You have a good point that, if in my case it happens only to ungrammatical ones, then it must be the star. In which case it is failing for the wrong reason and is not checking the grammar. Shouldn’t it be aware about the star though?? So odd. I will try to look into it more.
One thing I did recently which is actually related to the asterisk is I added a couple of lines to customize.py’s customize_test_sentences() to ensure that the star is printed to the test_sentence file. I made no modifications to the code related to writing things to lkb/script.
But there is a difference: before I had to manually add the asterisks to the ungrammatical sentences that I entered on the website. Now I just grab the file automatically written by the customization system. I will try to create another testsuite by hand and will see if the issue persists.
P.S.: Here’s a live example of fragmenting support :). This question is being discussed simultaneously on the developers list, since Woodley and others who might help with this question aren’t here. Not ideal ideed… But my hope is that if we resolve the question and summarize the solution here, it will then be more discoverable for the newcomers.