does the task involve simplifying expressions before applying values? I'm trying to find a reason for overflow (I'm sure Wolfram pre-manipulates the expression)

approved

Confirming this is an issue.

Fork addressing this issue here:
https://www.codewars.com/kumite/667cdbbc086a7f00f6dbca18

JS only issue.
Randomize function from test cases is incorrect.
That lead to not unique letters in funnel.

function randomize(arr) {
  for (let i = 0; i < arr.length; i++)
    arr[Math.floor(Math.random() * arr.length)] = arr[i];
}

Must be something like:

function randomize(arr) {
  for (let i = 0; i < arr.length; i++) {
    const j = Math.floor(Math.random() * arr.length)
    const t = arr[i]
    arr[i] = arr[j]
    arr[j] = t
  }
}

.

Yeah, I don't know how finding a square root is a binary operation.

in random tests used same array for solution and 'funk'. im added wrapper that copy array of family members

What means sqrt ranges?

Sorry, but I think this Kata is too much strict about performance (C++), or maybe the test dataset is really huge. My code does zero extra operations, I use references when possible. Still it fails at big_ranges_test - sqrt_ranges. I don't think there is need to be so strict, when my code is producing exact results, and is pretty effective...

associative operation may not have an identity, think about (x,y) => x

Sure! Without this one, the code is more complicated than expected.

I guess it's your implementation's problem. I faced the same but changed something in data structure and now it works. I'm sure it's because function composition and matrix multiplication are non-commutative so the order of evalutation should be preserved.

I didn't have problems with min though.