A per-job readout probe rescues qubit selection on one IBM Heron r2 processor and hurts it on another: a preregistered three-device test (TESSERACT-MAP-04)
Preregistered (doi:10.5281/zenodo.24406640) test on ibm_marrakesh, ibm_fez and ibm_kingston (6 jobs, 121 s QPU) of ranking 40-qubit paths with a few-second all-qubit readout probe instead of IBM's published readout errors. ibm_marrakesh: probe path won by +25.0 pp (F_cl 0.313 vs 0.063); the calibration path again used qubit 46 (probe 34% readout error, listed 1.1%). ibm_fez: probe path lost by 3.6 pp; Qiskit's stock layout was best. ibm_kingston: no difference, despite a 20% readout qubit listed at 1.0% on the calibration path. After readout mitigation the calibration-ranked path was better on all three devices. Verdict by the preregistered rule: helps on 1 of 3 devices; does not generalise. Undisclosed readout faults were present on 2 of 3 processors. Record contains paper, source, probe and main-job counts, calibrations, locks, QPYs, analyses 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.24407000
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- preprint