I think the honor points counter is sometimes bugged (only clientside).
If you complete another kata (any level) , your counter will probably be updated with current kata points and your last kata points (kyu5 here)

For instance
// I got 600 points . I complete a kyu5 (8pts) . My counter goes to 602 ( :'( )
// I complete a kyu7 (2pts). My counter goes to 610 (602+8)

The major problem I see is that even the intended solution (talking about ruby here) is not really "smart": it's simple brute force (with caching), but for some strange reason the prime checking is slightly faster than the built-in one, and thus the normal efforts fail.

Hi again. 1. There is no recursive relation applied here.
2. Only a smart brute force algorithm will solve it, as I did it (under the 12.000 ms).
Advice: Try to use a fast primality test.
Cheers. Raúl

Hi Matei. First of all, thanks for your time solving the kata.
I'll give you my point of view, you may agree or not.
To solve a kata with a pre-computed list it's not fair but sometimes the users do it on purpose to show that the kata wasn't well prepared, specially if the precomputed list is available in the internet.
So, if the author prepared the kata , with precomputed lists as a solution, it's completely unfair.
But if the author's solution is visible in the solutions side, (once you solved the kata) and it's an acceptable solution, means that he/she's requiring something that was done by himself/herself first.
In katas that need memoisation, due to performance requirements, we may reduce our tests runtime. Instead of having two memoisations: the one of the user and the one of the tests suit, you'll have the user's one and your code will be checked in a better way (as we only have 12.000 ms to do it), with higher values. As I told you, only a POV. Regards, Raul.