It’s possible to choose the wrong words for a given context, but since we only get one sentence on bunpro it’s difficult to infer the context without the hints. For example, the sentence in the original post has the English note “long form normally found in written or formal language”, which also points at なければならない, but if we only look at the Japanese sentence then the part we’re supposed to fill out is what’s the biggest clue the text is written or formal. In other words, we must use the hints.
I think it’d be difficult to find a context where only one form is acceptable. E.g. let’s say for なければならない bunpro could establish formal writing as a context, so when we see a few formal sentences around the one with the gap, we choose this form. The problem is formal Japanese that’s formal enough to be an obvious case for this form could be difficult to read for a learner who is in the middle of N4. We must avoid misunderstanding it as “all written Japanese” etc. as well.
Or let’s consider the even more stuffy せねばならぬ. It means the same as しなければならない and in an otherwise stuffy context, that would probably still be preferred, so we must make the context look old-fashioned and be trained to react to old-fashioned written Japanese with this form even though it’s not necessarily the best one. For example let’s say a sentence like …_______と考えてゐる could trigger せねばならぬ. The distinction is just as artificial as a hint that clarifies which form to use, just more annoying, and then some learners would not realize it’s artificial and just think this is the normal way to use the phrase.
Another point is that if possible, a single SRS question should only test one thing. SRS doesn’t work very well if the SRS can’t figure out what part the learner doesn’t understand. Randomly getting questions wrong because the context wasn’t understood correctly breaks the SRS to some degree because it can’t tell whether we got the context wrong, or understood the context but got the grammar wrong. Hints as part of the question avoid this. If it asks for “long-form + ば + なる” and we get it wrong, the SRS knows we didn’t remember how to use this specific form.
Another reason for the reliance on hints is also that if you add the なければならない point to your reviews, you expect to be tested on it, and therefore the site must ask for it specifically, and can’t just accept one out of 500 other ways to say it. Otherwise we effectively never get tested on a lot of the grammar we select for reviews.