Define the logical event

A logical error rate can refer to a memory experiment, a syndrome-extraction cycle, a logical gate, or an entire algorithmic fragment. Rates per round, per unit time, and per operation are not interchangeable. The experiment may report bit-flip and phase-flip channels separately or combine them through a model.

Write down the protected object, number of rounds, and success criterion before comparing values. A lower per-round memory error does not by itself establish a better logical gate, and a short experiment may not expose slow drift or leakage that appears over longer operation.

Check the scaling evidence

Fault-tolerant proposals rely on a regime in which increasing code size suppresses logical error under stated assumptions. An experiment showing that a larger code performs better than a smaller one is therefore different from a single encoded qubit beating one selected physical component. Both can be informative, but they test different claims.

For a scaling result, inspect how many code distances or sizes were measured, whether hardware and noise conditions were comparable, and whether confidence intervals separate the trend. Threshold values depend on the code, decoder, operation set, and noise model; they are not universal device specifications.

Account for all resources

Physical-qubit count is only the visible part of the overhead. Syndrome rounds take time, decoders use classical computation, routing and lattice surgery require space, and state preparation or distillation can dominate a useful workload. Postselection can improve a reported conditional result by discarding runs, but the acceptance probability then becomes part of the cost.

A fair comparison reports measurement fidelity, gate set, connectivity, cycle duration, decoder latency, leakage handling, and the treatment of lost or rejected shots. If one experiment uses real-time feedback and another decodes offline, that distinction matters for claims about operation.

  • Match logical operation and error unit.
  • Compare code size, distance, and number of syndrome rounds.
  • Record decoder and whether decoding was real time.
  • Include rejected shots and state-preparation cost.
  • Separate demonstrated components from projected system overhead.

Read the result as a chain of conditions

The most accurate summary often has a conditional form: for a stated code, device, circuit, decoder, and experiment duration, the measured logical quantity behaved in a particular way. This may sound less dramatic than a universal error-rate claim, but it tells another researcher what would need to be reproduced.

Use the paper’s supplementary information for decoder settings and exclusion rules, and look for raw or derived data where available. Compare results only across aligned columns; leave a blank rather than convert incomparable quantities into a single leaderboard.