This can be tricky, and, like you, I cannot point to a rule that specifically addresses blanks in a Nemeth-within-UEB transcription.
I *can,* however, affirm that it is inappropriate for the transcriber to give more information than the print does about what an answer should/will be like.
What I’m trying to say is, transcribe a low-line/omission dash/underscore according to where it appears in print and NOT what will/should replace it.
So, in your (very illustrative!) examples:
The low line that is part of a math expression (involving the “is congruent to” symbol) should be in Nemeth Code.
The low lines following the words “corresponds to” should be in UEB.
Really, I would put the low line in the sentence “So <fraction> is ____.” in UEB ... following the reasoning that the braille reader is getting the same information as the print reader from the instructions “The answer will be in decimal form.” and to transcribe the low line in Nemeth Code would be overkill.
Of course, this doesn’t solve all of our dilemmas, but I hope it shrinks some of them.
... Please let me add that I would transcribe consistently in Nemeth Code a <b>shape</b> that stands for something missing within a math expression and is then repeated and asked about outside of a math expression.