Thanks, I really like the pattern matching in python :)

At least it allows to benchmark a user solution without the slowness of the previous referee ;-)

Would be interesting to update the referee solution as the current one is very slow

The author is inactive.
Do you want to propose a fork to fix this issue?

Added in this fork

Added for Ruby, re-raised for other languages

Approved

25 is feasible, maybe optimum

Maybe just try to dump some full game notations from reputable sites like lichess.com and add it to tests, a hand-picked (or big enough) sample will be helpful and might just be better than the random test generator.

Try changing your approach. This greedy algorithm takes ~2^65 steps for "worst-case" 128-bit numbers.

I passed 26k with hidden communication.

Probably not - and there probabaly is a smarted way to prevent simple cheating.