1. I experienced the same issue of everyone

2. After working on dtype of index and things like this, I checked if a hunch was correct,because in the random tests, one test was right and the following wrong and viceversa

3. I created a global variable and each time my function was called I increased the variable and I find out that my function was called only when the test was wrong and it wasn't called when the test was right, in the end i tried to modify my solution to get wrong results on purpose and I still got 50% of successful test.

So i think the problem is how they check our solutions!!!

for me it's been usefull to spend little time writing several lines of code to solve a kata and find out a buit-in function or a faster way to do the same thing, this way i will hardly forget it

I wrote the same solution , but instead of 35 25 15 I used 200 200 100 because it works even with bigger nunbers, so you don't need to be accurate