scpn-quantum-control

Active Development — scpn-quantum-control is under intensive development. The public status wording is anchored to the Hardware Status Ledger, which separates theory, simulator, unmitigated hardware, mitigated hardware, and noise-limited claims. Current promoted hardware evidence is narrowed to artefact-backed ibm_fez baseline rows, the April/May 2026 ibm_kingston DLA parity raw-count datasets, and the May 2026 SCPN/FIM falsification artefacts. Stable core contracts and backend capability artefacts are now part of release/repro hardening and are kept separate from non-artefact scientific claims. APIs may evolve as this work progresses.

Version: 0.10.0 Status: Kuramoto-XY compiler + hardware runners + analysis stack + bounded differentiable-programming surface | generated capability inventory below | CI coverage gate 90% | IBM Heron r2 evidence ledgered

Honest scope — read first. At the system sizes reachable today (n ≤ 16 qubits) classical ODE and exact solvers are faster and more accurate than the quantum routes here; there is no demonstrated broad quantum advantage. The value of the quantum approach is characterisation (entanglement, MBL, synchronisation witnesses, DLA parity) and auditable hardware evidence, not speed. See the Limitations section below for the full disclosure.

What this repository is for

scpn-quantum-control is an evidence-governed workbench for turning coupled-oscillator models into quantum-control experiments, simulator studies, gradient-bearing optimisation loops, and auditable hardware-result packages. It is for teams that need a repeatable path from a physical model to a result that can be reviewed, reproduced, cited, or used in an application pilot.

If your team needs reproducible research or product-grade experimentation, the repository provides:

• a stable route from K_nm/omega problem definitions to simulators;
• deterministic artefact-led evidence for method claims;
• bounded optimisation and differentiable-programming surfaces for training, verification, and convergence analysis;
• hardware campaign management that separates committed raw-count rows from simulator and scoped-failure classes.

It is positioned for use-cases where decision quality depends on evidence: engineering teams that need a consistent pathway from model specification to observable contracts, and research teams that need explicit failure modes rather than implied coverage.

What you can do first

Need	First path	Output
Understand the software	Onboarding -> Quickstart	A small local Kuramoto-XY run and a clear claim boundary.
Bring your own coupled system	Physics-First Kuramoto-XY	A validated K_nm/omega problem compiled to simulator-ready quantum objects.
Train or inspect gradients	Differentiable Tutorials -> Differentiable Programming -> Quantum Gradients	Exact, finite-shot, framework-comparison, or fail-closed gradient evidence.
Review hardware claims	Hardware Status Ledger -> Hardware Result Packs	Raw-count-backed evidence or an explicit blocked promotion route.
Use v0.10 control surfaces	API Overview -> Tutorials -> Example Gallery	QRNG health checks, PQC trigger signing, UltraScale+ HLS emission, realtime telemetry, Studio federation, and sensing workflows.
Evaluate adoption	API Overview -> Release Readiness Gate	Stable integration surfaces, release gates, and licensing boundaries.

Application and commercial value

The practical value is not "quantum black box" experimentation; it is the ability to reduce integration risk while preserving future hardware options. This package is designed for organisations that want to:

• de-risk algorithm ideas on simulators and strong classical baselines first;
• compare evidence classes before claiming hardware-level results;
• standardise proof surfaces (contracts, manifests, and ledgers) across pilots;
• move between R&D notebooks and integration-friendly stable facades without collapsing into undocumented internal APIs.

Application lanes:

• Synchronisation diagnostics — explore where oscillator networks lock, decohere, or separate into sectors.
• Control prototyping — map power-grid, plasma, EEG/MEG, Josephson-array, and other coupled-system candidates into common K_nm/omega workflows, with topology-only candidates separated from measured-magnitude claims.
• Hardware evidence management — keep raw-count evidence, simulator output, and open claims separated before public release or paper citation.
• Differentiable computation and gradient-informed optimisation — supported compiler-AD primitives and parameter-shift building blocks with fail-closed unsupported paths (full surface: the Differentiable Programming Route below).
• Product route — AGPL for open research; proprietary deployment uses the commercial licence route described below.

Commercial value comes from reducing unclear research risk: every promoted result must have a named artefact, every unsupported route fails closed, and closed-source or SaaS use has a defined commercial licensing route. Open boundaries remain explicit: the package does not claim broad quantum advantage, clinical validation, or externally validated SCPN biology — it provides a reproducible computational workbench and the governance required to promote claims only when the evidence exists.

scpn-quantum-control Capability Inventory

Surface	Current inventory
Package version	0.10.0
Public API exports	835
Python source modules	518
Public Python classes	1003
Domain package families	31
Rust PyO3 function bindings	177
Rust source modules	46
Notebook files	100
Example files	37
Optional extras	43
Python test files	893
Public documentation pages	269
GitHub Actions workflows	24

Evidence boundary: this snapshot is a static inventory. Performance, coverage, hardware, and scientific-fidelity claims require their own committed evidence artefacts.

---

Status Snapshot — 2026-06-26

Area	Public status
Generic compiler surface	scpn_quantum_control.kuramoto_core validates arbitrary K_nm/omega inputs and compiles Hamiltonians, dense matrices, Trotter circuits, and order-parameter measurements.
v0.10 public surfaces	QRNG streaming and health reports, ML-DSA-65 trigger signing, UltraScale+ HLS pulse emission, realtime loop telemetry, NV magnetometry simulation, FRC pulsed-shot QAOA scheduling, control-scope boundary docs, and Studio federation manifests plus evidence bundles.
Release and reproducibility scope	Stable core contracts and backend capability artefacts for Kuramoto-XY synchronisation are included in release/readiness checks and promoted only with deterministic evidence manifests.
Hardware evidence	ibm_fez baseline rows are legacy artefact-backed observations; ibm_kingston Phase 1, Phase 2 A+G, Phase 2 B-C, and popcount DLA datasets are promoted with raw-count artefacts. The SCPN/FIM ibm_kingston result is promoted as a negative/falsification result for the tested digital circuit family.
Simulator and methods evidence	BKT, OTOC, Floquet, MBL, FIM, VQE, GPU, tensor-network, and classical comparison claims stay marked as simulator/classical/methods unless a hardware artefact is named. Generated benchmark artefacts are indexed from the benchmark dashboard and reproducibility CLI.
Licence boundary	The possible lightweight core split is documented, but all in-repository code remains AGPL/commercial unless a future release changes metadata and SPDX headers.

For claim classes, raw-artefact pointers, and promotion rules, see the Hardware Status Ledger.

Plain-Language Summary

scpn-quantum-control turns a coupled-oscillator network into quantum circuits, simulator workflows, hardware-result ledgers, and analysis tools. SCPN here denotes the Scale-Coupled Phase Network — a generalised-Kuramoto coupling model whose K_nm matrix provides a structured, reproducible test problem. The first supported physics lane is the Kuramoto-XY mapping: provide a coupling matrix K_nm and oscillator frequencies omega, then compile the matching XY Hamiltonian, run local or provider-backed circuits, measure synchronisation, and compare the result against classical and exact-simulation baselines.

The repository is not only a circuit generator. It also carries the operational surfaces needed to use the software responsibly:

• stable public facades for notebooks and integrations;
• Rust acceleration for hot numerical kernels;
• differentiable-programming kernels and conformance rows for supported scalar, vector, matrix, broadcast, selection, shape, assembly, cumulative, and reduction primitives;
• no-QPU release gates and hardware-result pack verifiers;
• claim-boundary documentation that separates simulator, hardware, and open scientific questions.

Who It Is For

User	Primary value	Start here
Quantum algorithm researcher	Compile oscillator networks into XY circuits and inspect synchronisation, entanglement, topology, and control observables.	Quickstart, Tutorials
Applied physicist or control engineer	Convert domain coupling graphs into reproducible quantum/simulator experiments while retaining classical baselines.	Physics-First Kuramoto-XY, Application Benchmark Plugins
Hardware experiment operator	Run or replay provider-backed campaigns only when the evidence ledger, raw counts, and release gates permit the claim.	Hardware Guide, Hardware Result Packs
Software integrator	Use stable facade APIs instead of binding to internal package layout.	Stable Facades API, API Overview
Commercial evaluator	Assess the route from research prototype to product lane: reproducible simulations, hardware-result governance, Rust kernels, and dual licensing.	Onboarding, Release Readiness Gate

Why Teams Would Adopt It

The package is useful when a team needs a single accountable route from a coupled-system model to quantum-control evidence:

Need	What the repository supplies	Boundary to respect
Turn oscillator data into quantum experiments	K_nm/omega validation, XY Hamiltonians, circuits, and local execution paths.	The mapping is a quantum XY analogue of synchronisation dynamics, not a proof of nonlinear classical equivalence for every system.
Optimise trainable objectives	Parameter-shift VQE, composed phase objectives, gradient certificates, and supported compiler/program-AD kernels.	Unsupported gates, adapters, backends, shapes, and transform nests fail closed.
Compare simulators, hardware, and papers	Raw-count ledgers, result packs, no-QPU gates, benchmark manifests, and reproducibility docs.	Claims must cite committed evidence artefacts; simulator output is not hardware validation.
Prepare a product or research integration	Stable facades, provider capability records, Rust acceleration hooks, API docs, tutorials, and licence boundary text.	Closed-source network services or embedded products need the commercial licence route.

Market-facing applications include quantum-control prototyping, power-grid and plasma synchronisation studies, EEG/MEG rhythm modelling under non-clinical claim limits, quantum-hardware campaign governance, gradient-informed VQE/QNN research, and reproducible benchmark publication.

Differentiable Programming Route

The differentiable-programming lane is now documented as a first-path product surface because gradient evidence is central to quantum optimisation, machine learning integration, and control. The bounded Phase-QNode promotion state is a promotion candidate until the claim ledger, external comparison rows, and isolated CI benchmark artefacts all pass:

Layer	Current status	Where to start
Quantum parameter-shift	Supported through scpn_quantum_control.phase.param_shift for callable expectation objectives, structured PhaseVQE gradients, and local gradient-descent VQE examples.	Quantum Gradients, Variational Methods
Program and compiler AD	Supported for registered scalar, vector, and matrix primitives with native lowering reports, deterministic alias/effect metadata summaries including bounded local rebinding/list-alias metadata, typed list-alias provenance, typed loop-carried state provenance, typed control-path alias provenance, typed rebinding-alias provenance, local object-attribute aliases, expression-rebinding aliases, branch-local control-path alias blockers, loop-carried scalar state metadata, executed array-view aliases, program_ad_effect_ir.v1 round-trip conformance through program_ad_ir_roundtrip_contracts, runtime/source control-region and ProgramADPhiNode conformance through program_ad_control_phi_metadata_contracts, finite/dtype/shape result checks, trainable-mask derivative zeroing, bounded reverse-adjoint generation provenance with ProgramADAdjointStep rows through program_adjoint_replay_provenance_contracts, fail-closed derivative-losing sign/heaviside contracts, elementwise primitive conformance for bounded absolute-value, domain, denominator, and inverse-trig boundary contracts, structured numeric primitive conformance for product/interpolation/signal/stencil contracts, cumulative primitive conformance for bounded cumsum/cumprod/diff traces, assembly primitive conformance for like-constructors and static stack conveniences, reduction primitive conformance for bounded statistical, trapezoid, unique-selector, and scalar-q order-statistic reductions, shape primitive conformance for bounded reshape, axis movement, rank promotion, repeat/tile, roll/rot90, and flip-family contracts, broadcast primitive conformance for bounded broadcast_to, broadcast_arrays, and binary rank-broadcasting paths, selection primitive conformance for bounded static selection folds, strict sort, where, and clip paths with dynamic masks/selectors, ties, and integer-output selector differentiation still fail-closed, bounded 2x2 static-linalg native lowering, wider static-linalg MLIR-runtime-only verification, and fail-closed unsupported boundaries. Alias/effect summaries, IR round-trip evidence, control/phi provenance, generated adjoint-step provenance, and the source/bytecode frontend execution gate that attaches accepted reports to whole-program results and rejects hard gaps before objective execution with unsupported-semantics source/region/bytecode diagnostics are local bounded evidence, not malformed control/view/list/loop/rebinding-alias promotion, non-executed branch adjoints, full compiler phi lowering, a complete static alias lattice, executable Rust/LLVM/JIT lowering, or full reverse-mode compiler AD.	Differentiable Programming, Differentiable API
Gradient support matrix	Executable support planning now covers registered gates, observables, backends, transforms, and ML/provider adapters with explicit blocked reasons and alternatives.	Quantum Gradients, Differentiable API
Unified readiness ledger	run_differentiable_readiness_audit() aggregates the support matrix, transform nesting, QNode tape/transform suites, provider gradients, hardware policy, and provider hardware-preparation audit into one reviewer-facing pass/fail ledger.	Differentiable Programming, Differentiable API
Transform nesting governance	Executable planning now separates supported local grad, value_and_grad, hessian, nested-grad, tape, scalar jvp, scalar vjp, scalar jacfwd, scalar jacrev, vector-output native Jacobian execution, native manual vmap(grad), whole-program grad(vmap(f)) over trace-aware leaves, local Hessian over a whole-program AD scalar objective, JVP/VJP over whole-program AD Hessian transforms, and provider-callback QNode transforms from blocked framework-vectorized, adapter-nested, finite-shot curvature, malformed-provider, and hardware nesting routes.	Quantum Gradients, Differentiable API
Provider-gradient readiness	Executable audit evidence distinguishes deterministic callbacks, finite-shot callbacks, multi-frequency rules, hardware-blocked routes, unknown backends, malformed finite-shot samples, and policy-bound hardware-preparation records.	Quantum Gradients, Differentiable API
Hardware-gradient policy readiness	Executable dry-run policy decisions now gate hardware-gradient preparation by provider/backend allowlist, shot budget, required evidence IDs, and live-execution ticket status. prepare_provider_hardware_parameter_shift_gradient(...) packages that approval into provider-preparation evidence, and run_provider_hardware_gradient_preparation_audit() verifies supported and blocked preparation routes without submitting QPU jobs.	Quantum Gradients, Differentiable API
Differentiable claim ledger	The Phase-QNode evidence ledger maps implementation, tests, artefact IDs, documentation, known gaps, and promotion status; no promoted claim is accepted without an artefact ID, and support-surface alignment checks keep ledger paths consistent with the generated capability manifest.	Differentiable Programming, Claim Ledger
Differentiable public claim table	Public-facing differentiable wording is generated from the committed ledger. Every current row is bounded-candidate only, and the table blocks hardware, provider, QPU, GPU, production-performance, and isolated_affinity claims until promotion evidence exists.	Public Claim Table
Differentiable baseline scorecard	run_differentiable_baseline_scorecard() scores the lane against named JAX, PyTorch, PennyLane, Qiskit Runtime, Catalyst, Enzyme, Rust Program AD, provider/hardware, benchmark, docs/API, and adoption baselines. Every current category remains behind-baseline governance evidence until promoted ledger rows and isolated benchmark artefacts exist.	Differentiable Programming, Baseline Scorecard
Differentiable Rust/Python inventory	run_differentiable_rust_python_inventory() classifies differentiable Python, Rust, compiler, provider, hardware, metadata, and deprecation surfaces before broad rustification. Rows record owner modules, tests, docs, benchmark status, mypy targets, docstring status, Rust parity, polyglot status, and blockers without promoting Rust, LLVM/JIT, provider, hardware, GPU, or isolated benchmark claims.	Differentiable Programming, Rust/Python Inventory
Differentiable external-validation lock	The external-validation package records exact SHA-256 digests for runtime, development, Python 3.11-3.13 CI, CPU framework-overlay, and Enzyme-runner lockfiles. The artefact is reviewer reproduction evidence only and remains functional_non_isolated.	Differentiable Programming, Environment Lock
Differentiable CI reproducibility	The differentiable framework workflow runs sparse and full CPU profiles across Python 3.11-3.13, enforces the module-specific test audit, uploads scheduled benchmark metadata, and exposes a manual optional GPU contract lane that remains functional_non_isolated.	Differentiable Programming, Workflow
Differentiable artefact bundle	The external-validation package records a reproducible manifest over the committed claim ledger, public claim table, environment lock, domain dataset closure, gradient comparison, maturity audit, and local benchmark evidence. The bundle is checksum provenance only and remains functional_non_isolated.	Artefact Bundle
Differentiable isolated benchmark plan	run_differentiable_isolated_benchmark_plan() maps every current non-isolated differentiable benchmark/evidence artefact, including the compiler-promotion batch gate, to a reserved-host rerun command, required runner labels, expected output paths, and explicit blockers. It is planning evidence only and returns promotion_ready=False until validated isolated_affinity artefacts exist.	Batch Plan, Benchmark API
Differentiable external-validation report	The technical report summarizes the comparison package, provider-family status, reproducibility artefacts, and remaining promotion blockers without upgrading any row beyond bounded-candidate evidence.	External Validation Report
Hardening-slice gate	run_differentiable_hardening_slice_gate(...) records the required Ruff, mypy, module-specific pytest, test-quality audit, claim-ledger validation, and benchmark-classification checks for each differentiable hardening slice. CI, local preflight, and the pre-push hook additionally enforce a module-specific strict-mypy ratchet across the closed differentiable API, claim-ledger, benchmark-evidence, QNN/QGNN/QSNN training and evidence satellites, objective/domain evidence, optimizer-baseline, backend selection, parameter-shift/VQE foundations, structured-ansatz/methodology/benchmark/Kuramoto/UPDE solver foundations, typed trajectory-result containers, layered ADAPT-VQE, Trotter-error bounds, framework-overlay, provider/hardware-gradient safety, Phase-QNode, framework-bridge, transform-nesting, external-comparison, XY compiler, and PennyLane import modules while repository-wide strict mode remains open debt. The same gates now enforce a scoped NumPy-style Ruff docstring ratchet for the differentiable external-validation, module-hardening audit, and hardening-slice gate surfaces while repository-wide docstring enforcement remains open debt. It is checklist/classification evidence only, not benchmark execution.	Differentiable Programming, Differentiable API
Module-hardening audit	run_differentiable_module_hardening_audit() discovers every differentiable/gradient/QNode/bridge/compiler module in the promotion scope and verifies a module-specific test plus declared fail-closed diagnostics for each.	Differentiable Programming, Differentiable API
Bounded phase-QNN, open-system, and coupling-recovery evidence	A deterministic data-reuploading binary classifier is available through train_parameter_shift_qnn_classifier(...) with multi-frequency parameter-shift descent, prediction evidence, accuracy, convergence certificates, finite-difference gradient verification, seeded finite-shot gradient uncertainty and noisy-convergence evidence, optional named external-gradient agreement records, a conformance suite with unsuitable-scenario evidence, deterministic convergence suites, bounded PyTorch custom torch.autograd.Function backward plus SGD integration audit, bounded PyTorch module state_dict, Adam optimizer-state replay, CPU/CUDA-smoke-gated device-state replay, weights-only CPU checkpoint replay, long-lived checkpoint matrix diagnostics, multi-scenario training-loop matrix diagnostics, local torch.export save/load value replay, static export-shape matrix diagnostics, input-driven dynamic-batch torch.export replay, and local AOTAutograd forward/backward FX graph persistence through run_torch_autograd_function_audit(...), run_torch_module_state_audit(...), run_torch_module_device_state_audit(...), run_torch_module_checkpoint_audit(...), run_torch_long_lived_checkpoint_matrix(...), run_torch_training_loop_matrix(...), run_torch_module_export_audit(...), run_torch_export_shape_matrix(...), run_torch_dynamic_shape_export_audit(...), and run_torch_aot_autograd_export_audit(...), non-isolated optimizer-baseline comparisons across parameter-shift, finite-difference, SGD, Adam, L-BFGS-B, diagonal-Fisher natural-gradient, seeded SPSA, and derivative-free grid routes, known-ground-state convergence certificates across natural-gradient, Adam, L-BFGS-B, seeded SPSA, and COBYLA through run_ground_state_optimizer_convergence_suite(...), bounded Lindblad/MCWF objective rows with density-matrix invariant and same-seed trajectory replay certificates through run_open_system_objective_suite(...), bounded Kuramoto and XY coupling time-series recovery evidence through run_coupling_recovery_suite(...), and caller-supplied framework-gradient agreement checks.	Quantum Gradients, Differentiable API, Lindblad, Open-System Hardware
Registered Phase-QNode family	Local statevector execution, density-matrix execution with bounded single-qubit Kraus channels, arbitrary-depth registered circuit builders with deterministic depth/resource profiles, registered GHZ-chain and hardware-efficient multi-qubit templates, controlled-H/S/T plus Toffoli/CCZ/Fredkin gates with exact Toffoli/Fredkin decompositions, sparse Ising-chain Hamiltonian construction, parameter-shift gradients for pure-state routes, framework parity rows, native JAX deterministic statevector value-and-gradient plus grad/value_and_grad/jacfwd/jacrev/hessian/jvp/vjp/vmap/jit transform lowering for registered local circuits, native JAX PyTree transform lowering with flattened Hessian symmetry evidence for structured registered local circuit parameters, native JAX pmap sharding transform lowering with one row per local device, native PyTorch deterministic statevector value-and-gradient lowering, native PyTorch torch.func.grad/jacrev/vmap transform lowering, native PyTorch non-fullgraph torch.compile value-and-gradient lowering on CPU, PyTorch compile-boundary diagnostics, verified SCPN MLIR-runtime lowering adapters, and isolated-affinity benchmark metadata are available for the declared gate/observable subset. Unsupported gates, dynamic/provider paths, native LLVM/JIT lowering, interpreter fallback success, noisy-channel gradients/metrics, registered PyTorch fullgraph torch.compile promotion, dynamic-shape compile promotion, registered Phase-QNode AOTAutograd/export persistence, dynamic-shape export promotion, incompatible CUDA/device execution, finite-shot native framework lowering, and unregistered observables fail closed with support reports.	Differentiable API, Benchmark Harness
ML framework and tape roadmap	Gradient tape, QNode-style tape records, backend gradient planning, provider-safe callback execution with shot/variance accounting, convergence certificates, optional JAX host-callback parameter-shift interop, deterministic registered Phase-QNode JAX statevector transform lowering, deterministic registered Phase-QNode JAX PyTree transform lowering, deterministic registered Phase-QNode JAX pmap/sharding transform lowering, deterministic registered Phase-QNode PyTorch statevector, torch.func, non-fullgraph torch.compile transform lowering, compile-boundary diagnostics, bounded PyTorch module-state, device-state, checkpoint, long-lived checkpoint matrix, training-loop matrix, export replay, export-shape matrix diagnostics, dynamic-batch export replay, and local AOTAutograd FX graph persistence, PyTorch module/transform/compiler/device maturity routing, PennyLane gradient-agreement checks, TensorFlow host-boundary tensor bridges, and bounded framework parity rows are available. Full provider-backed QNode migration bridges, finite-shot native framework lowering, dynamic-circuit lowering, compatible CUDA/device artefacts, registered PyTorch fullgraph torch.compile promotion, cross-runtime AOTAutograd execution, dynamic-shape AOTAutograd export, dynamic feature-width export promotion, cross-runtime checkpoint/export portability, and arbitrary architectures remain staged surfaces, not yet advertised as production-complete.	Differentiable Roadmap

This matters commercially because optimisation users do not only need circuits. They need gradients, convergence evidence, framework interop, reproducible benchmarks, and clear failure modes. The SCPN route aims to combine Kuramoto-XY physics, quantum-control objectives, hardware-result governance, and compiler-backed AD under one support matrix.

Rust polyglot parity includes a claim-bounded Program AD IR metadata parser in scpn_quantum_engine::program_ad_ir plus program_ad_effect_ir_metadata_summary(...) and program_ad_effect_ir_interpret_forward(...) plus program_ad_effect_ir_interpret_value_and_gradient(...) for PyO3 consumers. program_ad_registry_metadata_mirror(...) validates the Python registry coverage snapshot, returns deterministic family/facet counts, and records only the primitive-name overlap with the already bounded Rust scalar/static-linalg plus compact interpolation, compact signal, compact stencil, compact cumulative, elementwise/static-structural, static diag/diagflat, static matrix_power, fixed multi_dot, 2x2 distinct symmetric eigvalsh, 2x2 distinct symmetric eigh eigenvalues/nonzero-offdiagonal eigenvectors, 2x2 real-distinct eigvals, 2x2 real-simple eig eigenvalue/eigenvector replay, and 2x2 distinct-positive svd(..., compute_uv=False) singular-value replay plus constant-full-rank rank-1/Nx2/2xN pinv replay. Metadata summaries validate program_ad_effect_ir.v1 evidence only; Rust value+gradient replay is bounded to opcode-bearing scalar, elementwise-array, static structural-array, static structural-assembly, static source-map indexing, compact interpolation, compact signal, compact stencil, compact cumulative, static product, corrected moment, strict order-statistic, and compact static-grid trapezoid reductions with dx/x/xfull metadata, plus static integer np.diag gather/scatter nodes, static on-diagonal np.diagflat construction nodes, static integer np.linalg.matrix_power output nodes, fixed-signature np.linalg.multi_dot matrix-chain output nodes, 2x2 distinct symmetric np.linalg.eigvalsh spectral output nodes, 2x2 distinct symmetric np.linalg.eigh eigenvalue and nonzero-offdiagonal eigenvector output nodes, 2x2 real-distinct np.linalg.eigvals spectral output nodes, 2x2 real-simple np.linalg.eig eigenvalue/eigenvector output nodes, and 2x2 distinct-positive np.linalg.svd(..., compute_uv=False) singular-value output nodes plus constant-full-rank rank-1/Nx2/2xN np.linalg.pinv output nodes, including executed runtime branch metadata when matched by runtime phi provenance. It still fails closed on legacy opcode-free metadata, aliases, mutation, non-lowered dynamic indexing semantics, dynamic axes, dynamic trapezoid-grid metadata, dynamic q/method metadata, dynamic ddof/correction metadata, zero-variance std gradients, broad linalg/spectral array adjoints beyond the bounded 2x2 eigvalsh, eigh, eigvals, and real-simple eig, static rank-2 SVD singular-value, and rank-1/Nx2/2xN pinv boundaries, source-level/non-executed branch semantics, general Program AD execution, LLVM/JIT execution, hardware, provider, and performance promotion. The Rust claim boundary now reports the BL-02 dynamic_boundary_fail_closed_audit for the audited dynamic-boundary fail-closed corpus. Python callers can use scpn_quantum_control.program_ad_rust_bridge for the typed fail-closed wrappers; scpn_quantum_control.differentiable re-exports the same symbols for backward compatibility. Static whole-program bytecode/source frontend inspection now lives in scpn_quantum_control.whole_program_frontend; scpn_quantum_control.differentiable and the package root re-export compile_whole_program_frontend(...) and its report objects for compatibility. The frontend remains no-execution preflight metadata. whole_program_value_and_grad(...) now requires a frontend_ready report before objective execution and attaches that report to WholeProgramADResult.frontend_report; hard gaps fail closed with the frontend digest plus source/region/bytecode diagnostics. This is not executable Rust, LLVM, JIT, provider, hardware, or benchmark evidence. Python compiler interchange lowers captured program_ad_effect_ir.v1 records into deterministic scpn_diff.program_ad_* MLIR-style operations through compile_whole_program_ad_trace_to_mlir(...), validated by program_ad_mlir_interchange_contracts. This remains metadata interchange, not executable Rust, LLVM, JIT, provider, hardware, or performance evidence.

The first production-grade differentiable workflows are deliberately bounded:

1. train small VQE objectives with parameter-shift gradients through PhaseVQE.solve(gradient_method="parameter_shift");
2. verify gradients against finite differences and analytic references through verify_parameter_shift_gradient(...) and verify_vqe_parameter_shift_gradient(...);
3. train bounded phase-QNN classifiers through train_parameter_shift_qnn_classifier(...), verify their QNN-specific gradients through verify_parameter_shift_qnn_classifier_gradient(...), record seeded finite-shot uncertainty through estimate_parameter_shift_qnn_finite_shot_gradient(...), package evidence with run_parameter_shift_qnn_conformance_suite(...), certify deterministic local convergence with run_parameter_shift_qnn_convergence_suite(...), replay seeded finite-shot convergence with run_parameter_shift_qnn_finite_shot_convergence_suite(...), verify bounded PyTorch custom-autograd backward, module, optimizer, device-state, checkpoint, long-lived checkpoint matrix, training-loop matrix, export replay, static export-shape matrix diagnostics, dynamic-batch export replay, and local AOTAutograd FX gradient replay with run_torch_autograd_function_audit(...), run_torch_module_state_audit(...), run_torch_module_device_state_audit(...), run_torch_module_checkpoint_audit(...), run_torch_long_lived_checkpoint_matrix(...), run_torch_training_loop_matrix(...), run_torch_module_export_audit(...), run_torch_export_shape_matrix(...), run_torch_dynamic_shape_export_audit(...), and run_torch_aot_autograd_export_audit(...), compare local optimizer baselines with run_parameter_shift_qnn_optimizer_benchmark_suite(...), certify known-ground-state optimizer convergence with run_ground_state_optimizer_convergence_suite(...), evaluate bounded Lindblad/MCWF objective evidence with run_open_system_objective_suite(...), and record caller-supplied framework-gradient agreement with verify_parameter_shift_qnn_framework_agreement(...);
4. execute registered local Phase-QNode circuits with execute_phase_qnode_circuit(...), compare installed framework parity with run_phase_qnode_framework_parity_suite(), lower deterministic registered statevector value-and-gradient routes into native JAX with jax_phase_qnode_value_and_grad(...), audit registered JAX native transforms with jax_phase_qnode_native_transform_audit(...), audit structured-parameter JAX PyTree transforms with jax_phase_qnode_pytree_transform_audit(...), audit local-device JAX pmap/sharding transforms with jax_phase_qnode_sharding_transform_audit(...), lower deterministic registered statevector value-and-gradient routes into native PyTorch with torch_phase_qnode_value_and_grad(...), audit registered PyTorch torch.func transforms with torch_phase_qnode_transform_audit(...), classify PyTorch compile boundaries with torch_phase_qnode_compile_boundary_audit(...), record PyTorch module/transform/compiler/device maturity with run_torch_ecosystem_maturity_audit(...), and lower supported subsets to textual MLIR metadata with lower_phase_qnode_circuit_to_mlir(...);
5. use compiler/program-AD kernels for supported classical objectives;
6. evaluate hardware-gradient preparation with evaluate_hardware_gradient_policy(...) and run_hardware_gradient_policy_readiness_suite() before any provider job is prepared;
7. summarize the focused readiness suites with run_differentiable_readiness_audit();
8. document unsupported gates, backends, shapes, and dynamic program paths before they can mislead users.

Future releases will extend this route toward native framework gradients beyond the current bounded JAX/PyTorch/TensorFlow and registered JAX/PyTorch statevector bridges, full PennyLane/Qiskit migration bridges beyond agreement checks, finite-shot and provider-backed native lowering, quantum neural networks, analog oscillator mappings, open-system gradients, benchmark leaderboards, and real-time feedback control.

Richer Presentation

For a richer presentation of the Phase 1 hardware results, methodology deep-dives, interactive plots, and architecture diagrams, see the project website:

anulum.li/scpn-quantum-control

Direct entry points:

• Onboarding — what the project is, who should use it, what is mature, and what remains claim-bound
• Hardware Status Ledger — claim classes, campaign evidence paths, and publication hygiene rules
• Hardware Result Packs — offline manifest and integrity verifier for promoted IBM raw-count datasets
• Physics-First Kuramoto-XY — start from arbitrary oscillator networks before SCPN-specific layers
• Differentiable Programming — current AD surface, support boundaries, and user routes
• Quantum Gradients — parameter-shift, VQE gradients, verification tests, and planned backend gradient planner
• Differentiable API — public scpn_quantum_control.differentiable reference and usage map
• Differentiable Roadmap — staged plan for tape, framework adapters, QNN/QGNN/QSNN, analog mapping, benchmarks, verification, and dashboards
• Stable Facades API — mkdocstrings reference for first-path public facades
• Phase 1 Results — raw-count reproduction of the DLA parity asymmetry on ibm_kingston, April 2026, with full Welch table and interactive Plotly plot
• Reproducibility Manifest — per-commit pinning, IBM job IDs, dependency constraints, rerun protocol
• Method: GUESS Mitigation — symmetry-guided ZNE, shot-budget-free for the XY Hamiltonian
• Method: DLA Parity Theorem — $\mathfrak{su}(2^{n-1}) \oplus \mathfrak{su}(2^{n-1})$ decomposition and hardware reproduction path
• Method: Pulse Shaping — ICI three-level and (α, β)-hypergeometric Rust fast paths (the 1,665× and 44× figures on the linked page are v0.9.5-era workstation measurements, not reproduced by a committed benchmark artefact)
• The Science — plain-language primer on SCPN, Kuramoto-XY, and why the DLA parity result matters
• Methods Benchmark Dashboard — generated Rust/VQE, GPU, tensor-network, FIM, and reproducibility artefact index
• Roadmap — canonical active work queue and completed release-safety tasks

---

Quick Start

pip install scpn-quantum-control

import numpy as np
from scpn_quantum_control.phase.xy_kuramoto import QuantumKuramotoSolver

# 8 oscillators, exponential-decay coupling, heterogeneous frequencies
N = 8
K = 0.5 * np.exp(-0.3 * np.abs(np.subtract.outer(range(N), range(N))))
omega = np.linspace(0.8, 1.2, N)

# Simulate: Trotter evolution → order parameter R(t)
solver = QuantumKuramotoSolver(N, K, omega)
result = solver.run(t_max=1.0, dt=0.1)
print(f"Final R = {result['R'][-1]:.3f}")
# → R rises from ~0.3 (incoherent) toward 1.0 (synchronised)

No IBM credentials needed — runs on local statevector simulator. Pass any coupling matrix; the built-in SCPN benchmark is just one example.

Any coupling network — bring your own K and omega:

from scpn_quantum_control import QuantumKuramotoSolver, build_kuramoto_ring

K, omega = build_kuramoto_ring(6, coupling=0.5, rng_seed=42)
solver = QuantumKuramotoSolver(6, K, omega)
result = solver.run(t_max=1.0, dt=0.1, trotter_per_step=2)
print(f"R(t): {result['R']}")

Built-in SCPN network (16 oscillators from Paper 27):

from scpn_quantum_control import QuantumKuramotoSolver, build_knm_paper27, OMEGA_N_16

K = build_knm_paper27(L=4)
solver = QuantumKuramotoSolver(4, K, OMEGA_N_16[:4])
result = solver.run(t_max=0.5, dt=0.1, trotter_per_step=2)

Detect synchronisation with witness operators:

from scpn_quantum_control.analysis.sync_witness import evaluate_all_witnesses

# After running X-basis and Y-basis circuits on IBM hardware:
results = evaluate_all_witnesses(x_counts, y_counts, n_qubits=4)
for name, w in results.items():
    print(f"{name}: {'SYNCHRONIZED' if w.is_synchronized else 'incoherent'}")

For development (editable install with test/lint tooling):

pip install -e ".[dev]"
pre-commit install
pytest tests/ -v

Hardware execution (requires IBM Quantum credentials)

pip install -e ".[ibm]"
python run_hardware.py --experiment kuramoto --qubits 4 --shots 10000

---

What This Package Does

A Kuramoto-XY compiler and hardware-evidence workbench for heterogeneous coupled oscillators. The repository contains legacy ibm_fez baseline artefacts, promoted ibm_kingston DLA parity datasets, and SCPN/FIM falsification artefacts with raw counts, job IDs, integrity checks, and count-to-statistic reproduction harnesses.

The package provides:

1. A Kuramoto-to-quantum compiler — any coupling matrix K_nm and natural frequencies omega compile directly into executable Qiskit circuits for IBM hardware. Rust-accelerated dense Hamiltonian construction is faster than the Qiskit SparsePauliOp path for small systems (a parity-checked local regression guard, ~96× at L=4 on the i5-11600K baseline, shrinking with system size) — reproducible and gated; see Native Speedup Benchmark.

2. Tracked research module families probing the synchronisation phase transition — synchronisation witnesses, OTOC scrambling, Krylov complexity, persistent homology, DLA parity theorem, and more. Novel constructions and first applications are documented in the research-gems and API pages; exact file counts use the package table below.

3. Hardware evidence with claim classes — legacy ibm_fez baseline rows, promoted ibm_kingston DLA parity datasets, and the SCPN/FIM negative hardware result are separated from simulator-only, frontier, queued-job, and aggregate-only outputs. Stable core contracts and backend capability artefacts are included in this hardening boundary and are replayed via reproducibility tooling.

Think of it as a quantum microscope for synchronisation: classical Kuramoto tells you when oscillators lock in step; this package tells you what the quantum state looks like at the transition, how entangled it is, how fast information scrambles, and whether the system thermalises.

Advanced benchmark: The built-in SCPN 16-layer coupling matrix (Paper 27) provides a heterogeneous-frequency benchmark for structured oscillator-network experiments. Publication-facing claims should treat this as a classical complex-network input to quantum-inspired Hamiltonian, tensor-network, topological, and DLA analyses, not as a quantum-biological or clinical causation claim. See SCPN Foundations.

Key Results

Hardware Evidence

Result	Value
ibm_kingston DLA parity Phase 1	342 circuits, raw counts, job IDs, integrity checks, and reproduction harness in data/phase1_dla_parity/
ibm_kingston DLA parity Phase 2	Promoted A+G n=4 replication, B-C mixed n=6,8 scaling, and popcount controls in data/phase2_dla_parity/, data/phase2_scaling_bc/, and data/phase2_popcount_control/; no DLA-parity-only or monotone-scaling claim
ibm_kingston SCPN/FIM hardware test	Pilot and repeated follow-up artefacts in data/scpn_fim_hamiltonian/; promoted as a negative result for simple digital lambda=4 hardware protection on the tested circuit family
ibm_fez baseline rows	Legacy QPU observations retained in results/ibm_hardware_2026-03-28/ and results/march_2026/; quote only artefact-backed values through the ledger
Quarantined IBM outputs	V2/frontier/queued-job/aggregate-only artefacts are not promoted until raw counts, retrieval manifests, and analysis code are reviewed

Simulation

Result	Value
Critical coupling K_c(∞)	≈ 2.2 (BKT finite-size scaling)
DTC with heterogeneous ω	15/15 amplitudes show subharmonic response
OTOC scrambling	4× faster at K=4 vs K=1 (n=8)
Schmidt gap transition	K = 3.44 (n=8, 60-point resolution)
DLA dimension formula	2^(2N-1) − 2 (exact, all N)

Software

Metric	Value
Rust engine bindings	177 exported #[pyfunction] bindings in the tracked Rust crate; low-level helper fn definitions are an implementation detail.
Source package surface	Tracked Python source files under src/scpn_quantum_control (excluding package initialisers) — the current count lives in the generated capability inventory above.
Research module families	Analysis, phase, hardware, bridge, mitigation, QEC, applications, forecasting, and benchmark families; exact current counts are listed in the package map below.
Publication figures	17 (simulation + hardware, including the Phase 1 DLA parity panels and exact-simulation crossover)
Test suite	CI-gated suite at a 90% aggregate coverage gate (--cov-fail-under=90); the non-refactor source tree is at 100% line coverage. See the generated capability inventory above for the current tracked test-file count.
Reproducibility CLI	scpn-bench reproduce-methods, scpn-bench fim-all, and scpn-bench all regenerate committed methods/FIM artefacts without IBM submission

Exact-Simulation Wall-Time (Not broad quantum-advantage claim)

This section covers exact Hilbert-space simulation crossover only. No broad observable-level quantum-advantage claim is closed yet. Any broader advantage claim remains blocked until external tensor-network or GPU baselines and explicit data-loading plus state-preparation costs are accounted for.

No quantum advantage at n ≤ 16 in this exact-simulation path. Classical ODE is faster for all accessible sizes. The value of the quantum approach is characterisation (entanglement, MBL, witnesses), not speed.

Method	n=4	n=8	n=12	n=16
Classical Kuramoto ODE (scipy)	0.4 ms	1.4 ms	2.8 ms	~11 ms
Exact diagonalisation (numpy eigh)	0.1 ms	164 ms	26.8 s	OOM (32 GB)
Qiskit statevector	~50 ms	~2 s	~minutes	impractical
Rust Hamiltonian + numpy eigh	0.02 ms	30 ms	~5 s	~2 min (est.)
IBM hardware (per-job, 4000 shots)	~5 s	~10 s	~20 s	~40 s

Measured on Ubuntu 24.04, AMD Ryzen, 32 GB RAM. Rust speedup applies to Hamiltonian construction only; the eigh bottleneck is LAPACK in all cases.

Publications

• Preprint: Quantum Kuramoto-XY on 156-qubit processor
• Paper: Synchronisation Witness Operators (novel NISQ-ready formalism)
• Paper: DLA Parity Theorem (exact closed-form)

Background: Kuramoto → XY Mapping

Any network of N coupled Kuramoto oscillators can be represented by a linear Kuramoto-XY Hamiltonian analogue or embedding. This is not a claim that a gate-model circuit directly Trotterises the nonlinear classical Kuramoto ODE; direct nonlinear simulation requires an explicit Koopman, Carleman, or equivalent embedding. The built-in SCPN example uses 16 oscillators with a coupling matrix K_nm:

K_nm = K_base * exp(-alpha * |n - m|)

with K_base = 0.45, alpha = 0.3, and empirical calibration anchors (K[1,2] = 0.302, K[2,3] = 0.201, K[3,4] = 0.252, K[4,5] = 0.154). Cross-hierarchy boosts link distant layers (L1-L16 = 0.05, L5-L7 = 0.15). See docs/equations.md for the full parameter set.

The 16×16 K_nm coupling matrix. White annotations: calibration anchors from Paper 27 Table 2. Cyan annotations: cross-hierarchy boosts (L1↔L16, L5↔L7). Exponential decay with distance is visible along the diagonal.

The classical dynamics follow the Kuramoto ODE:

d(theta_i)/dt = omega_i + sum_j K_ij sin(theta_j - theta_i)

The working quantum analogue uses the XY Hamiltonian

H = -sum_{i<j} K_ij (X_i X_j + Y_i Y_j) - sum_i omega_i Z_i

where X, Y, Z are Pauli operators. Superconducting transmon devices can compile XX+YY interactions through native-gate decompositions, making quantum hardware a useful test bed for the corresponding Hamiltonian dynamics. The order parameter R — a measure of global synchronization — is extracted from qubit expectations: R = (1/N)|sum_i (<X_i> + i<Y_i>)|.

Coherence R as a function of coupling strength K_base across 16 SCPN layers. Strongly-coupled layers (L3, L4, L10) synchronize first; weakly-coupled L12 lags behind, consistent with the exponential decay in K_nm.

Reference: M. Sotek, Self-Consistent Phenomenological Network: Layer Dynamics and Coupling Structure, Working Paper 27 (2025). Manuscript in preparation.

Hardware Results (ibm_fez, February 2026)

Experiment	Qubits	Depth	Hardware	Exact	Error
VQE ground state	4	12 CZ	-6.2998	-6.3030	0.05%
Kuramoto XY (1 rep)	4	85	R=0.743	R=0.802	7.3%
Qubit scaling	6	147	R=0.482	R=0.532	9.3%
UPDE-16 snapshot	16	770	R=0.332	R=0.615	46%
QAOA-MPC (p=2)	4	--	-0.514	0.250	--

Full results with all 12 decoherence data points: results/HARDWARE_RESULTS.md

Key findings:

• VQE with K_nm-informed ansatz achieves 0.05% error on 4-qubit subsystem
• Coherence wall at depth 250-400 on Heron r2 — shallow Trotter (1 rep) beats deep Trotter on NISQ devices

More Trotter repetitions improve mathematical accuracy but increase circuit depth. On NISQ hardware, decoherence from the extra gates outweighs the Trotter error reduction. Optimal strategy: fewest reps that capture the physics.

• 16-layer UPDE snapshot on real hardware — per-layer structure partially tracks coupling topology (L12 collapse, L3 resilience at the extremes; Spearman rho = -0.13 across all layers)

Per-layer X-basis expectations from the 16-qubit UPDE snapshot on ibm_fez. L12 (most weakly coupled) shows near-complete decoherence; strongly-coupled layers (L3, L4, L10) maintain coherence.

• 12-point decoherence curve from depth 5 to 770 with exponential decay fit

Hardware-to-exact ratio R_hw/R_exact vs circuit depth. The three regimes: near-perfect readout (depth < 25), linear decoherence (85-400), and noise-dominated (> 400).

Package Map

Counts below are tracked Python source files under src/scpn_quantum_control, excluding package initialisers.

graph TD
    subgraph Foundation
        bridge["bridge/ (13)\nK_nm → Hamiltonian\ncross-repo adapters"]
    end

    subgraph "Core Physics"
        phase["phase/ (29)\nTrotter, VQE, ADAPT-VQE\nVarQITE, Floquet DTC"]
        analysis["analysis/ (61)\nWitnesses, QFI, PH\nOTOC, Krylov, magic"]
    end

    subgraph "Applications"
        control["control/ (11)\nQAOA-MPC, residual VQLS-GS proxy\nPetri nets, ITER"]
        qsnn["qsnn/ (7)\nQuantum spiking\nneural networks"]
        apps["applications/ (15)\nFMO, power grid\nJosephson, EEG, ITER"]
    end

    subgraph "Hardware & QEC"
        hw["hardware/ (63)\nIBM runner, backends\nGPU offload, cutting"]
        mit["mitigation/ (12)\nZNE, PEC, DD\nZ2 post-selection"]
        qec["qec/ (13)\nToric code, surface code\nrep code, error budget"]
    end

    subgraph "Field Theory"
        gauge["gauge/ (5)\nWilson loops, vortices\nCFT, universality"]
        crypto["crypto/ (9)\nQKD + PQC\nML-DSA signing"]
    end

    bridge --> phase
    bridge --> analysis
    bridge --> control
    bridge --> qsnn
    phase --> analysis
    phase --> apps
    hw --> phase
    mit --> hw
    qec --> hw
    analysis --> gauge

    style bridge fill:#6929C4,color:#fff
    style analysis fill:#d4a017,color:#000
    style phase fill:#6929C4,color:#fff

Loading

Subpackage	Modules	Purpose
analysis	61	Synchronisation probes: witnesses, QFI, PH, OTOC, Krylov, magic, BKT, DLA
hardware	63	IBM Quantum runner, plugin backends registry, AsyncHardwareRunner, trapped-ion backend, GPU offload, circuit cutting, fast sparse, qubit mapper (DynQ), provenance
phase	29	Time evolution: Trotter, VQE, ADAPT-VQE, VarQITE, AVQDS, QSVT, Floquet DTC, Lindblad
applications	15	FMO photosynthesis, power grid, Josephson array, EEG, ITER, quantum EVS, QRC+ESN baseline
bridge	13	K_nm → Hamiltonian, cross-repo adapters (sc-neurocore, SSGF, orchestrator)
control	11	QAOA-MPC, residual-certified VQLS Grad-Shafranov proxy, Petri nets, ITER disruption, topological optimiser
mitigation	12	ZNE, PEC, dynamical decoupling, Z2 parity, CPDR, symmetry verification, GUESS, compound
qec	13	Toric code, repetition code UPDE, surface code, biological surface code, error budget, multi-scale, syndrome flow
benchmarks	7	Classical vs quantum scaling, MPS baseline, GPU baseline, AppQSim
crypto	9	Entanglement QKD, topology authentication, ML-DSA signing, key hierarchy
identity	6	VQE attractor, coherence budget, entanglement witness, fingerprint
qsnn	7	Quantum spiking neural networks (LIF, trace STDP, synapses, dynamic coupling, training, neuromorphic bridge)
gauge	5	U(1) Wilson loops, vortex detection, CFT, universality, confinement
psi_field	4	U(1) compact lattice gauge: lattice, infoton, observables, SCPN mapping
ssgf	4	SSGF quantum integration
accel	3	Multi-language dispatcher + Julia tier (Rust → Julia → Python fallback chain)
dla_parity	4	DLA parity helpers and campaign analysis support
fep	2	Friston Free Energy Principle: variational free energy, predictive coding
forecasting	1	Held-out synchronisation forecasting over hardware traces and source-backed topology replays
benchmark_harness	4	Reproducible benchmark harness entry points
tcbo	1	TCBO quantum observer
pgbo	1	PGBO quantum bridge
l16	1	Layer 16 quantum director

Data Flow

The pipeline from coupling matrix to measurement follows a fixed sequence:

graph LR
    A["K_nm\ncoupling matrix"] --> B["knm_to_hamiltonian()\nSparsePauliOp"]
    B --> C["Trotter / VQE\nQuantumCircuit"]
    C --> D["Transpile\nnative gates"]
    D --> E["Execute\nAer / IBM"]
    E --> F["Parse counts\n⟨X⟩, ⟨Y⟩, ⟨Z⟩"]
    F --> G["Order parameter\nR(t)"]

    style A fill:#6929C4,color:#fff
    style G fill:#2ecc71,color:#000

Loading

Examples

31 standalone scripts in examples/:

#	Script	What it demonstrates
01	qlif_demo	Quantum LIF neuron: membrane → Ry rotation → spike
02	kuramoto_xy_demo	4-oscillator Kuramoto dynamics, R(t) trajectory
03	qaoa_mpc_demo	QAOA binary MPC: quadratic cost → Ising Hamiltonian
04	qpetri_demo	Quantum Petri net: tokens evolve in superposition
05	vqe_ansatz_comparison	Three ansatze benchmarked on 4-qubit Hamiltonian
06	zne_demo	Zero-noise extrapolation with unitary folding
07	crypto_bell_test	CHSH inequality violation certification
08	dynamical_decoupling	DD pulse sequence insertion (XY4, X2, CPMG)
09	classical_vs_quantum_benchmark	Scaling crossover analysis
10	identity_continuity_demo	VQE attractor basin stability
11	pec_demo	Probabilistic error cancellation
12	trapped_ion_demo	Ion trap noise model comparison
13	iter_disruption_demo	ITER plasma disruption classification
14a	frc_pulsed_shot_qaoa_demo	FRC pulsed-shot QAOA schedule
14b	quantum_advantage_demo	Advantage threshold estimation
15	qsnn_training_demo	QSNN training loop with parameter-shift
16	fault_tolerant_demo	Repetition code UPDE
17	snn_ssgf_bridges_demo	Cross-repo bridge roundtrips
18	end_to_end_pipeline	Complete K_nm → IBM → analysis pipeline
19	sync_witness_operator	Synchronisation witness operator demo
20	quantum_persistent_homology	Persistent homology analysis
21	biological_qec_scpn16	Biological surface code on 16-layer SCPN
22	quantum_neuromorphic_bridge	QSNN quantum LIF + trace STDP + dynamic coupling bridge
23	differentiable_api_workflow	Unified differentiable API, diagnostics, compiler report, and bounded QNN training
24	differentiable_benchmark_reproduction	Local differentiable benchmark evidence bundle reproduction with non-isolated classification
25	qrng_streaming_quickstart	QRNG stream-health checks
26	nv_magnetometry_20T_demo	NV-centre 20 T calibration surface
27	pqc_trigger_signer_demo	ML-DSA trigger signing
28	pulse_to_hls_quickstart	UltraScale+ HLS pulse source generation
29	kuramoto_handbook_workflow	Kuramoto facade diagnostics, stability, clusters, and coupling design
30	diff_first_path	Canonical differentiable namespace path and compatibility facade

All examples run on statevector simulation (no QPU needed).

Notebooks

100 tracked Jupyter notebooks in notebooks/ — including the core tutorials and retained investigation notebooks. Core notebooks:

#	Notebook	Level	Key Output
01	Kuramoto XY Dynamics	Beginner	R(t) trajectory, quantum-classical overlay
02	VQE Ground State	Beginner	Energy convergence, ansatz comparison
03	Error Mitigation	Intermediate	ZNE extrapolation plot
04	UPDE 16-Layer	Intermediate	Per-layer R bar chart
05	Crypto & Entanglement	Intermediate	CHSH S-parameter, QKD QBER
06	PEC Error Cancellation	Advanced	PEC vs ZNE, overhead scaling
07	Quantum Advantage	Advanced	Scaling crossover prediction
08	Identity Continuity	Advanced	Attractor basin, fingerprint
09	ITER Disruption	Domain	Feature distributions, accuracy, CONTROL bridge contract
10	QSNN Training	Advanced	Loss curve, weight evolution
11	Surface Code Budget	Advanced	QEC resource estimation
12	Trapped Ion Comparison	Advanced	Noise model comparison
13	Cross-Repo Bridges	Integration	Phase roundtrip, adapter demos
48	Kuramoto Handbook Workflow	Intermediate	Phase 5 facade workflow summary

All run on local AerSimulator. No IBM credentials needed.

Architecture

scpn_quantum_control/
├── analysis/       59 modules — synchronisation probes
├── hardware/       63 modules — IBM runner, backends, GPU, cutting, provenance
├── phase/          76 modules — time evolution + variational + Lindblad
├── bridge/         14 modules — K_nm → quantum objects + cross-repo
├── applications/   14 modules — physical system benchmarks
├── control/        14 modules — QAOA-MPC, residual VQLS-GS proxy, Petri, ITER, topological
├── mitigation/     12 modules — ZNE, PEC, DD, Z2, CPDR, symmetry
├── qec/            13 modules — error correction + biological surface code
├── benchmarks/     21 modules — performance baselines
├── identity/        6 modules — identity continuity analysis
├── qsnn/            7 modules — quantum spiking neural networks + neuromorphic bridge
├── crypto/          9 modules — entanglement QKD, topology authentication, ML-DSA signing, key hierarchy
├── gauge/           5 modules — U(1) gauge theory probes
├── ssgf/            4 modules — SSGF quantum integration
├── tcbo/            1 module  — TCBO quantum observer
├── pgbo/            1 module  — PGBO quantum bridge
├── l16/             1 module  — Layer 16 quantum director
└── scpn_quantum_engine/  Rust crate (PyO3 0.29, 177 exported PyO3 bindings)

Dependencies

Package	Version	Purpose
qiskit	>= 2.2,<3.0	Circuit construction, transpilation
qiskit-aer	>= 0.15,<1.0	Statevector + noise simulation
numpy	>= 1.24	Array operations
scipy	>= 1.10	Sparse linear algebra, optimisation
networkx	>= 3.0	Graph algorithms (QEC decoder)

Optional:

• matplotlib >= 3.5 for visualisation
• qiskit-ibm-runtime >= 0.40,<1.0 for hardware execution
• cupy >= 12.0 for GPU-accelerated simulation

Limitations

• NISQ benchmarking only. Current hardware results are proof-of-concept. Circuit depths >400 hit the Heron r2 coherence wall; the 16-layer UPDE snapshot (46% error) confirms this. Real tokamak control requires <1 ms deterministic latency on radiation-hardened hardware — cloud QPUs cannot provide that.
• SCPN is an unpublished model. The 16-layer coupling structure comes from a 2025 working paper (Sotek, Paper 27) with no external citations yet. The Kuramoto→XY mapping is standard physics; the specific K_nm parameterisation is not independently validated.
• Small-scale broad advantage not demonstrated. At N=4-16 qubits, classical ODE solvers outperform quantum simulation in both speed and accuracy. The exact Hilbert-space crossover is a resource boundary, not a demonstrated general advantage.
• IBM hardware claim hygiene. Do not cite queued-job, placeholder, aggregate-only, or frontier JSON as hardware validation. The promoted raw-count campaign is data/phase1_dla_parity/; legacy ibm_fez observations must name their committed artefact row.

Documentation

Full docs at anulum.github.io/scpn-quantum-control:

• Onboarding — project purpose, user routes, application value, and claim boundaries
• Installation — pip install + all optional extras
• Quickstart — first experiment in 5 minutes
• Differentiable Tutorials — runnable gradient workflow with diagnostics, compiler report, and bounded QNN training
• Differentiable Programming — bounded AD surface, gradients, compiler kernels, and roadmap boundaries
• Quantum Gradients — parameter-shift and gradient-evidence route for VQE and quantum-control objectives
• Differentiable API — public differentiable namespace and support matrix
• Differentiable Roadmap — staged gradient, adapter, benchmark, verification, and control roadmap
• Tutorials — 4-level learning path, 14 tutorials
• Stable Facades API — first-path public API for notebooks, tutorials, and integrations
• Kuramoto Standalone Package Decision — the oscillatools package split, CEO/IP-approved 2026-07-04
• Sparse Kuramoto CPU Path — SciPy sparse force/Euler/RK4 route with 1M-node ring scaling evidence
• API Overview — stable facade route first, advanced module references second
• Research Gems — 33 analysis modules with theory and API
• Equations — every equation in the codebase
• Architecture — dependency graph + 20 subpackages
• Analysis API — advanced reference for 46 analysis modules
• Phase API — advanced reference for 29 evolution algorithms
• Application Benchmark Plugins — EEG, plasma, power-grid, and FEP datasets through QPU artefacts
• Classical Baselines — SciPy ODE, QuTiP Lindblad, and MPS TEBD provenance surfaces
• TN/MPS Baseline Design — CPU-first N=30-40 tensor-network baseline plan
• TN/MPS Crossover Stage-1 Gate — QWC-5.1 N=30-40 row schema and claim boundary
• Josephson K_nm Magnitude Study — N=14 rho=0.990 topology candidate plus N=20/30/40 measured-magnitude gates
• p_h1 Open-Claim Guard — public wording guard that keeps the 0.72 threshold open until reproduced
• Hardware Guide — IBM Quantum setup
• Notebooks — 99 tracked notebooks
• Bridges — cross-repo integrations
• Language Policy — Rust / Julia / Go / Mojo accel chain
• Pipeline Performance — every module's measured wall-time + multi-language benchmarks
• Issue Triage — label taxonomy, SLAs, routing
• Falsification — 8 named claims + falsifiers

Related Repositories

Repository	Description
sc-neurocore	Classical SCPN spiking neural network engine (v3.15.34)
scpn-fusion-core	Classical SCPN algorithms: Kuramoto solvers, coupling calibration, transport (v3.9.11)
scpn-phase-orchestrator	SCPN phase orchestration: regime detection, UPDE engine, Petri-net supervisor (v0.9.0)
scpn-control	SCPN control systems: plasma MPC, disruption mitigation (v0.21.0)

Citation

@software{scpn_quantum_control,
  title  = {scpn-quantum-control: Quantum-Native SCPN Phase Dynamics and Control},
  author = {Sotek, Miroslav},
  year   = {2026},
  url    = {https://github.com/anulum/scpn-quantum-control},
  doi    = {10.5281/zenodo.18821929}
}

License

AGPL-3.0-or-later — commercial license available.

Dual licensing is explicit: the public repository is AGPL-3.0-or-later, and proprietary integration requires a separate commercial licence grant. The generic Kuramoto-XY facade is documented as a possible future core-package boundary, but it is not a separate permissive package today. Until an explicit release changes SPDX headers and package metadata, all in-repository code remains under the AGPL/commercial terms above.

Use case	Route
Academic research, teaching, and individual experiments	Use the AGPL terms in LICENSE.
AGPL-compatible open-source redistribution	Use the AGPL terms and preserve notices/source obligations.
Closed-source product, internal proprietary tool, SaaS, consulting deliverable, or embedded deployment	Obtain a commercial licence before distribution or network service use.
Future lightweight core package	Not available yet; no permissive relicensing is implied by the facade docs.

Commercial Licensing

AGPL-3.0 requires derivative works and network-service modifications to provide corresponding source under the AGPL. If you need to integrate scpn-quantum-control into proprietary software without publishing your source code, use the commercial route:

1. Email protoscience@anulum.li with organisation name, product or service description, deployment model, expected users, and whether source redistribution is required.
2. Select the licence tier below or request an enterprise quote.
3. Execute the commercial licence grant before shipping the proprietary integration or offering it as a network service.

Tier	Price	Includes
Indie	CHF 49/month	Single developer, one product
Pro	CHF 199/month	Team up to 10, unlimited products
Perpetual	CHF 999 one-time	Permanent license, one major version
Enterprise	Custom	SLA, priority support, custom modules

Reference files: LICENSE, NOTICE.md, docs/core_package_boundary.md, and docs/licensing_faq.md.

Contact: protoscience@anulum.li | anulum.li

---

    
Developed by ANULUM / Fortis Studio