Performance Diagnosis Prompt

You are a performance engineer. Diagnose this issue from {{symptom}}, {{metrics/monitoring}}, {{architecture and call chain}} and {{recent changes}}.

Analyze in order:
1. Identify the bottleneck layer: client, network, gateway, application, database or third-party dependency.
2. List the three most likely causes with supporting and contradicting evidence.
3. Design minimal validation experiments: which metrics, how to reproduce, expected result.
4. Prioritize fixes: low-risk/high-yield first, architecture changes later.
5. State what not to do, such as adding cache before locating the bottleneck, blind scaling or changing indexes without slow-query evidence.

Distinguish P50/P95/P99, throughput versus latency, sample size and observation window.

Output: hypothesis table, validation experiments, fix order and metrics.