Showcase¶
examples/showcase.py is the five-minute demo of what SuperQuantX does
today. It runs in about 10 seconds on a laptop, entirely on local
simulators, and checks every quantum result against something classical.
- Backends. Bell and GHZ circuits on every installed gate-model backend. Only the correct outcomes may appear.
- Quantum machine learning. A quantum kernel SVM on
make_moons, compared with a tuned RBF SVM and logistic regression using the protocol of arXiv:2610.42029. See Benchmarking. - Optimisation. QAOA on a 5-node MaxCut instance, checked against brute force.
Real output¶
Captured on 2 November 2026 (Python 3.11, PennyLane 0.45, Qiskit 2.6, Cirq 1.7, Braket SDK 1.127, pytket 2.18). Timings vary by machine; shot counts vary with sampling.
SuperQuantX 0.2.0 showcase. All runs are local classical simulations.
1. Bell and GHZ states on every backend (1000 shots each)
=========================================================
backend Bell 00/11 GHZ 000/111 correct time
simulator 495/505 507/493 yes 0.00s
pennylane 492/508 510/490 yes 0.01s
qiskit 511/489 507/493 yes 0.19s
cirq 517/483 485/515 yes 0.01s
braket 485/515 522/478 yes 0.03s
tket 491/509 497/503 yes 0.03s
Working: simulator, pennylane, qiskit, cirq, braket, tket
2. Quantum kernel SVM vs tuned classical models (make_moons, pennylane)
=======================================================================
model accuracy wall time grid QPU circuits
----------------------------------------------------------------------------------
QuantumSVM (PennyLane), tuned 0.930 +/- 0.051 3.43s 25 119,000
QuantumSVM (PennyLane), untuned 0.670 +/- 0.125 0.58s 1 -
SVC (RBF), tuned C and gamma 0.950 +/- 0.055 0.85s 25 -
Logistic regression, tuned C 0.860 +/- 0.097 0.19s 4 -
----------------------------------------------------------------------------------
100 samples, 2 features, 5-fold stratified CV (identical folds for every model), seed=0
quantum minus best classical: -0.020; no significant difference from SVC (RBF), tuned C and gamma (p=0.54)
Protocol: arXiv:2610.42029. Quantum kernels are classically simulated.
3. QAOA MaxCut vs brute force (5 nodes, 7 edges, p=2, pennylane)
================================================================
QAOA best sampled cut : 01101 value 6
Brute-force optimum : 01101 value 6
Expected cut of state : 5.019 (approximation ratio 0.837; a random cut averages 3.5)
Matches optimum : yes (484 circuit evaluations, 2.4s)
Total time 7.8s
What this shows, and what it does not¶
- Six SDKs run the same circuits through one API and agree, after bit-order normalisation.
- The tuned quantum kernel is competitive with classical models on this dataset, but not better: the difference from a tuned RBF SVM is not statistically significant. That matches the published evidence. Bandwidth tuning, not the circuit, is what lifts it from 0.67 to 0.93.
- QAOA at depth 2 finds the optimal cut among its samples. Its expected cut (approximation ratio about 0.84) is what would matter on hardware, and brute force is instant at this size.
- None of this is quantum advantage. All circuits are classically simulated.
Agents demo¶
examples/quantum_agents_demo.py runs the portfolio agent (CVaR-VQE on a
QUBO, checked against brute-force enumeration and a Markowitz reference) and
the classification agent (which wraps compare_to_classical).