Four days later, the best qubits were the worst: a preregistered replication on ibm_marrakesh in which readout faults invisible to the published calibration reversed a 10 pp layout effect (TESSERACT-MAP-03)
Preregistered follow-up (doi:10.5281/zenodo.24406496, published before submission) to TESSERACT-MAP-02 (doi:10.5281/zenodo.24406426). Job dc0ajv5s2omc738915k0 on ibm_marrakesh, 80 PUBs, 47 s QPU, same physical qubits as MAP-02 four days later. The 10.5 pp best-vs-median path gap did not replicate: it reversed to −4.29 pp (95% CI [−5.24, −3.33], Holm p = 6.4e-6). Cause: qubit 46 on the best path read |0> as 1 in 54% of shots while IBM's calibration listed ~1.1% readout error; q31 and q37 also degraded. After tensored readout mitigation the best path leads again by +17.0 pp. The path effect is not additive across chain and routed star (interaction −5.64 pp). Today's stock level-3 layout beat the MAP-02 best path by 1.75 pp. Layout rankings built on published calibration can be overturned within days by single-qubit readout faults; a per-job union readout check catches them. Record contains the paper, source, preregistration, lock, QPY, calibrations, raw counts, analysis and code.
Authors
- Devin Phillip Davis
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24406524
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint