- Docs
- Build tests
- Maintenance
- Troubleshooting
Troubleshooting
Find the message or symptom you see. Each entry says why it happens, what to check and how to fix it.
In short
On this page you can
- Fix load and save errors on the Auto-heal, AI Steps and Network Calls tabs.
- Find out why a failed step was not repaired, from its run-log reason.
- Handle a test that fails again after you rejected a repair.
- Solve conversion errors from Save as Action.
- Act on every warning in the Network → API steps dialog.
How to use it
- Copy the exact message from the screen or the run log.
- Find it below; the section follows the tab or the run log.
- Read Why, do the Check, then apply the Fix.
- Still stuck: write to info@smartestqa.ai with the project name and the run ID.
Auto-heal tab#
“Auto-heal records could not be loaded”Auto-heal
- Why
- The list request failed: the connection dropped, your session expired, or the project is not in your company. The tab shows the error instead of “Nothing waits for review”, so you never mistake an error for an empty queue.
- Check
- Other pages load; the project selector shows the project you expect.
- Fix
- Click Retry. If it fails again, reload the page and log in again.
“The decision could not be saved”Auto-heal
- Why
- The text after the colon says why. “Auto-heal record #n is already APPROVED” (or REJECTED, PROMOTED, SUPERSEDED): someone else, an AI assistant or the runner decided it first. Only repairs that are Waiting for review can be decided. “Access denied” or “not found”: the repair belongs to another project or company.
- Fix
- Change the filter to All to see the current status and who decided. If the repair was replaced, decide the newer card instead.
A step failed, but “Nothing waits for review”Auto-heal
- Why
- The runner refused to repair, or the repair was not saved. A card is created only when a repair passes every stage of How auto-heal decides.
- Check
-
- The project type: MOBILE projects are not healed; DESKTOP is healed only with Auto-Heal on (project types).
- The failed step’s run log: find “auto-heal did not repair the step — …” and look up the reason in Run-log reasons.
- The step type: only steps that find an element are healed. API, database, mail and AI steps are not.
- An assertion that failed on a real difference is not a repair case.
- Fix
- Act on the logged reason. To repair by hand, add a working locator in Discover › Action List.
The test fails again after I rejected a repairAuto-heal
- Why
- That is expected. Reject puts the action’s old, broken locators back: “The next run will fail on this step until the test is fixed.” If the runner can find the element again, it may instead make the same repair again.
- Fix
- Open the action in Discover › Action List and add a locator that finds the right control, or change the step in Automate. See What Reject changes.
I rejected a repair, and the next run made the same repair againAuto-heal
- Why
- The runner does not remember rejections. Reject restored the broken locator, so the next run found it broken, found the same element again and saved the same repair.
- Fix
- Rejecting is not enough on its own. Fix the action’s locators in Discover › Action List, or fix the app if the repair pointed at the wrong control.
The same assertion creates a new card on every runAuto-heal
- Why
- An assertion repair changes nothing in the test. On each run the same two texts pass again on format alone, and each pass is recorded. Rejecting a card does not stop it either.
- Fix
- If the new format is correct, change the expected value in Automate to the format the app shows now. If it is wrong, report it as a defect. See Assertions.
The healed tag or the InSight line does not go awayAuto-heal
- Why
- The tag stays while the test case has at least one repair waiting; the InSight line stays while any repair in the project waits. One test case can have several waiting repairs, one per action plus assertion repairs.
- Fix
- Filter Waiting for review and decide every card of that test case. See The healed tag and the count.
An older repair is waiting for review againAuto-heal
- Why
- You rejected a repair that had replaced an earlier one. Reject restored the locators as they were, and those include the earlier repair, so it is in use again and waits for a decision.
- Fix
- Review it like any other card. Reject it too if that locator is also wrong.
A card shows “—” instead of a test case nameAuto-heal
- Why
- The repair is not linked to a test case. Desktop repairs are recorded this way; their card shows the method Previous auto-heal engine.
- Fix
- Use the action and page names on the card’s second line, and the run ID in Details, to find where it ran.
“Only an absolute XPath identifies this element. Adding a data-testid to it would make the test stable.”Auto-heal
- Why
- The element has no stable attribute; only its position in the page identifies it. Such a locator breaks on the next layout change, so the runner uses it for the current run only and does not save it.
- Fix
- Ask the developers for a
data-testid,idoraria-labelon the element, then add it as a locator in Action Detail.
A wrong repair was confirmed by runsAuto-heal
- Why
- Confirmation counts runs in which the new locator found an element. It cannot tell that the test checks the wrong thing. A confirmed repair can no longer be rejected.
- Fix
- Correct the action’s locators in Discover › Action List. Review waiting cards before three runs pass.
Run-log reasons a step was not healed#
When the runner does not repair a step, the step fails with “Element not found: … — auto-heal: reason” and the log line “auto-heal did not repair the step — reason”. The stages are explained in How auto-heal decides.
“not healed — the page itself returned HTTP …” / “the application's API returned a server error” / “the page shows an error message” / “the session ended”Run log
- Why
- The page check found that the app itself is broken or logged out. Repairing would hide a defect, so the step fails on purpose.
- Fix
- Treat it as a defect in the app or the environment. For “the session ended”, check the test user and the session timeout, and log in again earlier in the test.
“no element with the same identity, and AI auto-heal is turned off” / “… and AI service is not available”Run log
- Why
- No exact match was found, and the AI stage could not run: Enable AI or Auto-Heal is off for this run, or the AI provider could not be reached.
- Fix
- Turn both on in Project Panel › AI Options, or in the pipeline’s settings. If they are on, check the AI account in the same place.
“AI is not sure enough (best … < 85)” / “two candidates are too close (best …, second …; needs a lead of 15)”Run log
- Why
- The AI did not find one clear winner. See the acceptance rule. Common causes: the control was removed, renamed beyond recognition, or several similar controls sit side by side.
- Fix
- Check the app. If the control is still there, add a locator by hand; if several look alike, use one that includes distinguishing text or a
data-testid.
“no 'type' element is visible on the page” (or 'click', 'select'…)Run log
- Why
- No visible element on the page suits the step type, for example no input field for a Type step. The page may not have loaded, or a dialog covers it.
- Fix
- Look at the step’s screenshot in the report. Add a wait or the missing navigation step before it.
“the AI gave no usable answer” / “the AI scored only n of m candidates”Run log
- Why
- The AI reply was empty, unreadable, or incomplete. A missing score would make the 15-point lead automatic, so an incomplete reply is refused.
- Fix
- Run again. If it repeats, check the AI model and token limit in Project PanelAI Options.
“could not build a locator that matches only the chosen element” / “the chosen element could not be used”Run log
- Why
- The winner was found, but every locator the runner could build also matched other elements, or the element could not be used.
- Fix
- Add a unique attribute to the element, or a locator by hand that includes its text or its container.
“the repaired element did not behave like the original (…); nothing was saved”Run log
- Why
- The effect check failed: for example “last time this step opened another page; this time the page did not change”. The element was probably the wrong one.
- Fix
- Nothing was changed. Check the app and fix the locator by hand.
“step repaired for this run, but the locator was not saved”Run log
- Why
- The step passed with the repaired element, but the locator was not written to the action. The detail line says why: “no stable locator for this element (only a position-based XPath)…”, or a server error while saving.
- Fix
- For a position-based XPath, see the data-testid entry. For a save error, the next run tries again.
AI Steps#
“The list could not be loaded.”AI Steps · Network Calls
- Why
- The request for the AI Steps or Network Calls list failed.
- Fix
- Click Retry. If it fails again, reload the page and log in again.
My AI step is not listed (“No AI step is waiting to be converted.”)AI Steps
- Check
-
- It is a WEB project and the step is an AI Do step (it performs an action). Ask, Check and Chat steps have no element to save.
- It passed in a pipeline or scheduled run. Dry Run results are not saved.
- It is still an AI step in its test case.
- Look under Can't convert and Converted.
- Fix
- Run the test case in a pipeline, then reopen the tab. See What the tab lists.
“This step has already been converted to an action.”AI Steps
- Why
- Someone converted it, for example in another tab. The list refreshes and the row moves to Converted.
- Fix
- Nothing to do. Under Converted the row shows the new action as “Action #id”; find it in Discover › Action List.
“This AI step cannot be saved as an action: its action type is no longer supported.” / “"…" has no action type; it stays an AI step.”AI Steps
- Why
- The AI used a verb with no action type, such as
press_enterorscroll_to. The Save as Action button is disabled. See the list. - Fix
- Keep the AI step, or replace it in Automate with a regular step or a Python or JS step.
“Only a successful AI (Do) step can be converted to an action.”AI Steps
- Why
- No saved result shows the step passing with an action and a locator.
- Fix
- Run it in a pipeline until it passes, then convert.
“Action saved, but the original AI step no longer existed” / “Original AI step not found, but action was created”AI Steps
- Why
- The action and its new step were created, but the old AI step could not be found or removed, for example because the test case was edited meanwhile.
- Fix
- Open the test case in Automate. If both the AI step and the new action step are there, delete the AI step.
“You do not have permission to delete this step”AI Steps
- Why
- The server refused the change (HTTP 403), usually because the project is not in your company or your session changed.
- Fix
- Reload and log in again. If it repeats, ask an admin to check your project access.
“Required fields are missing, please fill in all fields” and other conversion errorsAI Steps
- Why
- “Required fields are missing…”: the saved result has no locator or action. “Project ID is missing or invalid”: select the project again. “Could not save the action”, “Could not retrieve Action ID, please try again”, “Could not create the new action step”, “Operation could not be completed, a database error occurred”, “An error occurred while saving”, “Unknown AI action type, not supported”: the conversion was rolled back.
- Fix
- Reload the tab and try again. If the same step fails twice, write to info@smartestqa.ai with the project and test case name.
Network Calls#
My test case is not listed (“No Network capture is waiting to be converted.”)Network Calls
- Check
-
- The test case has a Network step (WEB projects only).
- It ran in a pipeline or scheduled run; Dry Run results are not saved.
- Look under Converted.
- Fix
- Run it in a pipeline, then reopen the tab.
“The capture could not be loaded.”Network Calls
- Why
- The captured calls of the test case could not be fetched, or the test case was deleted.
- Fix
- Reload the tab and click Make API test again.
“There are no requests to convert in this report. Run the Network step in "List" mode (whole session, API + pages + errors) and try again.”Network Calls
- Fix
- Set the Network step to List mode in Automate, run the test case in a pipeline, then convert. See Tips.
“No requests in this filter.”Network Calls
- Fix
- Choose another filter, or All. Your app’s calls may have been classed as Third party when they go to another domain.
“Requests use {{authToken}}, but no selected request produces a token.”Network Calls
- Fix
- Also select the request marked Login, or enter the token in the steps yourself. See What gets created.
“{{baseUrl}} must be defined at run time (e.g. as a suite parameter): …”Network Calls
- Why
- You chose to replace the host with
{{baseUrl}}. The shown address is the host that was replaced. - Fix
- Define
baseUrlwith that address, for example as a suite parameter in the pipeline.
“The list report has no request headers. If the app expects custom headers (tenant, language etc.), add them to the steps.”Network Calls
- Fix
- After saving, open the API Request steps and add the headers your API needs.
“You need to define values for these variables: …”Network Calls
- Why
- Masked secrets in headers, query or body became
{{variables}}. They are also listed in the new test case’s description. - Fix
- Give each a value before the first run, for example as a suite parameter or global parameter. Never paste production passwords.
A request is marked “truncated”Network Calls
- Fix
- “The body is truncated in the report. Complete it after saving.” Open the step and complete the body.
“Could not create the API steps”Network Calls
- Why
- Creating the test case or saving its steps failed, for example because your company’s licence limit for test cases is reached (an Admin sees the licence in Company Settings › Billing).
- Fix
- Look in the folder: a test case may have been created without steps; delete it. Then try again.
A converted test case is back in PendingNetwork Calls
- Why
- The API test case created from it was deleted.
- Fix
- Convert it again if you still need the API test. See After the conversion.
Still stuck? Write to info@smartestqa.ai with the project name, the test case and, for auto-heal, the run ID shown in the card’s Details.
Didn’t find your answer?Write to support with your project name and what you see on screen.
info@smartestqa.ai