[exmadscientist]

It's both much less wasteful and much more useful to add an error correction (FEC) block instead. With random errors spread across two copies, how do you know which chunks of each are correct? Sure, you could use a Merkle tree or similar, but FEC is a better solution. For size penalties of order 10% in practical use, FEC blocks can detect and correct errors not just in the data but in the FEC blocks themselves too.

Their biggest downside is that their computational complexity can be high. This is only a problem during creation and decoding on error; if the data is checksummed, you only need to take the slow path and touch the FEC block if an error is encountered. This is a great tradeoff for archival uses, and good for many other uses.