I think the not to check if the *.out files are expected should come before this step. If you generate the outputs and immediately run the test, it will always pass.

It should be clear also that if generating change other tests that you didn't add, you should revert it back and not commit.

Not sure if I understand correctly. Is it possible to run tests for new scripts that don't have *.out files generated for them yet? Not sure what would be the point of that, since there will be nothing to compare the output against.

I agree that it can be clarified that those generated out files which fail should not be committed, but a regression should reported (yet this likely won't be an issue, since regressions are going to be caught early at CI checks stages). This is also not a big deal for someone who participates in implementing GDScript language itself.

But current version does tell:

Make sure the output does have the expected values before submitting a pull request.

What I mean is that, the process describes how to write new tests, not updating existing ones. If you believe it's beneficial to describe those processes separately, then it may be worth it. But I'm afraid this could potentially make the documentation more confusing.

Calinou went through those steps successfully while making godotengine/godot#48029, for instance. If someone is going to have a problem, the documentation could be further clarified I guess.

That said, not sure what kind of changes to make without breaking the currently described workflow.

Added an explicit warning: