Question. Which parts of the Steane stack have run on hardware, and which stronger conclusions do those experiments not imply?
The fixed seven-qubit block has supported a progression from encoded states to active recovery, inter-block gates, universal protocol components, and code switching. None of those steps turns Steane into a growing-distance family.
| work | platform | demonstrated boundary | not established by that result |
|---|---|---|---|
| Nigg et al. | seven trapped ions | a full seven-qubit color-code/Steane logical qubit and encoded operations | repeated below-threshold scaling with distance |
| Ryan-Anderson et al. | trapped-ion QCCD | repeated syndrome extraction with real-time classical lookup and feedback | an intrinsic threshold for the fixed lookup decoder |
| Postler et al. | trapped ions | fault-tolerant Clifford operations combined with logical magic-state preparation/injection | a transversal Steane non-Clifford gate |
| Ryan-Anderson et al. | trapped-ion QCCD | encoded entangling gates and Bell-pair preparation between code blocks | zero routing or recovery overhead on another architecture |
| Bluvstein et al. | reconfigurable neutral atoms | many small color-code blocks and transversal inter-block operations in a movable-atom processor | increasing the distance of a Steane block |
| Postler et al. | trapped ions | repeated encoded-ancilla QEC and an experimental comparison with flag extraction | a circuit-independent ranking of extraction methods |
| Paetznick et al. | trapped-ion QCCD | repeated correction and better-than-physical metrics for tested logical qubits including Steane | a universal normalization shared by every memory experiment |
| Daguerre et al. | trapped ions | Reed–Muller-to-Steane magic-state switching and logical certification | native Steane T transversality |
Each boundary retains the code blocks, ancilla protocol, decoder, noise, and metric of its source. “Fault tolerant,” “better than physical,” and “universal” describe different acceptance tests and must not be merged into one scalar ranking.
The Steane checks have weight four, and transversal two-block gates require seven corresponding physical pairings. Architectures with all-to-all interactions, ion transport, or atom rearrangement can realize those pairings without embedding the block in a fixed nearest-neighbor square grid. This is an architectural fit, not proof that connectivity and movement are free: transport, rearrangement, crosstalk, leakage, and idle faults remain inside each experiment's circuit model.
The card therefore names trapped-ion and reconfigurable-neutral-atom platforms without turning that fit into a hardware exclusivity claim.
For a scaling memory claim, increasing a code parameter must increase distance while the experiment compares consistently normalized logical performance. constructor.py has no parameter and always returns distance three. Repeating a recovery cycle more times, running more blocks, or improving device fidelity does not create a distance sequence.
Concatenating Steane blocks or moving to larger triangular color codes can create a scaling family, but then the family, decoder, and syndrome gadgets must be named. The T11 realization sweep found no qualifying demonstration of below-threshold scaling obtained by increasing “Steane distance,” because that parameter does not exist for the fixed block.
All rows above are literature claims; the Atlas constructor cannot recompute a device experiment. The formal verifier establishes the encoded object to which those protocols refer, while the citations establish what each platform executed.
The T11 ranked experimental set contains trapped-ion and neutral-atom anchors. A platform not listed in card.yaml is an honest unverified gap, not evidence that a realization is impossible. A future addition must identify a literal seven-qubit Steane/color-code demonstration and preserve whether it showed detection, active correction, a logical gate, or a scaling result.