Every piece of digital information — a photo, a password, a boarding pass — has to survive the real world: scratched disks, flaky Wi-Fi, wet paper. The field of data and storage is largely about one problem: how do you keep information intact when the world is constantly trying to corrupt it? Different mechanisms solve different parts of that problem. This lesson is about one of the most elegant: error-correcting codes, and the reason a QR code can lose 30% of itself and still tell your phone exactly where to go.

A QR code isn't just a picture of data. It's data plus a deliberate, calculated amount of redundancy. When engineers designed the QR standard in 1994, they built in something called Reed-Solomon error correction — a mathematical technique that encodes the same information multiple times, spread across different parts of the square, in a way that lets a reader reconstruct the whole from any sufficient fragment. Think of it this way: if you had to send someone a 10-digit phone number by shouting across a noisy street, you might say it twice, or spell it out phonetically, or give them the area code three times. Reed-Solomon does the algebraic equivalent of all of that, automatically, invisibly, inside every QR code you've ever scanned.

QR codes come in four error-correction levels: L, M, Q, and H. Level L can recover from about 7% damage. Level H can recover from 30%. The catch — and here's the tradeoff that shows up everywhere in data storage — is that higher resilience means a bigger, denser square. Level H QR codes have more black-and-white modules crammed into the same grid, because they're carrying more redundant information alongside the original data. So when you see a QR code with a company logo printed right over the center of it, that's not reckless design. The code was generated at Level H specifically to absorb that logo as if it were damage. The logo covers maybe 20% of the surface. The error correction handles it without breaking a sweat.

The deeper principle here is that information and redundancy are always in tension. A perfectly compressed file has zero redundancy — every bit carries unique information, and a single corrupted bit can ruin the whole thing. A maximally error-corrected file is the opposite: so much of it is redundant that you could destroy half and still reconstruct the original. Real systems live somewhere in the middle, tuned to the noise level of the environment they're operating in. Your QR code doesn't know it's going to get a coffee stain. It just carries enough redundancy that when it does, the math already has it covered.