Reading results
A run opens on a live summary — pass, fail and error counts, elapsed time and progress. Click any test case for its detail. For a printable record of the whole run, click Evidence report — see Reports.
The step strip
Each step appears as a chip with its number, text, duration and a status dot. Click one to see the screenshot taken at that moment and the AI's reasoning. Badges flag anything notable — 🩹 self-healed, 🧠 resolved from memory, ⚠ ambiguity warning.
Screenshots
Every step is captured. Visual Diff and Side-by-Side compare consecutive steps, which is the fastest way to see what a click actually did.
The AI Decision Inspector
For each step this shows what the engine did and why: the action, the selector or coordinates it used, its confidence, its reasoning, and the alternatives it considered and rejected. When a step goes wrong, this is where you find out whether the engine misunderstood the step or the application misbehaved.
Failure triage
Every failure is classified automatically, with evidence:
| Verdict | Meaning | What to do |
|---|---|---|
| APP_DEFECT | The application misbehaved — JavaScript exceptions, server errors, wrong screen | Raise a defect. A ready-to-paste report is generated. |
| TEST_DEFECT | The test no longer matches a working application | Apply the suggested rewrite. |
| ENVIRONMENT | Network, DNS or availability trouble | Check the environment and re-run. |
| DATA | The test data was rejected — validation, credentials, duplicates | Fix the data. |
Self-healing
When a test case fails, the engine retries it once with adaptations — rewording the failed step from its own diagnosis, bypassing cached selectors, waiting longer. A healed pass is labeled "PASS (self-healed after 1 adaptive retry)" and carries the suggested permanent fix.
Script-defect advisories
Sometimes a step passes for the wrong reason. If your step quotes 'Complete Purchase' but the button actually reads 'Proceed to checkout', the engine finds it anyway — and flags it. The step passes, and you get a warning with a suggested rewrite. Worth fixing: right now that step only works because the AI rescued it. These advisories are collected for review in Autonomous Tuning, with the evidence and a suggested correction.