I found the problem in Java version in both random tests generator and reference solution, and hopefully fixed it. Let me know if there's still some issue.

I assumed Python because they completed the kata in Python, but it's certainly possible they encountered the bug in another translation. This is all under the assumption that the bug being reported is actually legitimate.

MAybe it could be one of my translations (Java or C#)? Other than Ruby, I can review all translations and check random tests generators, but if you know exactly what language version causes touble, let me know.

Impossible: the bug was found and fixed a year ago. The only explanation I have is that @makerGeek attempted the kata before that but has not reset the trainer since then. If the issue is not with the Python version, raise the issue with more info than "something is wrong".

When raising an issue, you should specify the language (as this kata has several different translators). Looks like you're referring to the Python translation, so @FArekkusu would be the guy for that.

@docgunthrop I found a wrong random test: '-2o' should equal '-2o+0'

Comparing to other competitive programming platforms, I think codewars'katas lack complexity constraints (which are important in real life and in programming interviews).
Or is there problems with time and memory constraints but I didn't see any of them?

@Blind4Basics: any explanation why checking only one subsquare is enough ?

[JAVA]

It will be great if java.util.* becomes imported by default!

thanks a lot, makes sense now, and yeah quite clever :)
I also found this post that explains how to do the summation useful :
http://www.trans4mind.com/personal_development/mathematics/series/sumNaturalSquares.htm