Performance

When to use

Invoke after system-design has approved bounded contexts, service interactions, major data flows, and user-visible journeys, and before backend implementation, frontend rendering decisions, persistence tuning, infrastructure provisioning, or a scale event. Use it whenever user-visible latency matters, workloads may saturate, growth projections exist, cost-per-request matters, or scale posture changes materially.

Do not use for availability and disaster-recovery architecture (use reliability), telemetry pipeline design (use operations), one-off profiling of a production bottleneck, database query or GC tuning, CDN/vendor optimization flags, implementation-level micro-optimizations, or incident retrospectives.

Inputs

Required:

• Approved system-design.md and its relevant ADRs.
• The performance scope: user-visible journeys, externally observable workflows, and batch or background jobs whose timing is contractually or operationally meaningful.
• Expected load shape: peak and steady-state traffic, concurrency profile, and growth horizon.

Optional: