Make releasing boring
Shipping should be an ordinary event, not a decision. This quarter takes the fear out of deploying so the team can hold a weekly rhythm for the rest of this year — measured on cycle time and change failure rate, not on output.
Median pull-request cycle time 4.5 days → 1.5 days
How it is measured: Hours from first commit to merged, median across merged PRs
Baseline: 4.5 days · Target: 1.5 days
Initiatives
- Split the review queue by area and give each area a named first reviewer
- Cap open PRs per engineer at two and enforce it in standup
Change failure rate 18% → 6% of the changes we put in production
How it is measured: Production changes needing a fix or rollback, over all production changes
Baseline: 18% · Target: 6%
Initiatives
- Add a one-click rollback and rehearse it in a game day
Time to restore service after a Sev-1, 6 hours → 45 minutes
How it is measured: Median minutes from Sev-1 declared to service healthy
Baseline: 6 hours · Target: 45 minutes
Initiatives
- Write a one-page runbook for the three systems that page us most