Results
Reading the report
What every number in a load-test report means, how the app decides pass or fail, and what failing status codes like 429 or 503 are telling you.
The result
At the top of every report is the Result — the app's plain summary:
- Looks healthy — nothing worrying at this load.
- Usable, with caveats — read the lines underneath before you rely on it.
- Not production-ready at this load — something clearly failed.
- Blocked by a rate limit or firewall — most requests were refused on purpose (429 or 403). The numbers show that limit, not your API's speed.
Underneath, each part is judged on its own:
| Part | What it looks at |
|---|---|
| Errors | The share of failed requests, and what the most common failure means |
| Latency | How long 95% of requests took (p95) — not judged when most responses were errors |
| Stability | Whether response times grew from the start of the run to the end |
| Capacity | Signs of overload: timeouts, 5xx errors, requests that could not be sent, being refused |
| Checks | The share of answers that passed your checks |
Without your own goals, the app uses starting rules: fewer than 1% failures is fine and 5% fails; 95% of requests within 500 ms is fine and 2 seconds fails. Set your own as soon as you know what your API promises.
Key numbers
| Number | Technical name | What it means |
|---|---|---|
| Requests per second | throughput, RPS | How many answers per second, on average. The hint shows the busiest second. |
| Response time | p95 latency | 95 of every 100 requests finished within this. A better guide than the average, because people notice the slow ones. |
| Typical response time | p50, median | Half the requests were faster than this. |
| Slowest 1% | p99 | 1 in 100 requests took at least this long. |
| Slowest | max | The single slowest request. |
| Failed requests | error rate | Requests with no answer, or a status of 400 or higher. |
| Checks passed | checks rate | Answers that passed your checks. |
| Simulated users | VUs | The most users at once, and the average over the run. |
| Server time | TTFB | From sending to the first byte of the answer — mostly your API working. |
| Requests not sent | dropped iterations | In requests-per-second tests: requests that could not start on time. The API — or your computer — could not keep up. |
| User satisfaction | Apdex | 0 to 1: how many answers were fast (within your goal) or bearable (within 4×). |
Before / after
When you've run the same request before, the report compares this run with your reference run (or the previous one): a plain sentence, every measure marked better, worse, or about the same, and both runs on the same charts. See History and reference runs.
The charts
- Response time — p50, p95, and p99 second by second. A line that climbs means the API is slowing down.
- Requests per second — how much load was actually delivered.
- Responses by outcome — 2xx, 4xx, 5xx, and no answer, second by second. The moment the colours change is the moment trouble started.
- Where the time went — waiting for a connection, DNS, TCP, TLS, sending, server time, download. High connection times point at the network or your computer; high server time points at the API.
- Response time distribution and the heatmap — how response times were spread, not just their averages.
Views — arrange the report your way
The report is a set of widgets on a grid. Which widgets, and where, is a view. Pick one from the View dropdown above the report. The view that opens by default is marked ★.
| Built-in view | Shows |
|---|---|
| Overview | Key numbers, the Result, response time, requests per second, responses by outcome, the distribution, where the time went, goals, and test settings. Opens by default. |
| Developer | Response time with more percentiles, server and connection time, where the time went, the spread, responses by outcome, the heatmap, goals, and response codes. |
| Capacity | Load against response time, requests per second, simulated users, failed requests, and requests not sent — for finding the limit. |
| Executive | The Result, a few key numbers, the user satisfaction score, p95 response time, and goals. |
| By topic | The same story grouped under three headings: Speed, Reliability, and Load, each with big single numbers and charts. |
| Compare overview | Used on compares: every API side by side in a table, the results, and the charts overlaid. |
To make a view the default, open it and choose Make default.
Customize a view
Press Customize. The grid gets faint column guides and a toolbar:
- Add widget opens the widget list, grouped by Summary, Over time, Distribution, Correctness, Details, and Layout.
- Undo and Redo — or Ctrl+Z and Ctrl+Shift+Z (Ctrl+Y works too).
- Save as my view — built-in views can't be changed, so your changes become a new view; give it a name and choose Save view. On your own views the button is Save, and Save as new… keeps the original and saves a copy.
- Discard changes throws your edits away. With nothing changed, the button says Done.
- Delete view removes one of your own views. Reports are not affected — only the layout.
Each widget has three buttons on its title bar while you customize:
- Size — a width: Full width, Three-quarters, Two-thirds, Half, Third, or Quarter; and a height: Short, Medium, or Tall. Text and table widgets size their height to their content.
- Settings — the Title, Chart type (only the types that read correctly for that data), and what to draw: which percentiles, which numbers and in what order, which timing phases, the line under a heading.
- More — Duplicate, Move to top, Move to bottom, Remove.
The grid has 12 columns and widgets snap to it. Drag a widget by its title bar, or resize it from its sides or bottom. From the keyboard, click a widget and use the arrow keys to move it and Shift + arrows to resize it; Enter opens its settings, Delete removes it, Ctrl+D duplicates it.
Changes apply when you save. Your views are kept in views.json in your app-data folder and are offered on every report. The HTML export follows the view on screen.
Widgets
| Widget | Shows |
|---|---|
| Key numbers | The headline numbers — pick which ones and their order. As tiles or a table. |
| Single number | One number, big, with the change against your reference run. |
| Section heading | A title, with an optional line under it, that groups the widgets below it. |
| Result | Pass / warn / fail in plain words, with the reasons. |
| User satisfaction score | The Apdex score, 0 to 1. |
| Response time | Percentiles second by second — pick which to draw. Can show your goal as a line. |
| Requests per second | Throughput second by second, with the target rate in requests-per-second tests. |
| Simulated users | How many users were sending requests at each moment. |
| Failed requests | The share of failed requests, second by second. |
| Responses by outcome | 2xx, 4xx, 5xx, and network failures, second by second. |
| Data transfer | Bytes received and sent per second. |
| Server and connection time | Server time, TCP connect, TLS handshake, and waiting for a connection, second by second. |
| Requests not sent | Requests a requests-per-second test could not start, each second. |
| Checks passing | The share of checks passing, each second. |
| Response time spread | How many requests fell into each response-time band. |
| Response time over time, by speed | A heatmap of response-time bands over time; darker means more requests. |
| Response time distribution | Min, p50, p90, p95, p99, and max for the whole run. |
| Where the time went | The average split of a request: DNS, connect, TLS, server, download. |
| Load vs response time | Each second as a dot. Where the dots bend upward, the API starts to struggle. |
| Pass/fail goals | Each goal met or missed, and each check's pass rate. |
| Response codes | A count of every status code returned. |
| Test settings | Method, URL, auth type, load shape, times, and bytes. |
On a compare, only the widgets that work for several runs at once are offered — Responses by outcome, the heatmap, Pass/fail goals, and Test settings are left out. Some widgets need per-second data that runs made with older versions don't have.
What happened — the written debrief
The HTML export starts with What happened in this run: the load, the speed, errors and when they started, where the time went, and whether the API looked near its limit — followed by suggested next steps. It is written by fixed rules from the measured numbers, so the same run always gets the same words.
Failing status codes, in plain words
| Status | Means | Usually |
|---|---|---|
| no answer | Timeout, refused connection, or network problem | The API is down or overloaded, or a firewall drops the traffic |
| 400 | Bad request | Something in the request is wrong |
| 401 | Not signed in | Missing or expired token — see Run first |
| 403 | Forbidden | Refused on purpose, often by a firewall that blocks heavy traffic |
| 404 | Not found | Wrong address |
| 429 | Too many requests | A rate limit: the server or its CDN is slowing you down on purpose |
| 500 | Server error | The API crashed or hit a bug — look at its logs |
| 502 / 504 | Bad gateway / gateway timeout | A proxy in front of the API couldn't reach it, or waited too long |
| 503 | Unavailable | The API is overloaded or down |
If failures start at a certain moment, the debrief says when, at what load, and whether they stopped when the load dropped. Failures that continue after the load drops are typical of a rate limit or firewall that blocks a caller for a while. See Troubleshooting.