Verdicts
A check ends Passed or Failed. What each one means, what the report holds, what it posts on GitHub, and what happens when Tada itself could not finish.
A check ends Passed or Failed. When Tada itself could not finish, the check has no verdict and says Didn’t finish. Tada keeps the three apart on purpose.
Overview
| What it means | Check on GitHub | In the report | |
|---|---|---|---|
| Passed | Every step happened, and every word a test expected was on screen. | Green, All 3 tests passed. | The demo, with a link you can share |
| Failed | Words a test expected were missing, or a step could not be done. | Red, Checkout stopped at step 3 of 5: click Pay. | What broke, on film, private to your workspace |
| Didn’t finish | Tada stopped before it could decide. | Neutral, Tada couldn’t finish this check. | The reason |
A feature takes its standing from its checks. It is Passing when its last check passed and Failing when its last check failed.
Passed
Every test found the words it expected on screen, and every step happened. A pass covers the tests the check walked, at that address, on that check. It says nothing about the rest of the app.
A test that expects no words has nothing that can go wrong. Its pass proves
only that its steps happened, which is why Tada does not save a test without
an expect:.
Failed
A test did not find the words it expected, or a step could not be done, such as a click on a button that is not there. That test stops at that step, and the tests after it in the same check may be Not reached.
The report names the step that stopped, the words the test expected and the words that were really on the screen. The cause is usually one of these:
- A real bug in your app.
- The test needs new words, because the app’s wording changed on purpose.
- Data the test needs is missing from that environment.
- Tada could not sign in, for example because a saved sign-in expired.
A failed check never gets a public link. Its film is visible only to your workspace.
Didn’t finish
Tada itself stopped, so nothing was proven about your app either way. For example, the address did not answer, or the check ran out of time. The check on GitHub is neutral rather than red, so a problem on Tada’s side never blocks your merge. A check that didn’t finish never counts toward your plan.
No AI decides a verdict on its own
A check follows the steps as written. A step that expects words passes when they are on the screen and fails when they are not.
The one exception is a step marked as a judgment, for what no rule can decide, such as the shape of a chart or what an image shows. A model answers it from a screenshot and the text on the screen. A no is asked again in a fresh call, and the step fails only when both answers are no; when the two answers differ the step passes and says so. The report names every step that was judged. A test can still be flaky when your app or the network behaves differently from one check to the next.