Your change failure rate is probably wrong
Most teams count incidents, not failed changes — and the two diverge badly once you deploy more than once a day. How to define it so the number stays useful.
Read articleInsights from the pipeline
Field notes from real engagements — what worked, what we got wrong, and the numbers behind it. No vendor whitepapers, no thought leadership about digital transformation.
Featured guide
The four DORA metrics, and how to instrument them without a consultant
Deployment frequency, lead time for changes, change failure rate and time to restore. Everyone quotes them; far fewer teams measure them honestly. This guide walks through pulling each one from GitHub Actions and Argo CD, the definitions that trip people up, and the three ways a dashboard can flatter you into thinking things are fine.
dora-baseline.sh
# deployment frequency, last 30 days gh api "/repos/$REPO/deployments?environment=production" \ --paginate --jq '.[].created_at' \ | awk '$0 > start' start="$(date -v-30d +%F)" \ | wc -l # lead time: merge commit → production deploy ./scripts/lead-time.sh --env production --window 30d
Latest articles
Guides, postmortems and opinions
Where 40% of a cloud bill usually hides
Idle non-production, over-provisioned databases, log retention nobody chose, and cross-AZ traffic. The order we check them in, and what each one typically returns.
Read articleYou probably do not need a service mesh yet
Meshes solve real problems, but they add an operational surface most teams are not ready for. A decision checklist, and what to do at each stage instead.
Read articleMaking SOC 2 evidence a build artefact
If a control is enforced in the pipeline, the evidence can be generated by the pipeline. A practical mapping from common controls to automated checks.
Read articleTerraform modules that survive contact with a second team
Versioning, sane defaults, escape hatches and the interface design mistakes that force a rewrite eighteen months in.
Read articleCutting 4,000 alerts down to 180
A postmortem of an alert cleanup: how we decided what to delete, what we kept by accident, and the two outages that got through during the transition.
Read articleBuild caching that actually saves time
Most cache configurations are net-negative once you count restore time. How to measure whether yours is earning its keep.
Read articleGolden paths fail when they are mandatory
Adoption follows convenience, not policy. What we changed after a platform rollout that teams quietly routed around.
Read articleTreating prompts like code you can roll back
Versioning, evaluation gates and canary releases for LLM-backed features — using the delivery machinery you already have.
Read article
Monthly, and short
One email a month, no fluff
New guides, an interesting postmortem and one number from a real engagement. Unsubscribe in a click.
Rather have the analysis done on your stack?
The free two-week health check applies all of this to your own pipelines, infrastructure and cloud bill, and gives you a written report with prioritised fixes.