I also find myself having issues with this. I’m especially having issues on when it does or does not want an honorific prefix on a verb. Sometimes it absolutely requires the honorific お and sometimes it demands that it not be present, even sometimes differing on the same word across different situations. I can’t seem to identify a pattern. One recent example was:
買っていただけませんか being mandated, and お買っていただけませんか being incorrect. I could really use explanation on why in this case, or in other cases, as to why this is the case on each sentence.
Another issue I have is with requests, where some requests will allow くれる but not くれないか and won’t allow くれるか. It feels like almost a complete guess which variation the sentence demands, and it feels like several simply reject many variations arbitrarily with no indication why they are rejected or why they may be invalid.
If we have an entire arsenal of ways to make requests of slightly varying politeness levels, but which still mean effectively the same thing, what are these questions trying to reinforce by rejecting half of them with no explanation?
e: For some of these problems, would multiple choice questions simply be easier to curate? A pool of correct choices with more possibly unlocked by future grammar points, and a pool of incorrect choices. Display, for example, 3 of the incorrect pool and 1 of the correct pool each time the question comes up. Then just prompt to select the most correct / a correct choice. Provide explanations on why each incorrect answer is wrong, and the specific nuance of the correct answer. Would this provide better learning in situations where there are potentially MANY correct choices?