I also stored waaaaay too much data.. I spent way too much time on the math for this one too.

The order of results requirese on xi

It is bad practice to create a list in the sum function. A generator will not allocate memory. Also, I find x and y to be better names for points in this, i.e. math, domain.

Someone makes a Kumite and has a requirement eg. "Turn a string into hexadecimal" and players can submit code in response to other code to improve it by either making it more concise and/or faster. May the best code win

My solution has a problem about round() when the input number is very large.

This is how I would do it, as it is more declarative vs. iterating over the list and building a string.

View these discussions, I have better understanding what the author wants.
The regular expression. I will have a try.

Using package not in the standard python library like numpy, to test others is not very good.
I think we come here to train language , not the special library.

How to play Kumite, I don't understand.

If adding the gap between the nearnest primes, I will have better understanding.
When I view the solution, I find that I miss the author says. And find two primes between
m and n with gap g, regadless whether they are near to each other.

Just wonderful, I store too much data. But what this keta need is just two numbers!

The same, maybe the author leave the chance to write tests to us

The same problem, 450010 should equal -1