Qubit counts alone can’t tell you how close quantum computers are to breaking encryption. The machine also has to correct its errors fast enough to finish the attack, and that depends partly on software running on ordinary computers. Last week showed why: Iceberg Quantum estimated RSA-2048 could be broken with 72,000 physical qubits in 3.7 days, assuming a 1µs correction cycle. Its decoder averages 4µs per round. How decoding keeps pace with that assumed cycle rate is an important implementation question. Quantinuum reported an estimated logical-memory error rate of 5 × 10⁻⁶ per correction cycle, about five errors per million rounds. It decoded the experimental data offline. The same paper benchmarks decoders fast enough for real time, but the experiment didn’t use them live. Leiden and Riverlane found standard correlated decoding erring about 4.4 times more often than simpler matching in one tested configuration with very low hardware error rates. They also proposed a fix. When a smaller attack estimate makes headlines, check the conditions: 1. How reliable must the qubits be? Iceberg’s estimate assumes a physical error rate of roughly one in 1,000. 2. Has the complete error-correction system demonstrated the speed the estimate assumes?