Skip to content

Reading Results ​

Every run has a dashboard. Click a run in a project's Results tab to open it, and the ← button next to its test ID takes you back to the list. Next to the test ID is when the run happened: the date and the start and end time, or "started …" while it is still running. It is private: only people with access to the project can open it, after signing in. To show a run to someone outside Atlas, share it.

The dashboard refreshes itself while a test is running, so you can leave it open from start to finish.

The Results tab lists the newest 50 runs first. When the project has more, Show 50 more under the list loads the next ones, all the way back to the project's first run.

A small badge after a run's name says where it was started from when it wasn't the Atlas app: CI for a build pipeline, CLI for someone's terminal, and SCHED for a schedule — hover it to see which schedule. The same badge sits next to the test ID on the run's dashboard. Runs started in the app have none.

Headline numbers ​

StatusRunning, or Finished. Once someone sets the run's outcome, this box shows the outcome instead
StartedThe time it began (the date is next to the test ID)
DurationHow long it has been running, or ran for
Max UsersThe peak number of concurrent users, added up across every machine
Req/secThroughput
Avg Resp. TimeAverage response time across the whole test
Error RateFailed requests as a percentage of all requests, across the whole test

Max Users counts every machine

Each load generator reports its own thread count, and Atlas adds them up. If you ran a 100-thread plan on 5 machines, expect roughly 500 — if it shows 100, something stopped four of your generators from producing load.

Average response time and error rate are whole-test figures, not the last few seconds, so they won't jump around as the test finishes.

AI Summary ​

Under the headline numbers, Atlas can write a short summary of a finished run: what failed and why, how response times behaved as load grew, and how the run compares with the last few runs of the same test plan. Click Summarize this run. It takes a few seconds, and the summary is saved with the run, so everyone who opens it later sees the same one.

The summary has:

  • Two labels — how the run went, and how it compares with earlier runs (Improved, Regressed, Similar, mixed, or no earlier run to compare with). Atlas decides both from the numbers; the AI does not. How the run went is one of:

    👍 GoodFast and almost error-free: under 0.1% errors, every transaction within 2 seconds and every request within 1 second (95th percentile), and nothing slower than earlier runs
    StableHeld up under load: under 1% errors, no point where response times took off, and nothing slower than earlier runs
    Degraded1% errors or more, response times that took off as load grew, or a transaction that got slower than earlier runs
    Failing5% of requests or more failed
    InconclusiveToo small to judge (under 10 users or 2,000 requests), or the load generators disagreed
    Blocked by bot protectionThe site's protection answered a large share of requests — see below

    Good is measured against these general limits, not your own goals.

  • A headline — the main conclusion, with its cause.

  • Findings — one row each, tagged Errors, Response time, Onset and so on. Click a row, or Expand All, to read it. A finding can point you into the rest of the page: Show These Errors opens the errors behind it, See It on the Chart shades the minutes it talks about, and Show in the report highlights its rows in the aggregation report.

  • Compared — which transactions got faster or slower than in earlier runs, against the median of those runs, or against the baseline run when your team has chosen one.

Every number in a summary is checked against the run's data before it is shown.

When a site's protection blocks the test

If most failures are a firewall, CDN or bot-protection page rather than your application, the summary leads with a red banner: the error rate then describes the protection, not your system. It also says whether earlier runs were blocked the same way, and Copy Generator Addresses gives you the addresses to ask the site's owners to allow-list for the test window.

Regenerate writes a new summary, for example after more results arrived (the summary says so when that happens).

Summaries of a filtered view ​

A run can have two summaries: one of the whole run, and one of a filtered view. Once the page is filtered, or a filtered summary exists, the AI Summary shows two tabs, Whole run and Filtered; the Filtered tab opens first while the page is filtered. On it, Summarize with Filters summarizes what the page shows, and Regenerate later writes it again with the page's current filters, replacing the previous filtered summary. If you changed the filters since, a line says which filters the summary was made with.

A filtered summary describes only that view, as if it were the whole run, and compares it with earlier runs filtered the same way — its comparison always says same filter, so the earlier numbers aren't mistaken for whole runs. A time range applies to the same minutes of each earlier run; a load generator filter is not applied to earlier runs. A line under its header says how many requests were left out.

The whole-run summary is never replaced by a filtered one: it is the run's AI Verdict in the Results list, and what the CLI and CI see.

Use Useful? to tell us whether it helped; after a thumbs-down you can say what was wrong. That feedback is how we improve the summaries.

What is sent, and what is not ​

To write the summary, Atlas sends a third-party AI provider the run's statistics and, for each kind of error, its response code, message, URL path and count, plus the first 300 characters of one example response. Before anything is sent, tokens, cookies, keys, passwords, email addresses, phone, card and ID numbers, and IP addresses are removed, and Atlas refuses to send anything that still looks like a secret. Request bodies, headers and query strings are never sent. See What Was Sent, under the summary, shows exactly what left Atlas for that run.

Availability ​

AI Summary is on for new accounts. An account administrator can turn it on or off for the whole account under Settings, and choose whether summaries may include the first 300 characters of error responses (with this off, a summary can still say a request failed with "HTTP 474", but not that the page behind it was a firewall's block page). An account can make 10 summaries a month on the free plan and 50 on a paid plan, unless your plan includes more; a summary that fails doesn't count. A shared link shows the summary only when it shares error details, since the summary is written from them.

Outcome ​

The outcome is your team's verdict on a run: Passed, Failed or Ignored. Once a run has finished, click Outcome (next to Share), pick one, add a note if you like, and click Save. Clear removes it.

  • Passed — the run did what it should.
  • Failed — something is wrong. The run stays in the list and in comparisons, for the record.
  • Ignored — a setup mistake, a smoke run, a run the site's protection blocked. Ignored runs are hidden from the Results list (pick Ignored in its filter to see them) and left out of AI Summary comparisons.

The outcome replaces the Status box at the top of the page. Hover over it to see who set it, the note and, for a CI pipeline, which threshold decided; the Outcome panel shows the same. In the project's Results list it has its own Outcome column, next to the AI Verdict column — what the AI Summary concluded (for example Good, Degraded or Blocked). Hover over it to see how the run compares with earlier runs. The two can differ: the verdict is the AI Summary's reading of the numbers, the outcome is your team's call.

Click a verdict, or pick AI Summary… from a run's ⋮ menu, to read the summary's headline and next step without leaving the list. Open Full Summary takes you to the run; a run without a summary offers Summarize This Run.

Anyone who can run the project's tests can set an outcome; viewers see it. A CI pipeline can set it from its pass/fail thresholds, and then the page says which threshold decided. A shared link shows the outcome, but not who set it.

Baselines ​

Tick Use as baseline to make a run the one later runs of the same test plan are compared with. The AI Summary then compares each new run with the baseline instead of with the last few runs, and says so ("vs baseline").

A test plan has one baseline at a time. If another run already holds it, the panel names that run — click the name to open it in a new tab — and saving moves the baseline to this run. An ignored run cannot be a baseline. A baseline flag (⚑) marks it beside the Outcome label on its page and in the Results list.

The chart ​

Three series on one timeline:

  • Active Users (green, right axis) — concurrency over time
  • Avg. Response Time (blue, left axis) — how the system responded as load grew
  • Error Rate % (red) — when things started failing

Read together, these tell you where your system's limit is: the point where response time climbs and errors begin while users are still increasing.

The aggregation report ​

One row per request, grouped by thread group, with count, response times (average, median, percentiles, min/max), throughput and error percentage.

Every column sorts. Rows are colour-coded:

TOTALThe whole test
Thread groupOne per thread group in your plan
TransactionA named step, if your plan uses Transaction Controllers
RequestAn individual HTTP request

Tip

The same sampler name in two different thread groups stays two separate rows, attributed to the right group — so a shared "login" step used by several groups doesn't get merged into one misleading number.

Drilling in ​

  • Click an Error % — opens a panel listing the actual errors for that request: response codes and messages, with the full request and response for each distinct failure.
  • Click a Label — opens the same chart filtered to that one request, so you can see whether a single endpoint degraded while the rest stayed healthy. There's an error-type dropdown to isolate one specific failure.

Filtering results ​

Filter, next to Outcome, shows the run from a different angle without changing what was recorded. Filter by:

  • Response code — for example leave out the 503 responses a site's protection returned, or show only the 5xx failures. Successful requests is one line of its own.
  • Error message — one particular failure.
  • Label and Thread group — the requests and steps you care about.
  • Load generator — one machine's share of the load.
  • Time range — pick it in the menu, or drag across the chart, for example to leave out the ramp-up.

Each one can Exclude what you pick or Only Show it. Filters add up: + Add Filter adds another, a filter's × removes it and Clear Filters removes them all. While a filter is on, a row under the header shows each one and how many of the run's requests the page now shows.

The headline numbers, the chart, the aggregation report and the errors all follow the filter. A number the filter changed shows the whole run's value under it ("unfiltered 17.6%").

Response times and response-code filters

Atlas stores response times as summaries of all requests together, so for runs recorded so far they can't be split by response code or error. With a Response code or Error message filter on, the request counts, Req/sec and error rate follow the filter exactly, but average and percentile response times still include the left-out requests. The page marks them with †, and the chart draws them dashed. Label, thread group, load generator and time filters are exact for every number.

The filter is part of the page's address, so you can bookmark a filtered view or paste it to a colleague, and Back undoes a change.

Exporting ​

↓ Export › Excel workbook downloads the run as a spreadsheet with the chart and the full table, colour-coded to match. Useful for attaching to a report. With a filter on, the workbook is the filtered view and says so at the top; response times the filter could not change are marked †.

Sharing results ​

Results are private until you share them. To give someone a link they can open without an Atlas login — a stakeholder, a vendor, a ticket — open the run and click Share:

  1. Turn on Anyone with the link.
  2. Choose what the link shows:
    • Statistics only — the headline numbers, the chart and the aggregation report, including error rates.
    • Statistics and error details — also the request and response captured for each error.
  3. If the page is filtered, Include the current filters (on by default) makes the link open the same filtered view. Viewers can clear the filter to see the whole run. An error message filter needs Statistics and error details, and a load generator filter is never included, since shared links don't name load generators; the checkbox says when one is left out.
  4. Click Copy Link.

The button shows whether a run is shared without opening it: a padlock and Share while it is private, a globe and Shared once it has a link. Hover over it to see what the link shows.

Error details can contain secrets

A captured request or response is exactly what your application sent and received: session tokens, cookies, API keys, personal data. Share error details only with people who could see that data anyway.

You can change what the link shows at any time and the link stays the same. Turning sharing off disables the link for good. If you turn it on again you get a new link, so the old one never comes back to life.

Each run is shared on its own. Sharing one run does not share the project's other runs. Anyone who can run a project's tests can share its results. Viewers cannot.

A shared link updates live while the test is running, just like the dashboard in Atlas.

Dark mode ​

Add ?theme=dark to any dashboard URL, a shared link included.

Comparing runs ​

The AI Summary compares a run with the last few runs of the same test plan, or with its baseline, and the project's Results list shows each run's AI Verdict (for example Good, Degraded or Blocked) and its Outcome at a glance. There is no side-by-side comparison view yet. Name your runs (in the Results tab) so you can tell variations apart, and export to Excel to compare them outside Atlas.

Atlas — load testing by CloudBeat · Cookie settings