There were too many different strategies for this puzzle, to keep myself sane I used a simple heuristic of how many total checks on top of a unification glyph was needed. The following was the results I calculated for all strategies I could think of from least to most complex:

1: Sort a match pattern and substring for each substring. (576 checks)

2: Test for inequality between the match pattern and each substring by loading both at the same time. (432 checks)

3: Sort the match pattern once then,
  3a: Sort each substring, testing if the 3 atom pairs match. (300 checks)
  3b: Sort the sample once, using logic to reconstruct each substring, then do 3a. (156 checks)
  3c: Use the match pattern to selectivly take out an atom from a complete set of 4, then test each atom in the substring if it is, or isn't said atom. (84 checks)


After much testing I landed on a choice between 3b and 3c. 3c had half as many checks needed compared to 3b, but something to note is that a 3b sorter is much easier to build then a fast 3c. Based on my own testing, a 3b core could do a 17 cycle check with 400g and 40 area, while an 8 cycle 3c core was 1600g and 170 area. Knowing the earlier heuristic I was able to tell that 3-4 3b sorters could match the metric wsum of my 3c 8 cycle core. So what to do?

What drove me to 3c was simply time, I had already built a 3c core, and I was gonna use it. 3b's approach of 3 parallel sorters while using logic to reconstruct each substring also sounded like logic hell. Ah well, who knows maybe the winner of the puzzle will be exactly that kind of approach using brains to beat my solves brawn. This note is already too long and rambly, so I'll leave it here. 

- Starficz