If you split for example "a__b" (double space instead of underscores) into a list, it will result in ["a", "", "b"].
When you join it later, there will be one space between "a" and the empty string "" - and another space between the empty string and "b". Therefore you end up keeping the double space like in the original string.

Your solution is not far from the above.

I'm sorry for my bad English, but in order to solve this kata, I needed an additional condition, because the tests did not pass, and everything went without it...
Strange tests, the second time passed without a condition :)

https://docs.python.org/3/library/stdtypes.html#str.split

If sep is given, consecutive delimiters are not grouped together and are deemed to delimit empty strings (for example, '1,,2'.split(',') returns ['1', '', '2'])
...
If sep is not specified or is None, a different splitting algorithm is applied: runs of consecutive whitespace are regarded as a single separator

How does this account for string casing, i.e uppercase, lowercase?

OP's solution is better imo. More readable, and doesn't add clutter (Not that yours does)

Haha I like that

As someone whose last math class was before 9/11, I sometimes feel this way, but not this time!

This was clearly explained, with great examples.

Plus, it was an actual coding challenge, not some Math-trick riddle that's useless in real life.

behind the scenes, the JS interpreter will have to:

• instantiate a new RegExp object
• invoke the RegExp engine
• compile the RegExp
• parse the input string, which can get very slow with some backtracking patterns

Why does using regex make it slow?

It uses regex, so answer is "no".

But it's slower

How about mine? What's the time and space complexity

Well, one call to match is performed in O(N) and you cannot solve the problem faster.
But it also allocates extra O(N) memory only to take the length of the array[-ish structure] later.

TIME = O(N), optimal

MEM = O(N), non-optimal, O(1) is possible.