Really nice kata. I was going to suggest a potential sequel for writing a proper transpiler, but seems like many people already did that for this kata!

I would say easier 4 or harder 5, yeah.

kata is approvable, 4 kyu seems right?

Let's stick with printable characters. I updated the description.

Currently, random tests use char code 32 - 126. So, I don't think this needs to be handled.

Leaving author to resolve this, since author might have a change of mind later.

Does print_string need to be able to handle \x00?

Fixed

Its not, thanks

Right, I did not have performance in mind. Made the tests smaller.

Is mem[12] correct in the description of Addition?

0:  -2,  9,  3,   ; mem[9] -= mem[-2], go to 3  (now mem[9] = -A)
3:   9, -3,  6,   ; mem[-3] -= mem[9], go to 6  (now mem[12] = -(A+B))

I agree, multiplication should also be possible well below 1 second.

There are fixed tests with large numbers (up to ~10^6). IMHO 10^4 is sufficient.

The other tasks are faster (my times: multiplication: 2740ms, division: 150ms, print: 10ms, copy: 670ms)

Error pane crashes with out of memory on error. You are printing these larger arrays, there is no need to.

Are there performance constraints? I just implemented multiplication, and all tests (fixed/random) together, already take me about 5 seconds.

Fixed, thanks! :)