I am thinking of my trace prints and they are usually text which actually does not compress too well with RLE. English doesn’t have too many repeated letters. LZ (even with a tiny window) would work better.

That is to say: I am doubting the 33-70% compression

The 33-60 % figure is for repetitive firmware logs (periodic status lines, hex dumps, etc.) where the XOR-delta between successive bytes is zero most of the time.

The headline should read “compresses repetitive logs”; sorry for the noise.