Arguably you want the least amount of data per byte, because then when Corruption shows up, it has less of a chance of removing information :V
This is why plain text is actually the best *in general*. Although corruption is difficult to spot, it's blisteringly easy to repair by hand because our entire mind is built to recover context from partial-context situations
And because each individual byte only contains a blisteringly small part of the overall structure, it's resilient to certain types of corruption (There is actually no 'one size fit all' corruption mechanisms)
LDPC is actually perfect because it gets as close to the Shannon limit as is possible (there's still room for improvement)
What I haven't seen thusfar is an assessment of the types of bit rot on storage mediums. What form does bitrot take? Literally nobody knows
To expand on that last point: The pinnacle of anti-corruption software consists of both turbo encoding and LDPC encodings. But the problem with that is if you have a huge single chunk of the data wiped out, those codes fail. They're designed for variable errors in a bitstream, if someone runs a pin down a strip of the magnetic tape or whatever you're using, you're stuffed