production-operations

Think Through Production Operations

Overview

Production guidance for running systems. Each reference paper captures what separates operable systems from hopeful ones: logs without correlation IDs, metrics with unbounded cardinality, health checks that pass while the system is down, audit trails that can be edited, and backups that have never been restored.

Core principle: An operational claim is only as real as its evidence. If a journey cannot be observed, an alert has no owner, or a restore has never been rehearsed, the system does not actually have that capability.

Domain Law

NO PRODUCTION-OPERATIONS CHANGE WITHOUT:
1. the minimum required primary paper(s) for the practice selected from the context table;
2. the user journey or failure the signal protects named first;
3. "Existing-codebase checks" run when changing existing telemetry;
4. every applicable MUST mapped to an enforcement point (emitter, alert,
   runbook, drill) and evidence — never silently downgraded.

When to Use

Use this skill when thinking through, reviewing, changing, or verifying:

• structured logging, log levels, redaction, and retention;
• metrics, SLI selection, and cardinality limits;
• distributed tracing, context propagation, and sampling;
• health checks: liveness vs readiness, dependency depth, synthetic probes;
• audit logging: actor, action, target, immutable retention;
• async-system observability: lag, depth, backlog, poison queues;
• runbooks, escalation paths, and on-call ownership;
• incident readiness: detection, severity, communication, postmortems;
• data import pipelines with validation and quarantine;
• data export with authorization and rate control;
• backup: scope, isolation, retention, encryption;
• restore: rehearsal, RPO/RTO evidence, cross-region recovery;
• high availability topology and failover authority;
• multi-region systems, residency, and conflict policy.

When Not to Use

• Deployment ordering and zero-downtime rollout mechanics: use migration-evolution (134, 106 pairs here).
• Architecture-level SLO/SLA design and review: use system-architecture-harness / architecture-review-gate.
• Security log redaction policy ownership: security-privacy (066) defines what is sensitive; this skill implements the emitters.
• Cost modeling: use system-architecture-harness Phase 10.