The oldest error-correcting trick in the book is also the dumbest, and that is exactly why it is a good place to start. If you want to send a single bit down a noisy wire and you worry it might flip, send it three times. Instead of 0, transmit 000. If a burst of noise corrupts one copy and the receiver sees 010, a majority vote decides the intended value was 0. This is the classical repetition code, and it works because you can make copies freely and read them whenever you like.
A qubit can do neither. You cannot photocopy an unknown quantum state, a rule called the no-cloning theorem. And you cannot simply peek at a qubit to check on it, because measuring a superposition collapses it and destroys the very information you were trying to protect. The repetition code had to be rebuilt from scratch to survive in this world, and how it was rebuilt shaped everything that came after.
Spreading a qubit thin
The quantum version starts with a different kind of redundancy. Rather than copying a state, you entangle one logical qubit across three physical qubits so that the information lives in the correlations between them, not in any single one. You never store the answer in a place you can point to. You store it in how the qubits agree with each other.
Checking for errors then becomes a question of agreement rather than value. You bring in extra helper qubits, called ancillas, and use them to ask a careful question: do qubit one and qubit two hold the same value, and do qubit two and qubit three hold the same value? These are parity checks, and the beauty is that the answer reveals nothing about whether the data is a 0 or a 1. It only reveals whether two neighbors disagree. Measuring the ancilla tells you "they match" or "they differ" without ever collapsing the protected information.
If both checks come back clean, nothing has flipped. If the first check says two neighbors disagree and the second says they agree, the middle qubit is the odd one out and you know exactly which one drifted. A classical computer running alongside the machine takes those parity results, called the syndrome, and works out the correction. This is the same majority-vote logic as the radio trick, translated into a language that never looks at the message directly.
The catch that broke the toy
Here is where the quantum world extracts its price. A classical bit can only fail one way: it flips from 0 to 1 or back. A qubit can fail two ways. It can suffer a bit flip, swapping its poles, or a phase flip, scrambling the delicate sign relationship that makes superposition and interference possible. The three-qubit repetition code catches bit flips beautifully and is completely blind to phase flips. Protect against one and you leave the door wide open for the other.
You can build a mirror-image version that guards phase instead, but then bit flips walk right in. The repetition code, on its own, is only ever half a solution. That single fact is why quantum error correction is so much harder than the classical kind, and why it took decades of theory to escape the trap.
Why it still matters
The escape came from nesting the two versions together into two-dimensional layouts, most famously the surface code, which watches for both flavors of error at once. But the repetition code never left. Hardware teams still run it constantly, because it is the cleanest possible stress test. Google, among others, has used repetition-code experiments to prove a crucial claim: that adding more physical qubits to a logical qubit actually lowers the error rate rather than raising it. If you cannot show suppression on the simplest code, you have no business attempting a harder one.
Those experiments also expose the ugly surprises that theory glosses over, like correlated errors that strike several qubits at once and fool the majority vote, or the leakage of a qubit out of its two computing states entirely. The repetition code is where engineers first see whether their machine behaves the way the math promises.
Say it three times and take a vote. The idea is almost embarrassingly simple. Making it work on hardware that refuses to be copied or observed is the entire story of quantum error correction in miniature.
