Approved!

Btw, I like the "manual" svg graphics in the description

yeah it if boils down to micro optimisation, better leave it as is

Hm.... How so, because $2 \cdot 10^5$ for 15 tests requires performance right. My solution has 26 NlogN operations. When I made it 15 tests of length 1 million, mine took ~7 seconds and yours took 12 probably because of recursion, but the time complexity is pretty similar.

Thanks, I'll combine both other ideas, without extreme performance requirements.

I totally understand your frustration! I've spent days trying to fix bugs myself. It's tough when a formula just won't behave. Turns out I was the one with the bug—guess I need to read the instructions too! I'm glad you figured it out in the end. Great job!

I tried to reduce my code for days, not understanding why my formula was wrong!

Great concept and very well-presented. I think parts II & III could be combined into a single kata (one that I look forward to trying), although the more performance-hungry among us might demand a separate kata devoted only to that.

Yes, apologies.

should be 533

35 and 53 are not totally good (having 3 and 5 from 355)

No reason to use spoilers here.

• hmm... would error tolerance of ±(1 kg/m^3) be wide enough, though...
• or maybe in the [floor(ρ),ceil(ρ)] range?