I thought your code is similar to mine.

Fixed by avermakov

Ruby now has sample tests (added some time ago)

I totally agree with OP. The description sucks. Guessing your way out of a bad description does not count as a special required skill on the part of the solver, but of an obnoxious attitude of the author.

Can you specify which language this occurs in please?

Fixed.

Added a fork but wait for approval, there is something I don't get, it shows only one test in the log, when there are a lot more.

The C# version has the same problem, the precision of double starts to break down after 15 or 16 digits and the tests sometimes pass and sometimes fail depending on how high the random length value happened to go.

Should be fixed now.

It's easy to fix this warning: just replace np.random.random_integers by np.random.randint. Shouldn't affect performance.

Unfortunately, only users with more than 10k honor can edit the tests.

absolutely this.
in fact, generating the random strings dominates execution time. I tried a fork where the random string is generated once and provides substring slices for each test but even that times out when generating a single 100,000,000 term string.

Edit: i've created a fork which can generate random testcases fast enough, but guess what? even more numpy

numpy random utilities are used most probably because Python's random module is (was?) affected by a performance issue: https://github.com/codewars/runner/issues/58

Please use spoiler flag next time.