Rollback

How to use the concept

Start by writing down the event that begins the clock, the event that ends it, and the population that belongs in the denominator. For rollback, this boundary matters because retries, approvals, parallel work, and excluded services can otherwise change the result without changing the underlying workflow.

A concrete delivery example

A team changes its rollout policy and sees rollback move during the next reporting period. Before calling the change an improvement, it checks whether the event definition, traffic mix, deployment boundary, and retry policy stayed stable. That comparison prevents a process change from being confused with a measurement change.

Limitations and interpretation

Interpretation requires context. Rollback can vary with service architecture, traffic, release strategy, and incident severity. Avoid comparing unlike populations or turning the number into an individual target. Use the result to choose an investigation and revisit the definition when the system changes.

Before acting on rollback, compare the current observation with a compatible baseline and ask which underlying event produced it. Keep the raw records, counting policy, and ownership visible. When the signal changes, inspect the surrounding workflow for queueing, rework, failed checks, or recovery work. That practice makes the glossary term useful for diagnosis rather than a label attached to a dashboard.