Reproduce the workload
Capture route, request shape, timing and concurrency. A global CPU number is not enough to explain a slow commerce journey.
Performance guide →Performance readiness
Optimize from evidence,
not from anxiety.
Performance work should protect transaction correctness and identify the real bottleneck—database, application, provider, network, asset or browser—before adding complexity.
Find the boundary
Separate application time, database time, provider wait time and frontend work before changing infrastructure.
Architecture →Keep transactions correct
Do not trade idempotency, validation or reconciliation away for superficial speed. Move expensive work off the critical path when safe.
Reliability →Know the before state
Record latency distribution, query evidence and provider timing before an optimization so the effect can be verified.
Keep the known-good path
After change, verify the same commerce journey and protected routes—not only the new code path.