The experiment is best understood as a completed implementation with an unfinished advantage test. We know how to construct the random graph-state circuit, sample it on hardware, check its Clifford prefix and challenge several classical methods. What remains open is a common quality target that turns unlike runtimes into a fair race.
What was implemented
- a frozen 70-qubit random-graph sampling circuit;
- a Qiskit parser and structural validator for the exact QASM;
- small exact circuits for software and simulator cross-checks;
- a guarded hardware workflow for planning, validation, submission and retrieval;
- graph-state stabilizer measurements and spacetime error-detection experiments;
- classical statevector, MPS, extended-stabilizer and observable-specific baselines.
This is the main achievement of the repository: the claim has been converted into an executable, inspectable research object.
The clock result in context
The complete hardware circuit returned 256 samples in 19 QPU-seconds. A fit to locally measured extended-stabilizer runtimes at much smaller widths projects about 6.89 million years for one classical sample at 70 qubits.
The contrast is scientifically interesting because both sides refer to the same nominal width. It is not yet a quality-matched speedup for three reasons:
- the large classical time is extrapolated rather than measured at 70 qubits;
- one projected classical sample is compared with a batch of quantum samples;
- the quantum and classical outputs have not been shown to meet the same accuracy criterion.
The large ratio belongs on a results dashboard, but it should not dominate the theory or implementation story.
What the verification says
The graph-state prefix supplies efficiently measurable stabilizer observables. The hardware data support a functioning end-to-end check protocol, but the predeclared high-confidence lower-bound test did not establish a positive fidelity-style signal. Relaxing a confidence level after seeing the data is useful for sensitivity analysis, not a substitute success criterion.
The full non-Clifford samples and the Clifford-prefix stabilizer estimates are complementary evidence. Neither alone proves that the hardware distribution is closer to the ideal target than every feasible classical approximation.
How to complete the advantage test
A stronger follow-up should begin with a benchmark contract rather than another runtime:
- Choose one output task. For example, full bit-string sampling or a predeclared panel of observables.
- Choose one quality metric. On small circuits this can be distribution distance to an exact reference; on larger circuits it may be a predeclared stabilizer or observable error.
- Set the threshold before running. Both quantum and classical routes must reach the same target.
- Measure time to threshold. Include the timing layers required by the scientific question and report the others separately.
- Scale through an overlap region. Measure both routes at intermediate widths before extrapolating beyond classical reach.
This design would reveal where the classical method stops meeting the quality target while the hardware still does. Only then does “seconds versus years” become a measured time-to-solution statement rather than a same-width clock scenario.
Final perspective
Random-graph sampling is attractive because graph states combine three unusual properties: a compact graph-theoretic definition, rich entanglement, and a stabilizer algebra that remains experimentally testable. The non-Clifford measurement basis then turns that verifiable resource into a harder sampling problem.
The repository demonstrates how to build and investigate that idea on real hardware. Its current result is a serious candidate for practical quantum advantage, not the final proof.
Explore the implementation · Inspect the evidence dashboard · Return to the Advantage List