Connect a Postman API key to check your work
You do each step in Postman, then LiftOff uses your key to look at your workspace and confirm it's done. Connect once and it works for every module. Your key is saved in this browser and is only used to check your work in Postman.
One fast Send doesn’t prove an API is healthy under traffic. Use a short Functional runner pass, then learn why avg vs p95 change the decision — before you open the Performance tab.
Why averages can mislead
This step uses the Collection Runner in Functional mode — not the Performance tab yet. You’re collecting a handful of timings so variability is visible, then learning how Postman summarizes many timings with avg vs percentiles.
1. Run a short Functional collection run
- Confirm the Performance environment is selected in the top-right environment picker (the one you created in Step 3 — not the Performance runner).
- Select Coffee API Load Test → open the Collection Runner (⋯ → Run or Run on the tab bar).
- Stay on Functional (leave Performance alone for now).
- Set Iterations to
10, keep the Performance environment selected, and start the run. - When it finishes, skim the results: each iteration has its own response time. They won’t be identical — that’s normal. Write down the fastest and slowest times you see for Get Hot Coffee.
2. What the summary metrics mean
When Postman aggregates many requests under load, it reports more than a single average. From the performance metrics guide:
| Metric | Plain-language meaning |
|---|---|
| Avg. response time | All response times added up, divided by the count |
| P90 / P95 / P99 | The time at which 90% / 95% / 99% of requests were at or faster than this value — the rest were slower |
| Error % | Requests that failed to complete normally (timeouts, connection/TLS issues, uncaught script errors) |
| Failure % | Requests that completed but failed at least one pm.test() assertion |
P95 is the one teams watch most often: “95% of users were at or under this latency; 5% were worse.”
3. Worked example — why the average can look “fine”
Imagine 20 requests with these response times (ms):
180, 190, 200, 200, 210, 210, 220, 220, 230, 230, 240, 240, 250, 260, 280, 300, 900, 2100, 3200, 4800
| Summary | Value | What a reader might think |
|---|---|---|
| Average | ~720 ms | “Under a second — acceptable?” |
| Median (p50) | ~235 ms | “Most requests are snappy” |
| P95 | ~3200 ms | “1 in 20 requests takes ~3+ seconds” |
| P99 | ~4800 ms | “The worst few are almost 5 seconds” |
The average is pulled up by a few slow requests, but it still understates how bad the tail feels. If you only watched average, you might ship; if you watched p95, you’d investigate.
Here’s a compact “report card” in the same shape Postman shows after a performance run:
| Metric | Value |
|---|---|
| Average response time | 720 ms |
| Median | 235 ms |
| P95 | 3.2 s |
| P99 | 4.8 s |
| Error rate | 0% |
Ask: Would you call this API healthy based on the average alone? What decision changes when you look at p95?
Expected outcome: You’ve seen real timing variance across 10 Functional iterations, and you can explain in one sentence why p95 (or p99) catches problems that average misses.