I found a bug. I wrote bad number to #CHCK instead of #KEY1, but my solution was accepted.

I feel like if you were allowed to stay in the KEYPAIR hosts for just 1 more cycle, this puzzle would be a lot less frustrating. 1 more cycle isn't enough time to make a file and save the #inpt to it, but it is enough time to use a JUMP command. But because there's not enough time to use a JUMP command, I ended up having to waste many lines reusing code over and over. Which made my solution very large.

I was able to shave off the 6 lines I needed to qualify for the histogram leaderboards in the end, but I'm almost certain I can't get it any smaller now, and it's exactly 100 lines.
Figuring out how to get the #inpt to the end was a fun and interesting challenge. Spending a week rearranging my code just to make it fit in 100 lines wasn't.

And I still don't understand at all how other people are beating it with so few cycles and small solutions sizes at all.

Well I managed to make a solution that is exactly 100 lines long, but it fails on 1 out of 100 tests (test #93), because my method of storing the #INPT makes it max out at 9999 on that puzzle. If I add code to test for 9999 it goes over the solution size limit. I'm so frustrated and disappointed right now. I don't understand how people are beating it with such small solutions at all.

EXTREME difficulty is right. I'm cramming data in placesI didn't know my EXAs had! Although it seems impossible, I can confirm that it is solveable, although I had to add special checks for certain test cases.

How are people doing this in <30k cycles?