Nadarasa · instrument · gate 0.4.9

Cross-experiment Selene benchmark suite

Every major experiment in this stack, re-run end to end through one harness and scored on a common scale: Simon's hidden-shift search, the ADAPT-GQE composed excitation, ethylene QPDE, the Floquet native port, the NHS discharge QUBO and the feed-forward conditional gate. Each experiment keeps its own accuracy definition — the fidelity column is that quantity normalised onto 0–1 against its exact or analytic reference.

verdict verifiedbenchmark_suite_complete3/3 criteria

Each criterion below is recomputed from the committed rows by quantum/verdicts.py; this chip reads the result rather than restating it.

passevery_experiment_scored
Every experiment in the suite produced a score.
measured 6 · threshold 6
passnothing_missing
No experiment is missing from the aggregation.
measured 0 · threshold 0
passfidelities_physical
Every reported ideal fidelity lies in [0, 1].
measured [] · threshold []

Experiments

6/6

Cells

474

Mean fidelity

99.20%

Worst 1x-H2 loss

8.3%

Results

Click a column header to sort; click a row to expand its raw metric.

	Experiment	Qubits		Shots/cell
G4	Feed-forward conditional gate	2	5	4000	99.03%	99.70%	0.30%	12.3s
G16	Ethylene QPDE (evolution-time trick)	2	20	4096	98.76%	94.70%	1.27%	2.5s↺
G20	Mixed-field Ising Floquet native port	6	384	512	99.76%	96.45%	3.32%	2.5s↺
G21	ADAPT-GQE composed H2 UCCSD excitation	2	2	512	97.66%	94.92%	4.89%	6.7s
G22	NHS discharge-flow QUBO (QAOA p=1)	12	36	512	100.00%	91.73%	8.27%	3.9s↺
G23	Simon's hidden-shift search	6-10	27	256	100.00%	99.18%	0.82%	2.5s↺

Methodology

Fidelity. Each experiment reports a different physical quantity — Hartree, a subharmonic peak, a probability deviation, a QUBO energy. The suite maps each onto agreement with its exact reference in 0–1 and keeps the raw number in the expanded row.
Noise. One H2-class depolarizing model everywhere: p_1q = 0.00003, p_2q = 0.00129, p_meas = 0.00135. QPDE, Floquet and Simon already carry full ladders; the feed-forward, ADAPT and discharge rows are probed by the suite at ideal / 1× / 5×. Degradation compares circuit fragility, not device quality — the circuits differ in depth and width.
Runtime. Selene emulator wall-clock seconds inside the build sandbox, not Quantinuum hardware time. Every driver is resumable, so a row marked ↺ was re-scored against a warm per-row cache rather than swept cold; use runtimes comparatively only.
Reproduce. Run python3 -m quantum.benchmark.runner then python3 -m quantum.benchmark.aggregate. The suite caches one JSON per experiment and re-emits src/data/demos/benchmark_suite.json.

Generated 2026-08-13T07:20:03+00:00 · Quest state-vector · TKET compile lane · back to the frontier map