I think there's probably a much more efficient way to pack the correction data than JSON. For example, if you wanted to embed a 10MB video file in there, the correction data would be huge.

In the project there, correction data is used to recover bytes that have been changed into LF when they are actually CR or CRLF.

One idea is to store the correction data as binary, then read two bits every time you see a LF byte. It's either an actual LF, a CR, or a CRLF. The downside is that binary data itself could need correction as well, and encoding nearly 1-bit data in 2 bits is still wasteful (but simple). Packing five 3-state values into a byte is less wasteful and would eliminate forbidden symbols, but is still not optimal.