The Gap
One of the most operationally useful concepts in resilience engineering is the distinction between:
-
Work-as-Imagined (WAI): The process as written in the runbook, the architecture diagram, the onboarding doc, or the manager's mental model. It is clean, linear, and logical. It assumes people follow steps in order, tools work as documented, and edge cases are handled by the official escalation path.
-
Work-as-Done (WAD): The messy reality of how engineers actually get the job done. It involves undocumented scripts, Slack messages that substitute for tickets, tribal knowledge held by two people, workarounds for tools that are broken in specific environments, and judgment calls made under time pressure.
Why the Gap Exists
WAI and WAD diverge because reality is more complex than any process document can capture. Engineers are intelligent and adaptive — when the official process doesn't fit the situation, they improvise. Most of the time, this improvisation is what keeps the system running. The gap between WAI and WAD is not a sign of a broken team; it is a sign of a team doing real work in a complex environment.
The Danger
Incidents often happen when management makes decisions based on WAI while the system actually runs on WAD.
- WAI assumption: "We can let Bob go; everything is documented in Confluence."
-
WAD reality: Bob manually restarts the Kafka consumer group every Tuesday morning because the auto-restart script has a race condition that nobody has had time to fix. This is not documented anywhere. It has never failed because Bob has never missed a Tuesday.
-
WAI assumption: "Our deployment pipeline is fully automated."
- WAD reality: The pipeline requires an engineer to manually approve a step in the UI because the approval API has been broken for four months and the ticket to fix it is in the backlog.
To build resilient systems, you must design for Work-as-Done. This means conducting workflow interviews with engineers ("walk me through what you actually do when X happens"), reviewing runbooks against reality, and treating undocumented workarounds as signals of process gaps — not as violations to be punished.