Actors and Their Journeys

Four human questions and two machine ones; the rest of the set exists to answer them.

Editable source SVG draw.io All views
Product engineering Product engineer 40 engineers Goal — Ship my migration on Tuesday without being the person who broke analytics. Core journeys Ship a schema migration see view 05 Add a table to replication Service owner 9 services Goal — Know that nothing downstream can make my database slow down. Core journeys Read the source-impact budget Data platform Data engineer 6 engineers Goal — Point a new consumer at data that is already correct, in a day, not a quarter. Core journeys Wire up a new sink see view 04 Rebuild a sink from the log Platform SRE on call, 24×7 Goal — Know within a minute whether lag is a stall, a burst or a wrong number. Core journeys Chase a wrong number see view 06 Pause a table safely Consumers of the data Revenue analyst 120 people Goal — Trust today's number enough to put it in front of the board. Core journeys Read the mirror with an as-of Support agent 300 seats Goal — See the change the customer just made while they are still on the call. Core journeys Open a freshly updated record Machines in the cast Reconciliation job daily, tier-1 tables Goal — Find divergence before a human does. Core journeys Publish a divergence verdict Partner system 14 integrated Goal — Receive a change event once, in order, per record. Core journeys Consume a webhook fan-out Who the Pipeline Is For, and What They Get to Do Person or role Journey / task Security / platform External / third party Four different questions arrive at this platform; the set answers all four before it draws a box. v 1.0 · owner Data Platform Architecture · date 2026-10

Why these four

  • The product engineer never asked for this platform and can break it by doing their job correctly — so the design assumes no notice of a migration.
  • The data engineer's cost is the onboarding of the second and fifth sink, not the first.
  • The SRE needs to distinguish stale from wrong in under a minute, which is a telemetry requirement, not a dashboard preference.
  • The analyst's trust is the actual product: a number nobody will put in front of the board is not worth replicating.

Assumptions

  • 40 product engineers across 9 services; 6 data engineers; 120 analysts; 300 support seats (assumptions).
  • 14 partner systems consume the webhook fan-out.

What this rules out

  • A design that requires the application team to publish events itself — that is dual writes, and it fails silently (ADR-02).
  • A platform whose onboarding path runs through a ticket to the database owner.
◀  High-Level Architecture All views Journey — Wire Up a New Sink  ▶
Solution Architecture v1.0 · Google Cloud · Data Platform Architecture · 2026-10 · 21 views · 15 architecture decision records03-actors-and-journeys