Noise is inevitable — so qubits must be protected. But the two obvious defenses are both banned: you can’t copy an unknown qubit (chapter 7a), and you can’t look at it to check (chapter 03). Quantum error correction threads that needle — spotting and fixing errors without ever reading the data.
The escape is to stop storing the information in any single qubit. Spread one logical qubit across three physical ones — not as copies (that’s forbidden), but woven together so the meaning lives in how they agree. Then you never ask “what value is qubit 2?” You ask only “do qubits 1 and 2 still match?” — a question about agreement, not content. That single clever shift is the whole idea.
Three friends are told a secret word and sent into separate rooms. You’re not allowed to ask any of them the word (that would let it leak). But you are allowed to ask two of them, through the wall: “does your word match your neighbour’s?” If friends 1&2 say match but 2&3 say differ, you know friend 3 mis-heard — without ever learning the word. Tell friend 3 to fix theirs and the secret is whole again.
The logical qubit α|000〉 + β|111〉 is stored below. Click a qubit to hit it with a bit-flip. Two parity checks — “does 1=2?” and “does 2=3?” — light up a syndrome that pinpoints the culprit without ever revealing α or β. Then correct it.
The two checks are the operators Z₁Z₂ and Z₂Z₃ — the stabilizers. They report whether neighbouring qubits agree, and crucially they commute with the logical information, so reading them disturbs nothing. The two answers form a 2-bit address of the error:
Apply the matching X and the error undoes itself. This three-qubit code only catches bit-flips; stack it with a phase-flip version and you get Shor’s nine-qubit code — and modern surface codes scale the same idea across a whole lattice.
“No-cloning makes quantum error correction impossible.” It would — if the code copied the qubit. It doesn’t. α|000〉+β|111〉 is a single entangled state, not three copies of α|0〉+β|1〉; and the syndrome reads only parities, never α or β. Both bans stay intact.
The catch: correction only wins if your physical qubits are already good enough. Below a certain error rate per gate — the fault-tolerance threshold — adding more physical qubits drives the logical error down. Above it, you just add more things to break. That threshold is what the hardware of the final chapter is racing to beat.