I was pleased that ROSALIND and GANYMEDE were given as the example, because it made clear that it was permissible for two letters to encode as the same letter (in this case I -> E and D -> E). Had that example not been given, I would have assumed that it was like a normal substitution cipher where each letter encodes uniquely, which would probably have stopped me from getting the correct solution. I strongly suspect the example was chosen deliberately for that reason.
I agree that the instructions are ambiguous as to whether the same pair of names is used each time; "the clue answer is composed of letters in a name" should probably have said "a specified name" or something like that. I worked on the assumption that the puzzle would have been insoluble otherwise; if you could pick a different pair of names each time, then there would be multiple possible encodings for each word, and no way of knowing what to write in for the unchecked letters.
A model of clarity compared with this week's incomprehensible instructions though!