BalaGAN | art image-to-image translation methods | Machine Learning library

I have been struggling to do this (every DAY!) for at least a month. I have searched stackoverflow, I have read MDN array, string, regex, etc., references over and over and over again, and nothing has helped. I am somewhat familiar with regex, but this is over my head. I trust that somebody here will solve this with one line of code, which is why I waited until I'm about to throw my computer out the window before asking for help. I really wanted to find the solution for myself, but I simply cannot do it.

I was enjoying a game of cryptograms, where random letters are used to sort of 'encode' a poem or story, I probably don't need to describe it here, but here's a picture just in case.

So I thought it would be a good exercise to create a form where you can enter a pattern made up of a combination of letters, numbers, and "?" for unknown. In the image, you see the word represented with "YACAZ", there are two A's in that word, so you know those two letters are the same. So in my function, you would use any number 0 - 9 as placeholders, so using the same example, you would enter "?1a1?".

Here's what I have at the moment. Every time I try to iterate through the arrays that regex gives me, I end up at the same place, trying - and failing - to compare two sets of nested arrays with each other. No matter how I try to break them down and compare them, it ends up becoming a huge non-functioning mess. I can get the placeholder indexes, but then what?

I have nothing against lodash, but I have very little experience with it, so maybe it could help with this? It doesn't do anything that cannot be done with plain vanilla javascript, does it?