About this guide
- Written for
- QA leads and test managers · Security and compliance reviewers · People evaluating or buying tools
What test evidence is for
A pass or a fail is a claim. Evidence is what lets three different people check that claim: the developer who has to fix a failure, the QA lead who has to decide whether it is a real regression or a flaky run, and the reviewer or auditor who has to trust a result they did not watch.
Reproducible test results start from the same place: if the record says which element was used, which requests went out and what came back, a second run can be compared step by step instead of by impression.
The seven files a recorded session produces
Each file can be downloaded on its own from the recording page, as JSON. One button downloads the whole set as a ZIP bundle: the seven files, the screenshots, the generated code and a scrubbed copy of the sign-in state, so a reviewer gets one archive rather than seven downloads.
- Element details: every element the flow touched, with typed and categorised selectors for each.
- Network data: the API calls, WebSocket messages and GraphQL operations the flow caused, with their requests and responses.
- Performance data: Core Web Vitals, cache behaviour and resource timing for the pages visited.
- Security data: cookies, headers, authentication tokens, CORS and TLS details observed during the flow.
- UI elements: size, font, visibility and link-type details of what was on screen.
- Session metadata: the browser, the starting URL, timestamps, and the registry of IDs that links the other six.
- The rerun definition: the replayable step list, which points at elements and requests by ID.
Where the recordings live
The Recordings page lists every captured session with its steps, elements and network calls, a security-header score, and the runs it produced; a filter separates browser recordings from API-only ones, which have no browser launch and hold HTTP steps only. From the list you re-run or delete several recordings at once. Opening one shows the recorded steps in an editor (add, edit, reorder, disable or delete a step, insert a wait), the files the recording downloaded, the seven data files, the export menu and a visual playback that replays the steps in a browser on the server with a screenshot after each step.
Deleting a recording removes its runs, its download evidence and its place in any upload file it used; a file in Test Files that nothing else uses is released. The deletion is final, and the list asks before it deletes.
How the files point at each other
Elements, requests and cookies carry stable IDs such as elem_0001, req_0001 and cookie_001. A file that needs another file’s data refers to it by ID instead of copying it, so the security file can name the request that set a cookie, and a replay step can name the element it clicks.
The format only grows: new fields are added, existing fields and IDs are kept, so evidence recorded today stays readable by later versions of the platform.
What a failed step shows
That is usually enough to tell a changed page from a slow service from a broken test, which is the first decision in any failure analysis.
- The step and the element it targeted, with every locator that was available for that element.
- The network calls the step caused, by request ID, with their status and timing.
- The assertion and what was found instead.
- A screenshot of the page at that moment.
Evidence for heals and security findings
A rescued step keeps its own record: the locator that failed, the one that matched and the strategy that found it, shown in the run detail. A security finding is reported as confirmed only with the request that proved it, the response, and a proof that reproduces it, linked back to the recorded request it came from.
What is redacted, and what is never guessed
A recording captures what the browser sees, including what was typed. A login typed during a recording is kept, encrypted with AES-256-GCM, so the replay can sign in, and passwords are hidden in recording views, exports and generated test code. Make recordings with test accounts and synthetic data.
When a value cannot be determined from the browser, the field is left empty rather than estimated. Evidence with a plausible guess in it is worse than evidence with a gap, because the gap is visible.
Limits
- Evidence shows what happened in the flows that were recorded. It does not show that untested paths work.
- A screenshot and a request log explain a failure; they do not fix it, and a judgement about whether the page is wrong or the test is wrong is still a person’s.
Questions people ask about this
- What is test evidence?
- The record that lets someone check what a test did and why it passed or failed without running it again: the elements used, the requests made and their responses, the assertion results and screenshots, linked to one another.
- Can I get the files out of the platform?
- Yes. Each of the seven files downloads from the recording page as JSON, and a recording also exports as Playwright, Selenium, Robot Framework or pytest code.
Source: /documentation/test-evidence, updated 2026-11-16. Answers are checked against the shipped product on that date.
Found something out of date? The people who build the module answer the contact form on the home page, and corrections to this guide are dated above. Tell us.
This guide is about the test evidence every run keeps, and what it does not show. Twenty tools, including ours, are compared by job.