QR code error correction, explained simply

Error correction is expressed as a percentage of the code that can be lost and reconstructed. The default level is fine for clean digital use; printed codes often benefit from more.

The percentages describe how much of the pattern the scanner may fail to read — because it is covered, scratched, faded or badly lit — while still rebuilding the original data.

A QR code is not a picture of your URL. It is a grid of data words with error-correction words woven in, computed from the data with a checksum-style algorithm. When the scanner reads the grid it can spot which words came out corrupted and solve for the originals.

Crucially, the words are interleaved: consecutive modules come from different words. That is why damage gets spread out. A scratch through a fifth of the code damages a little of many words rather than destroying a few — and each of those words has its own correction allowance to spend.

More redundancy means more modules for the same content, so the code gets denser. That is why a heavily styled code needs to be printed larger: you are asking it to carry more backup data.

Density is the thing to watch at print size. The module count rises with the correction level, so a code at H has smaller individual modules than the same content at L printed at the same physical size. Scan distance depends on module size, not on the size of the square.

Treat the level as a function of what the code has to survive. On a screen, nothing goes wrong and L is fine. On paper, something usually does.

It recovers missing modules. It cannot recover a code whose modules are the wrong size for the viewing distance, or too low in contrast to be read at all — those are failures of the print, not of the data.

The same applies to a missing quiet zone, a code on a reflective surface, and a code smeared by motion blur. In each case the bits that arrive are confidently wrong, and redundancy has nothing to reconstruct from.