Your averagehides the tail.

186ms avg - green check. 1 in 20 past 800ms at p95. The slowest 1% in seconds. The dashboard never flinched.

Run the endpoint. Read the tail. p50, p95, p99 from one sequential pass. Web for smoke tests · npx @delayt/cli for 50+ runs and --assert-p95 in CI.

↳ add -u and --assert-p95 to your pipeline

// trace · example CLI output

// example receiptrun_ocln34nl · 50 requests

endpointGET api.example.com/v1/resource

requests50 sequential · ~12s wall time

avg186ms ← smooths over outliers

p50143ms ← median request

p95812ms ← 5% of requests slower

p992.31s ← slowest 1% of requests

errors0% · all 2xx in this sample

verdictCRITICAL: example budget p99 < 800ms (with --assert-p99=800)

* fictional sample, not a live run

// distribution

50 requests bucketed by response time. The long tail on the right is what an average silently smooths away.

0ms· p50 143ms· p95 812ms3s+

* fictional sample, not a live run

// manifesto

Sequential latency checks, not load tests. Print p50, p95, and p99 from real HTTP responses. Pick your own thresholds; we don't pretend 200ms fits every API.

- no signup on the hosted web app
- browser UI for quick runs (up to 20 requests); CLI for 50–200 and CI
- CLI runs locally; auth headers stay in your terminal
- hosted web runs send HTTP from the server (see docs)
- exit code 1 when --assert-p95 / --assert-p99 fails
- MIT · npm install @delayt/cli