About this guide
- Written for
- QA leads and test managers · SDETs and automation engineers · Developers who own their tests
The list
The page opens with the totals (runs, passed, failed, running, pending) and the pass rate, then the runs newest first with the run id and the recording id. A run id can be renamed; several runs can be selected and deleted for good, after a confirmation that says what goes with them: run data, step results, trace data and screenshots. Test Runs is part of every plan.
A summary panel over the list is labelled by its source: "Rule-based summary (no model involved)" or "Generated by a local AI model, it can be wrong. Review before acting." The label is the point: nothing in it is a confirmed finding.
One run
The header names the recording, the browser and where the run executed; a run still going says so and fills in as it runs; a run that spent its whole time in the queue says that it never executed a step. Three actions sit at the top: Stop Test Run, Re-run Test and Download Screenshots Zip.
The step table has a row per step: the step, what happened, the screenshot and a Raise Defect action. A status is one of passed, failed, bug, slow, healed or skipped. When fewer steps executed than were recorded, the page says so with the three counts and the usual reason (execution stopped early after a failure), and a note explains how a failure is counted. A healed step reads "Original selector failed, found element using …" with the strategy and its confidence; the strategies are an accessibility match on role and name, an alternative locator from the recorded chain, fuzzy text, an XPath variant and a partial class match. A heal is marked auto-healed, pending review or rejected.
The trace
When the run was executed with tracing on, the trace viewer shows a timeline, the network calls (downloadable as a HAR file), the console lines, the DOM snapshots and a performance tab with total and average step time, the slowest steps and a self-healing report (steps healed, healing rate, confidence). A run executed without tracing says "Trace Data Not Available" and names the setting to turn on next time.
The observability strip above it counts what the replay met: overlays dismissed (cookie banners, chat widgets), bot walls it cannot solve on its own, credentials fetched from the vault, the fallback heal cascade, the last-resort heal tier and its confidence gate (a zero there is the gate refusing a wrong element, not a failure), audited heals waiting for a person in the review queue (not applied; the step failed until someone decides), the egress country requested and observed, locators written back to the recording for the next replay, HTTP steps and their assertions, visual checks, drags and keyboard chords.
When a run needs a person
A step the replay cannot pass on its own, a verification challenge for example, pauses the run with "Waiting for a person": solve it in the browser the run paused on, type the answer or the word solved, and resume. Only the run's owner, the team owner or a platform administrator can answer; anyone else is told so.
Re-run, defects and baselines
- Re-run with a configuration: playback speed from a quarter to four times, a network profile from slow 3G to offline (Chromium family), a reused browser profile, and failed steps only.
- Raise Defect copies a defect summary to your clipboard, ready to paste into your issue tracker.
- Approve as new baseline promotes this run's snapshot as the reference future visual checks compare against. It is tenant-scoped, recorded in the audit log with an optional comment, and applies at once.
- A silent screen recording of the run is kept beside the screenshots; the step table lists every action as text for anyone who cannot watch it.
- A run of an API-only recording launched no browser; its page shows each HTTP step with its request and response headers and bodies.
Questions people ask about this
- Why does a run show fewer steps than the recording?
- Execution stopped early, usually after a failed step. The page prints the recorded, executed and missing counts so the gap is visible, not hidden in a total.
- Why is there no trace on my run?
- The run was executed with tracing off. Turn it on in the execution configuration and re-run; the page says so in place of the trace.
- Is the healed step a pass?
- It is a step that found its element by another strategy. It is marked healed, with the strategy and confidence, so you can judge whether the right element was found. A heal held for review was not applied, and the step failed until someone decided.
Source: /documentation/test-runs, 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 regression testing, the job that suites, runs and planning serve. Twenty tools, including ours, are compared by job.