Turns out to be a math challenge, rather than programming. Very challenging test suite.

I agree.

On one hand it's the temptation and ease of regex, on the other is the temptation of writing a beautifully structured interpreter toolchain

Great kata! Touches a couple of side problems. I learned something.

Implemented a deterministic solution with complexity O(n^3) in Python. Solves 70-by-70 puzzles in <10s on my laptop.

Nice game. This is 5% coding, 95% thinking.

Great kata. Sounds simple, turns out to be very challenging.

Couldn't you copy and paste it here? Having to go to a different kata to see information you need in this one shouldn't be needed. Even more when this is the easier one, and the normal order for doing this should be easier first, harder later, not the other way around.

Thanks.
CONTIGUOUS. There is only one 2 next to 3.

Sorry, I don't understand the example {2,2,2,3}.
If subarray {2} is counted three times, why is {2,3} only counted once?

(unfortunately, the answer to nekoman's question is hidden)

Well done!
I'm not a big fan of regex, but this demonstrates it can help to concentrate on the essential stuff, leaving all the tokenization on a single line.

Perfect iterators!
I think four conditions should cover everything.
In order not to miss anything, shouldn't the exceptions be caught around each condition separately?

https://oeis.org/A002326

I tried with strings of length 0-100. It seems to be always cyclic, but the number of repetitions is not a trivial sequence.
Results for length 0,2,4,...,100 (length+1 is the same cycle size):
1,2,4,3,6,10,12,4,8,18,6,11,20,18,28,5,10,12,36,12,20,14,12,23,21,8,52,20,18,58,60,6,12,66,22,35,9,20,30,39,54,82,8,28,11,12,10,36,48,30,100
How to calculate that for any length?