Syndrome decoding is the classical step that converts the syndrome measurements produced by a quantum error correction code into a decision about what correction to apply. A syndrome tells you that a parity check was violated — that an error occurred somewhere — but not which error occurred, since many distinct error patterns produce the same syndrome. The decoder's job is therefore inference: given the observed syndrome history and a noise model, find the most likely error consistent with it. Crucially this must happen in real time. Syndromes stream out of the device every error-correction cycle, roughly every microsecond on superconducting hardware, and a decoder slower than that rate builds an unbounded backlog that grows without limit and eventually destroys the computation. Common decoder families include minimum-weight perfect matching for surface codes, union-find for speed, and belief propagation with post-processing for qLDPC codes. The tradeoff between speed and accuracy is harsh: a faster but less accurate decoder raises the logical error rate and can push an otherwise viable device back above threshold. The classical co-processor is genuinely part of the quantum computer.