Do you mean that I should put the "doseq" in the "deftest"?

There are randomized tests that are pregenerated, as described in the comment:
; these are were generated offline as random string creation is very time consuming in clojure
; they are based on C++'s algorithm as follows:
; first and last cells are always clear, maze size between 4 and 11 (not including line breaks), 1 out of 3 chance for a wall

I've added the code I'm using for maze creation to the tests also.
(defn random-maze-builder
[size]
(let [maze-size (+ size (* size size))
break-idx (+ size 1)]
(str
(reduce
(fn [maze idx]
(let [next (cond
(= idx 1) "."
(= (mod idx break-idx) 0) "\n"
(= idx (- maze-size 1)) "."
:else (if (= 0 (rand-int 3))
"W"
"."))]
(.append maze next)))
(StringBuilder. maze-size)
(range 1 maze-size)))))

Wow I wasn't aware it would be twice as fast with mapv!
Thanks, this is a tip to remember.

Fixed the comparator, unless I'm missing something additional (in case I'd appriciate you'd point out)

Could you please elaborate? When I tested it, it worked fine

Having the frequencies call within the every makes it an O(n^2), so it's probably better to calculate them ahead of time

Saying what's the issue with this solution would be much more helpful than just saying "is terrible"

Original signature is "std::string DNAStrand(const std::string& dna)" in which you can't change dna because of the "const"

You should probably "dedupe" only after the sort, as there is no promise array1 is sorted

Using "replace-first" misleads the code reader about you're trying to do, and adds un-needed steps, you could use "includes?" instead

In "substring-of-any?" you could use "when" instead of "if"