Standardized benchmark circuits, experiment protocols, and results.
benchmarks/
├── circuits/ # 10 algorithm families, 20-50 instances each
│ ├── ghz/
│ ├── qft/
│ ├── grover/
│ ├── bernstein_vazirani/
│ ├── qpe/
│ ├── qaoa_maxcut/
│ ├── vqe_ansatze/
│ ├── trotter_simulation/
│ ├── quantum_arithmetic/
│ └── random_clifford/
├── protocols/ # Experiment definitions (20 experiments)
│ ├── 01_ml_dataset_generation.yaml
│ ├── 06_fixed_vs_searched.yaml
│ ├── 11_framework_shootout.yaml
│ └── 18_noise_aware.yaml
└── results/ # Output data (results.json, trace.jsonl, dataset.parquet, summary.md, figure.png)
| Family | Sizes | Instances |
|---|---|---|
| GHZ | 4-100 qubits | 20 |
| QFT | 3-30 | 30 |
| Grover | 3-20 | 20 |
| Bernstein-Vazirani | 4-50 | 20 |
| Quantum Phase Estimation | 3-20 | 20 |
| QAOA MaxCut | multiple graph families | 50 |
| VQE ansatze | multiple molecule/ansatz types | 30 |
| Hamiltonian/Trotter simulation | varying steps | 20 |
| Quantum arithmetic | adders/multipliers | 20 |
| Random Clifford/general circuits | varying density | 50 |
Target: ~2,000 benchmark circuits.
Every benchmark run produces:
| File | Content |
|---|---|
results.json |
Structured metrics (depth, gates, time, equivalence) |
trace.jsonl |
One line per transformation step (provenance) |
dataset.parquet |
ML training data (s, a, s', r) tuples |
summary.md |
Human-readable summary with tables |
figure.png |
Visualization |
Reported as relative deltas for paired experiments:
Delta_D = (D_WQ - D_baseline) / D_baseline
Delta_G2q = (G2q_WQ - G2q_base) / G2q_base
Plus: compilation time, semantic-equivalence pass rate, target validity, dataset throughput, unique states, unique transitions.
Reported as: median/mean + bootstrap confidence intervals.