Your task is pretty easy: to write a regular expression that matches [snip] but doesn't match [snip]

Ehm, no. Apparently, we are supposed to read author's mind and come up with some incompletely specified regex.

That's not easy, and it's not good kata design.

Of course, you don't want to hard-code the given examples, because such a solution would not pass the final tests.

So that means there are more words tested than user can know about.

So that means the rules that apply to the known words, may be different to the rules that work also for the other unknown words.

So that means user can only guess and hope...

??