Documented capabilities

  1. 01QRegister, local-component, stabilizer, phase-graph, symmetry, tensor, and observable routes
  2. 02Reusable operation, tensor-contraction, Pauli-estimator, and parameter-sweep plans
  3. 03Exact reverse-mode adjoint gradients, bounded scheduling, causal pruning, and point batching
  4. 04Fail-closed execution broker, general exact fallback, and QSC v1 state transport

QSA 0.2.0

Measured release workloads

Results apply to the named benchmark and representation. They are not a universal multiplier for arbitrary circuits.

01

Fused native numerical kernels

Recorded real, complex, reduction, and small-matrix workloads vs matched NumPy work

4.52x-7.96x
02

Repeated 18-qubit exact tensor queries

Reusable bounded-width contraction plan vs rebuilding and direct contraction

32.6x
03

Repeated 100-qubit exact tensor queries

Reusable bounded-width contraction plan vs rebuilding and direct contraction

36.4x
04

18-qubit exact adjoint gradient

Best execution ratio vs matched Aer statevector parameter shift; setup plus first execution was 5.79x

96.2x
05

100-qubit exact adjoint gradient

Best execution ratio vs exact Aer MPS parameter shift; setup plus first execution was 4.51x

364x

Release validation

Numerical, compatibility, packaging, and hostile-input checks

Open validation record ↗
  1. 01Linux, macOS, and Windows build gates
  2. 02Randomized dense-reference comparisons across structural routes
  3. 03Exact tensor, estimator, parameter-sweep, and gradient checks
  4. 04QSC v1 round-trip and hostile-input coverage
  5. 05AddressSanitizer and UndefinedBehaviorSanitizer gates
  6. 06C ABI 1.5 compatibility, archive rebuild, and package checks