"The API feels slow" is not a metric
You have a number from APM or the network tab — say 1,200 ms — but is that bad? It depends on your budget. A 200 ms internal call and a 1,200 ms user-facing call mean different things.
Put the number next to a budget
Use the Latency Diagnostic: enter the measured time and your threshold (e.g., 800 ms). It classifies the result as healthy, slow, or critical and lists the first places to look — upstream dependency, unindexed query, missing cache, or an early client timeout.
Worked example
Measured 1,200 ms against an 800 ms budget → "Critical: more than double." The checklist points at a downstream DB query and a missing index. After indexing, re-measure; if it drops under budget, the verdict flips to healthy.
Boundaries
- The tool compares numbers; it does not measure the network. Get the raw figure from devtools or APM first.
- Watch p95, not the average — the average hides the slow tail that users actually feel.
FAQ
What budget should I use? A common user-facing target is 200–400 ms for core calls; internal service calls are often tighter.
Status is 200 but slow? Latency and status are independent — pair this with the Status Lookup to confirm the request itself succeeded.