Skip to main content

​
Two Ways UI Tests Fail on Rerun

When you rerun a UI test that previously passed, one of two things has changed: A plain rerun can’t tell these apart. Auto-Heal can: the test recovers from the first kind, and the second kind still fails, as it should.
In the web portal, Auto-Heal is built into UI reruns; there’s no toggle to switch on. Not every plan includes Auto-Heal. Compare plans on the pricing page.

​
What Auto-Heal Does

TestSprite runs your saved test. If the page changed but the flow still works, the test recovers and is marked Passed; if the flow is broken, it stays Failed. Tests stay green through refactors, design system swaps, and copy edits without anyone editing them by hand, while real bugs still surface.

​
When Auto-Heal Helps and When It Doesn’t

It helps with UI drift:
  • Minor refactors: renamed buttons (Submit → Save), navigation moved into a hamburger menu, a wrapper element added around a form field.
  • Design system swaps: a custom dropdown replaced with a shadcn Select, with a new DOM and accessibility roles. Auto-Heal re-binds the test to the new component.
  • Copy edits: “Continue” became “Next”. Anything matching by text breaks on plain replay; Auto-Heal sees the rename in context.
It doesn’t help with:
  • Real product bugs: a required field rejects valid input, an API returns 500, a redirect goes to the wrong page. The test stays Failed, which is the correct outcome. Auto-Heal absorbs UI drift; it doesn’t make tests pass.
  • Removed features: if the flow was deleted on purpose, remove the test from your suite. Auto-Heal can’t tell intentional removal from a regression.
  • A broken login or onboarding gate: if the sequence before the flow is what broke, recovery can’t get past it. Keep your test account in a known state.

​
Why Frontend-Only

Auto-Heal applies only to UI tests. API tests issue HTTP requests, and routes don’t drift the way a UI does without an explicit version bump. When an API test fails on rerun, the cause is usually one of these:

Endpoint behaves differently

A real bug. The failing assertion and the recorded request/response usually show what changed.

Auth token expired

Configure Auto-Auth so a fresh token is fetched before every run.

Required fixture was deleted

Check Auto Cleanup and your dependency chains: a producer’s record may have been removed before its consumer ran.

​
Edge Cases & Troubleshooting

Open the test detail page and look at the Steps list in its left column. Your saved test stays the source of truth; Auto-Heal doesn’t overwrite it unless you explicitly save updates.
Yes. Recovery covers the whole step sequence: steps that still work are reused, and broken ones are re-decided in context.
The test is marked Failed with the last-known error in the Error / Cause / Fix tabs of the test detail page. Click Run test to try again. If recovery keeps erroring on a test, contact support with the test ID.

​
Where to Go Next

Edit & Rerun

The rerun surfaces: single, batch, scheduled

Test Schedules

Set up nightly schedules

Insights

Regressions and quality trends across runs