This is a known issue currently. The existing spelling algorithm is quite sophisticated in terms of how it analyses pitch classes, morphs and chrome across a longer context (disclaimer: this is way outside my own music theory knowledge). I think it’s designed to work well where you have complex or ambiguous situations However, there are some cases like this where it doesn’t produce a useful result such as F rather than E# (or double sharps and flats).
We’re considering how to improve this in the future.