Shared Nonlocal-Game Probes for Small Quantum Workloads: IBM Mirror Comparisons and Cross-Vendor Coverage Tests

Can a small set of shared nonlocal-game probes stand in for one mirror circuit per workload when screening small quantum workloads? We evaluate shared-probe screening (SPS): eight GHZ/Mermin echo circuits reused across workloads, a frozen simulator-trained predictor and a suite-level residual-quantile lower bound. The target is computational-basis output-distribution quality (a Hellinger proxy), not state fidelity. Direct SPS-versus-mirror comparisons were run on IBM processors only; IQM and Rigetti runs test bound coverage. (a) On 132 IBM links in three jobs, at equal screening shots, mirrors ranked workloads better (per-circuit Spearman 0.678 vs 0.487; difference −0.191, 95% CI [−0.283, −0.115]); this replicated on a fresh calibration day (88 links, two jobs; −0.187 [−0.308, −0.082]). A post-hoc job-level analysis finds the same sign in all five jobs (t interval over jobs [−0.295, −0.089]), but these are only five calibration units on three devices. (b) Added value of the probes over vendor calibration plus circuit metadata was not established: SPS minus a metadata-only predictor was +0.069 [−0.004, +0.143] and +0.010 [−0.088, +0.115] on the two IBM days. (c) SPS was cheaper on some axes: half the circuit configurations, 0.24× the two-qubit gate executions and about 0.74× the QPU time (pro-rated from mixed jobs). (d) The suite-level bound covered IBM (30/30; 86/88 on the fresh day) and IQM (18/18, descriptive), failed on Rigetti (6/30), and was restored by one perdevice calibration (21/21, then 26/26 three days later) at about three times IBM’s width; that US$34.50 calibration is recovered after about 12 screening batches at list prices. An SPS-first cascade gave no QPU saving on hardware. Data, pre-registrations and code are openly released.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24210212
Primary Topic
Quantum Computing Algorithms and Architecture
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Shared Nonlocal-Game Probes for Small Quantum Workloads: IBM Mirror Comparisons and Cross-Vendor Coverage Tests

Asad Ali
Zenodo (CERN European Organization for Nuclear Research)
Quantum Computing Algorithms and Architecture
preprint

Shared Nonlocal-Game Probes for Small Quantum Workloads: IBM Mirror Comparisons and Cross-Vendor Coverage Tests

Asad Ali
preprint en

Abstract

Can a small set of shared nonlocal-game probes stand in for one mirror circuit per workload when screening small quantum workloads? We evaluate shared-probe screening (SPS): eight GHZ/Mermin echo circuits reused across workloads, a frozen simulator-trained predictor and a suite-level residual-quantile lower bound. The target is computational-basis output-distribution quality (a Hellinger proxy), not state fidelity. Direct SPS-versus-mirror comparisons were run on IBM processors only; IQM and Rigetti runs test bound coverage. (a) On 132 IBM links in three jobs, at equal screening shots, mirrors ranked workloads better (per-circuit Spearman 0.678 vs 0.487; difference −0.191, 95% CI [−0.283, −0.115]); this replicated on a fresh calibration day (88 links, two jobs; −0.187 [−0.308, −0.082]). A post-hoc job-level analysis finds the same sign in all five jobs (t interval over jobs [−0.295, −0.089]), but these are only five calibration units on three devices. (b) Added value of the probes over vendor calibration plus circuit metadata was not established: SPS minus a metadata-only predictor was +0.069 [−0.004, +0.143] and +0.010 [−0.088, +0.115] on the two IBM days. (c) SPS was cheaper on some axes: half the circuit configurations, 0.24× the two-qubit gate executions and about 0.74× the QPU time (pro-rated from mixed jobs). (d) The suite-level bound covered IBM (30/30; 86/88 on the fresh day) and IQM (18/18, descriptive), failed on Rigetti (6/30), and was restored by one perdevice calibration (21/21, then 26/26 three days later) at about three times IBM’s width; that US$34.50 calibration is recovered after about 12 screening batches at list prices. An SPS-first cascade gave no QPU saving on hardware. Data, pre-registrations and code are openly released.

Zenodo (CERN European Organization for Nuclear Research)
Quantum Computing Algorithms and Architecture
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon