An open-source, software-agnostic analyzer and optimizer for hybrid quantum-classical projects. The system inspects a user's quantum codebase and answers:

Which planned quantum operations actually need QPU hardware?

Core distinction

Existing hybrid quantum systems often solve: Where should task T execute?

WestQuant solves the deeper problem: Should task T require a QPU at all?

This is Semantic QPU Minimization.

Quick start

pip install westquant-qcsc

Command renamed in 0.2.0. The console command is now westquant-qcsc. Versions 0.1.x installed it as westquant, which collided with the westquant umbrella package's own command. Replace westquant audit|propose|run|policy|closed-loop with westquant-qcsc …; subcommands and options are unchanged. If you keep 0.1.x, do not install it alongside the westquant package.

Audit a project

westquant-qcsc audit ./my_quantum_project

Output:

westquant_report.html    # Full HTML report
westquant_report.md      # Markdown report
westquant_plan.json      # Machine-readable plan (SAFE/BALANCED/AGGRESSIVE)
westquant_workflow.json  # WQFlow IR graph

QPU Budget Predictor

Estimate the required QPU budget from problem structure — before any quantum execution:

from qcsc import QPUBudgetPredictor, ProblemProfile

profile = ProblemProfile(
    problem="MaxCut",
    n_qubits=16,
    p=3,
    graph_family="GEO",
    quality_target=0.95,
)
predictor = QPUBudgetPredictor()
budget = predictor.estimate(profile)

print(f"Shots: {budget.recommended_shots:,}")
print(f"Reduction vs naive: {budget.reduction_vs_naive:.0f}x")
print(f"Estimated quality: {budget.estimated_quality:.4f}")
print(f"Strategies: {budget.strategies}")

Or from the CLI:

westquant-qcsc budget --problem MaxCut --qubits 16 --depth 3 --family GEO --compare

Output:

  [B0_naive]
    Shots:         2,119,680
    Est. quality:  0.9961

  [B1_best_practice]
    Shots:         1,024
    Est. quality:  0.9961
    vs naive:      2070.0x reduction

  [B2_classical]
    Shots:         0
    Est. quality:  1.0000

The predictor is calibrated from 12,600 exact simulation results across 6 problems, 7 graph families, and depths p=1,2,3. See Paper A for the underlying research.

What it detects

Three plans

• SAFE: Only EXACT transformations. Mathematical result preserved.
• BALANCED: EXACT + EQUIVALENT + low-risk APPROXIMATE. Requires tolerance ε.
• AGGRESSIVE: All transformations including EXPERIMENTAL. Always shows uncertainty.

Architecture

qcsc/
├── qpu_budget.py              # QPU budget predictor (Paper A integration)
├── flow_ir/                    # WQFlow IR — directed graph of hybrid workflows
├── resource_model/             # QPU backend + local compute capabilities
├── transformation_registry/    # 15 QPU-saving transformations (T01-T15)
├── analyzer/                   # Framework-specific source analysis (Qiskit first)
├── detectors/                  # Detectors that produce recommendations
├── planner/                    # Pareto planner (SAFE/BALANCED/AGGRESSIVE)
├── reporting/                  # HTML + Markdown + JSON report generation
├── benchmark/                  # Benchmark projects with known opportunities
├── cost_model.py               # QPU cost estimation
├── closed_loop.py              # Closed-loop measurement tracking
├── learned_policy.py           # Learned policy for transformation ranking
├── patch_generator.py         # Advisory patch generation
├── runner.py                   # Supervised plan execution
└── cli.py                      # CLI entry point

Design philosophy

WestQuant asks three questions in order:

1. Does this operation need to happen? (Eliminate)
2. Does it need to be quantum? (Replace with classical)
3. Only then: which resource should execute it? (Schedule)

That ordering is the defining idea of the project.

CUDA-Q and cuQuantum

Version 0.1.1 adds advisory capability records for CUDA-Q CPU/GPU targets and cuStateVec, cuTensorNet, cuDensityMat, cuPauliProp, and cuStabilizer. Install westquant-qcsc[cudaq] to compose QPU minimization with westquant-cudaq. The integration is domain-neutral and does not include vertical-specific scientific methods.

V0.1 — Advisory only

V0.1 performs static analysis + cost estimation + recommendations. It does NOT:

• modify user code
• execute user code
• submit jobs to hardware
• automatically rewrite experiments

Community

• Discussions: Join the conversation
• Contributing: See CONTRIBUTING.md
• Code of Conduct: See CODE_OF_CONDUCT.md

License

MIT