Skip to content

Notes on the scoring curves

lead-time

Deploys wait on review, not on build.

Symptoms

Finished work sits in a queue. The wait is what people feel; the pipeline is not.

Signals & measures

  • Median review wait is 14h of a 19h lead time — 74% of elapsed time
  • Three approvers required on every change, regardless of size

wait-time

Nearly three quarters of elapsed delivery time is a review queue.

Symptoms

Finished work is not moving, and the team cannot clear the queue itself. The same delay is raised in retros without an owner, so people experience it as not being heard.

Signals & measures

  • waitTimePct: 74 is derived from 14h median review wait ÷ 19h median lead time across the dashboard's 90-day window
  • Boundary is commit to production; review wait is the only waiting state available in the dashboard, so this is a partial view rather than full flow efficiency
  • Confirmed against three of the last four retros; no correction factor applied

deploy-freq

The rate is a policy outcome, not a capability limit.

Symptoms

Read this next to lead time. The pipeline completes in under an hour; the gap between that and 0.4 deploys a day is queueing, not build capability. Shipping less often than they can is experienced as stuckness, not as a careful release policy.

change-fail

A low failure rate on a low change rate is not the same result as a low failure rate on a high one.

Symptoms

Read this one beside deployment frequency. The rate looks healthy, and part of why is that the risky work is not being attempted — large changes are batched until they can be approved together. A quiet failure number here is not the same as confidence that a change will land.

Signals & measures

  • Assumes the dashboard reports change-failure rate as a fraction (0.12 entered as 12%). Unlabelled on the dashboard; to be confirmed with the platform team
  • Same dashboard: 0.4 deploys/day, so the failure rate is on a low change rate, not a high one

recovery-time-cqa

Recovery depends on who is on call, so the spread matters more than the median.

Symptoms

The night goes well or badly depending on who answered. That is the opposite of calm: the rota is a lottery over knowledge that was never written down.

Signals & measures

  • Assumes MTTR is reported in minutes; 45 entered as 0.75h
  • No runbook for the two most frequent alerts, so time-to-diagnose dominates time-to-fix

recovery-time

Same reading, service-quality lens.

Symptoms

Recovery time feeds two signals, so it needs a note under each curve to appear under both. This is the QoS one; recovery-time-cqa carries the change-quality one. From inside the organisation it is the same night: wait to find out who is on call, then wait while they diagnose.

medhigh-incidents

Six a month across three teams is two per team, which is where this curve turns steep.

Symptoms

Customer-affecting incidents at this rate leave little room between pages. The load is felt as always being on, even when the median recovery time looks contained.

Signals & measures

  • teamsCount: 3 supplied, so the per-team division is right. Without it the group would be scored as one team and read far worse