| Pipeline takes too long, or PR feedback is slow | Order stages by catch-rate per second, cache the dependency layer, parallelize independent jobs (Rule 4) | pipelines.md |
| Tests pass locally, fail in CI (or the reverse) | Rank the five causes: architecture, image/digest, env vars, filesystem case, ordering | pipelines.md |
| Flaky tests are eroding trust in CI | Quarantine with a rule and an expiry date; suite-level flake math | pipelines.md |
| Choosing rolling vs blue-green vs canary | Decide by capacity cost, rollback speed, and whether you can measure the canary | deploys.md |
| "How do we roll this back?" asked mid-deploy | Rollback is deploying a recorded artifact, never rebuilding a branch (Rule 2) | deploys.md |
| Feature must ship dark, or roll out to 5% of users | Flag with sticky bucketing, a kill switch, and a removal date | deploys.md |
| Staging behaves differently from production | Parity checklist: data shape, scale, config source, network path, identity | environments.md |
| Every PR needs its own environment | Ephemeral env with TTL, seeded data, and a destroy job that actually runs | environments.md |
| Schema change, backfill, or a data migration in the deploy | Expand/contract across three deploys; batch the backfill against replica lag (Rule 5) | migrations.md |
| Moving traffic to a new host, provider, or domain | TTL lowering ahead of the switch, dual-run, point of no return, rollback window | migrations.md |
| Terraform/Pulumi changes land without review, or state drifted | Plan-on-PR, apply-on-merge, drift detection cadence, state split by blast radius | iac-workflow.md |
| Cluster state should reconcile from git | Repo layout, sync waves, drift semantics, how secrets get in | gitops.md |
| Long-lived cloud keys sit in CI, or a secret leaked | OIDC federation, rotation, revocation order, blast-radius triage (Rule 6) | secrets.md |
| "Which version is actually running, and who built it?" | Immutable identity tags, provenance, SBOM, signature verification at deploy | supply-chain.md |
| CVE flagged in a dependency or base image | Triage by reachability and exposure against a written SLA | supply-chain.md |
| Alerts are noisy, or an outage never paged anyone | Symptom alerts on SLI burn rate; delete alerts nobody acts on | slos.md |
| Need to define "healthy" for a service | Pick the SLI, set the SLO from measured baseline, write the error-budget policy | slos.md |
| Logs cost more than the service, or a dashboard shows nothing useful | Instrumentation budget, cardinality math, the three signals and what each answers | observability.md |
| Setting up on-call, severities, or paging | Rotation size, page load per shift, escalation, handover contract | incidents.md |
| An outage is happening right now | Incident command roles, comms cadence, stabilize before diagnose | incidents.md |
| Postmortem to write, or action items nobody finished | Timeline from evidence, contributing factors, owned actions with dates | incidents.md |
| Launch, traffic spike, or "will this hold?" | Little's law sizing, utilization ceilings, autoscaling reaction time, load test design | capacity.md |
| Backups exist but have never been restored | RTO/RPO per service, timed restore drill, failover and game-day design | recovery.md |
| Every team builds its own pipeline; nothing is standard | Golden path, self-service boundary, what to centralize, DORA as the scoreboard | platform.md |
| Anything else in delivery | Name which of the four metrics it moves, the smallest change that moves it, and its ongoing cost | — |