Repository navigation
SCPN Quantum Control 1.3.0 adds native differentiable transform qualification, signed readout correction and uncertainty propagation, and Studio workspace and provider-evidence interfaces. It also fixes solver, tensor-network and dependency-security issues accumulated since 1.2.0.
- Native JAX, PyTorch and TensorFlow qualification includes analytic references and exact native coverage. Fixed-axis PyTorch Phase-QNode compilation retains dependency guards. The release compiler qualification uses PyTorch 2.15.0 CPU.
- Signed readout correction preserves quasi-probabilities and propagates sampling and shared calibration uncertainty.
- Studio adds source-preserving workspace transactions, method-comparison inspection, provider observation history and a capability catalogue.
- Bounded ansatz/layout comparison, MCWF sampling evidence and the pre-registered KYMA v3 results retain their explicit scientific claim boundaries.
- Public installations require companion
oscillatools >=0.1.1, including the qualified solver time, termination and finite-input fixes. - The Rust extra requires
scpn-quantum-engine >=0.3.0, including native FEP/MPC and execution-admission corrections and zeroizing signing-key custody. The engine requires the matching parent 1.3.0 host contract. - Patched dependency locks, aligned action pins and source-archive privacy checks strengthen packaging and CI.
The compiled PyTorch qualification covers Pauli observables with the eager capture backend; dense NumPy observables retain an independent compiler limitation. PyTorch 2.13 retains a compiled training-loop/functorch compatibility gap; the release compiler profile is PyTorch 2.15.0 CPU. This release makes no new quantum-advantage, hardware-performance or calibrated-coverage claim.
ML-DSA support remains a research implementation without a constant-time or FIPS-140 validation claim.
Three unpatched upstream provider advisories retain documented codec-only dispositions with a fail-closed review deadline of 5 November 2026. They are not represented as patched vulnerabilities. See the hardware guide for the exact scope.
See the full changelog for details.