Improve performance with evidence

Start with a metric and measurement conditions. An improvement must be reproducible while preserving correct behavior.

Requirements

Reduce catalog search time on an agreed query set while preserving result content and order. Define the metric, data and acceptable improvement before starting.

Workflow

Baseline → Bottleneck → Approval → Change → Measure again → Accept

This is a learning outline to adapt. It is not an importable route or a promise of a specific speedup.

Comparable conditions

Keep the same data and queries, record environment resources and background load, define cache warmup, repeat counts and result aggregation, and verify correctness independently.

One fast response does not establish a stable improvement.

Stage results

Stage	Result
Baseline	Metric, conditions, commands and repeated results.
Investigation	Confirmed bottleneck and a change plan.
Approval	Agreed improvement criteria and boundaries.
Implementation	A change that preserves comparability.
Verification	New measurements and correctness comparison.
Acceptance	Decision against agreed criteria.

Commands

The profile must declare measurement and validation commands. Make data preparation and readiness checks repeatable. A command stage can compare measurements while an agent interprets profiling results.

If a required measurement is unavailable, stop and explain why. An assumption must not replace a missing result.

Rework

Ask for specific evidence: identical warmup conditions, rare queries with empty results or checks that preserve ordering. Repeat measurements and checks affected by the change.

Continue with project commands and route creation.