SPECTRAX-GK v1.7.0

SPECTRAX-GK 1.7.0

This release completes the current architecture and validation tranche while keeping research claims explicitly scoped to artifact-backed evidence.

Highlights

• Refactored the package into physics-first domains with zero unreviewed modules above 1,000 lines, compact public facades, and enforced architecture manifests.
• Added reviewed SOLVAX 0.7.x structured-solver ownership for batched tridiagonal solves and bounded-memory Jacobian workflows.
• Strengthened differentiable workflows with JVP/VJP and finite-difference gates for fixed-step, adaptive, IMEX, VMEC-JAX, Boozer, quasilinear, UQ, and optimization paths.
• Closed the equal-species diagonal finite-wavelength Coulomb, original-Sugama, and improved-Sugama Python research boundary through the converged P24/J10 Figures 12–14 protocol, including equation, dissipation, convergence, zonal-response, velocity-section, and AD evidence.
• Added scoped production CPU/GPU parallelization for independent ky scans and quasilinear/UQ ensembles with serial identity gates. Whole-state nonlinear sharding remains diagnostic and is not advertised as a production speedup.
• Expanded nonlinear transport, stellarator optimization, quasilinear model-selection, comparison, restart, and resolved-diagnostic gates. Universal absolute quasilinear flux and broad production nonlinear turbulent-flux optimization remain explicitly unpromoted.
• Improved executable progress reporting, default runs, saved-input provenance, and spectraxgk --plot / spectrax-gk --plot publication-quality plotting.
• Refreshed the benchmark, runtime/memory, collision, differentiability, optimization, and validation documentation while keeping the tracked repository below its 50 MB policy.

Validation and release boundaries

• Package-wide CI coverage remains at or above 95%; coverage supplements rather than replaces physics and convergence gates.
• Required-core linear/nonlinear release cases, package builds, strict documentation, repository hygiene, and installed-wheel smoke tests are release gates.
• Arbitrary finite-wavelength multispecies Landau/Sugama input-file models, universal absolute quasilinear flux prediction, broad W7-X TEM/zonal claims, and production nonlinear multi-GPU domain-decomposition speedup remain future research scope.

Installation

pip install --upgrade spectraxgk

Run spectraxgk for the default example or spectraxgk --help for input and plotting options.

SPECTRAX-GK v1.6.10

SPECTRAX-GK v1.6.10

This release finalizes the nonlinear turbulent-flux optimization tranche with explicit claim boundaries.

Nonlinear optimization evidence

• Added tracked broad-matrix negative-evidence artifacts for the max-mode-5 QA nonlinear transport campaign.
• Documented the strict 18-point matrix outcome across 3 surfaces, 2 field-line labels, and 3 k_y values.
• Accepted QA/ESS optimized candidate: broad matrix failed with 9/18 samples passing.
• Projected transport-weight 1e-3: stopped early after the all-sample gate became impossible to satisfy (1/18 passing samples, mean reduction below 2%).
• Projected transport-weight 5e-4: stopped after the first completed sample increased heat flux by about 2.99%.
• Strict t=1500 growth/QL/nonlinear-window rows remain negative-transfer evidence and are not promoted.

Claim boundary

SPECTRAX-GK keeps scoped single-point matched-audit optimization evidence and the VMEC-JAX-style optimization examples, but this release does not claim broad multi-surface nonlinear turbulent-flux stellarator optimization. That claim remains deferred until a future candidate family passes the full matrix gate.

Examples and docs

• Updated optimization examples and documentation for editable VMEC files, objectives, algorithms, and matrix-audit workflows.
• Added fail-closed instructions for the nonlinear transport matrix portfolio and release finalizer.
• Added negative-evidence figures and JSON ledgers to the documentation.

Release checks

• Main CI passed for 7817628e.
• Release workflow passed and published spectraxgk 1.6.10 to PyPI.
• Local release readiness, repository-size, artifact-manifest, package-architecture, performance, parallelization, Sphinx, package build, and twine metadata checks passed.

SPECTRAX-GK v1.6.9

Full Changelog: v1.6.8...v1.6.9

SPECTRAX-GK v1.6.8

SPECTRAX-GK v1.6.8

Maintenance release focused on code simplification, release hygiene, and benchmark-driver testability after the v1.6.7 differentiable architecture checkpoint.

Highlights

• Split the remaining long benchmark driver wrappers into stable public facades plus typed private request-routing stages.
• Reduced the audited source hotspot count at or above 100 lines to zero.
• Preserved public benchmark APIs and output schemas for ETG, KBM, TEM, kinetic-electron, and Cyclone validation paths.
• Refreshed the tracked release-readiness artifact for 1.6.8.
• Published wheel and source distribution to PyPI through the release workflow.

Validation

• GitHub release workflow passed: version check, release-readiness gates, package build, Twine metadata check, and PyPI publish.
• Local release checks passed before tagging: Ruff, MyPy, Sphinx docs build, package build, Twine check, release-version check, release-readiness check, and targeted CLI/release tests.
• Previous full CI for the source refactor head passed; CI for the metadata-only version-bump commit is running separately on main.

Scope

This release does not promote new physics claims. Universal absolute quasilinear flux prediction, W7-X zonal long-window recurrence/damping, TEM/kinetic-electron W7-X extension, and broad nonlinear domain-decomposition speedup remain explicitly scoped or deferred in the validation ledgers.

SPECTRAX-GK v1.6.7

SPECTRAX-GK v1.6.7

This patch release ships the differentiable architecture/refactor checkpoint and keeps the current validation, performance, and science claims explicitly scoped.

Highlights

• Reorganized internals behind domain packages and stable public facades for geometry, solvers, diagnostics, artifacts, validation, objectives, runtime workflows, and parallel utilities.
• Kept the executable/Python workflows release-ready while preserving documented claim boundaries for model-development and performance lanes.
• Added release-closure plan evidence for the remaining large validation/reporting refactor hotspots and the post-release cleanup path.
• Updated package metadata and release-readiness artifacts to 1.6.7.

Validation

• Local release readiness, repository-size, architecture, release-artifact, technical-release, package-build, Twine metadata, release-scope tests, and Sphinx docs build passed before tagging.
• CI is required to be green on the exact release commit before publication.

Claim Boundaries

• Quasilinear outputs remain scoped to diagnostics, model development, and screening guardrails; no universal absolute-flux predictor is promoted.
• Nonlinear domain decomposition remains diagnostic/identity-gated until a production end-to-end CPU/GPU speedup gate passes.
• Broad nonlinear turbulent-flux stellarator optimization remains scoped until long post-transient matched baseline-vs-optimized windows pass across broader surfaces, field lines, and geometries.
• W7-X zonal recurrence/damping and W7-X TEM/multi-flux-tube extensions remain post-release science lanes.

SPECTRAX-GK v1.6.6

SPECTRAX-GK v1.6.6

This patch release finalizes the release-scoped performance artifact refresh and tightens claim boundaries for parallelization and nonlinear domain decomposition.

Highlights

• Regenerated the shipped runtime/memory benchmark panel from the tracked runtime_memory_summary_ship_refresh.json sidecar.
• Added a concise README performance table for representative shipped CPU/GPU/reference rows, including W7-X and HSX nonlinear cases.
• Rebuilt parallelization and decomposition status panels from tracked artifacts.
• Clarified that production parallelization is closed for independent k_y scans and quasilinear/UQ ensembles.
• Kept nonlinear device-z/domain decomposition fail-closed as diagnostic-only: identity passes, but the two-GPU speedup gate is not met and observable diagnostics need fused device-side accumulation before promotion.
• Updated release-readiness metadata and version to 1.6.6.

Validation

Local release checks passed before tagging:

• focused nonlinear/parallel artifact pytest shard
• Sphinx docs build
• release-readiness checker
• release-version checker with PyPI uniqueness for v1.6.6
• parallel-scaling artifact checker
• performance-optimization manifest checker
• local package build for sdist and wheel

Deferred

Production nonlinear domain-decomposition speedup remains deferred until a full solver route fuses scalar diagnostics into the RHS/update path and clears matched CPU/GPU serial-vs-decomposed transport-window speedup gates.

SPECTRAX-GK v1.6.5

SPECTRAX-GK v1.6.5

Release focused on tightening the quasilinear validation guardrails and shipping the current manuscript/release evidence cleanly.

Changes

• Updated quasilinear diagnostic artifacts for the current 11-case validation ledger.
• Locked the current fail-closed absolute-flux result in tests and documentation: the best regularized spectral-envelope candidate remains above the transport-error gate, so no runtime/TOML absolute-flux predictor is promoted.
• Regenerated and compressed the quasilinear residual-anatomy and regularization preview figures.
• Fixed wide package coverage shards that still expected the obsolete ten-case quasilinear snapshot.
• Kept release-scope documentation explicit about deferred W7-X zonal/TEM lanes and scoped nonlinear optimization claims.

Validation

• GitHub CI passed on main for commit 5e845f1.
• Wide package coverage matrix passed.
• Release readiness, repository-size, artifact-manifest, docs build, package build, and twine check passed before tagging.
• PyPI publish succeeded for spectraxgk==1.6.5.

Scope

• Production-closed claims remain limited to the validated release surface: independent-work parallelization, scoped nonlinear optimization evidence, and guarded quasilinear model-development diagnostics.
• Universal absolute quasilinear flux prediction, broad nonlinear domain-decomposition speedup, W7-X zonal long-window closure, and W7-X TEM/multi-flux extension remain outside this release scope.

SPECTRAX-GK v1.6.4

SPECTRAX-GK v1.6.4

Release-scoped tightening for validation, release guardrails, and publication artifacts.

Highlights

• Refreshed the RBC(1,1) transport landscape panel with 16 strict long-window nonlinear overlay points: the negative-side scan plus the zero-offset baseline now use replicated post-transient ensembles.
• Added tracked runtime/memory provenance sidecars for the shipped performance panel.
• Hardened the release workflow so PyPI publication is gated by release-readiness, artifact, repository-size, performance, parallelization, and quasilinear-promotion checks.
• Tightened README/docs language around parallelization: production claims cover independent k_y scans, quasilinear batches, and UQ ensembles; nonlinear domain decomposition remains diagnostic until production numerical-identity gates are complete.
• Clarified quasilinear and nonlinear optimization scope: current figures support validated diagnostic/model-development claims, with absolute-flux promotion and broader long-window nonlinear optimization evidence continuing under the manuscript validation plan.

Validation Notes

• The RBC(1,1) positive-side long-window nonlinear overlays remain explicitly scoped as pending rather than silently inferred from reduced or startup-window estimators.
• Runtime and memory summaries now have versioned CSV/JSON sidecars under docs/_static/.
• Release readiness and scope tests cover the new guardrails.

SPECTRAX-GK v1.6.3

SPECTRAX-GK v1.6.3

This release tightens the release-readiness and validation guardrails around the 1.6.x optimization, quasilinear, parallelization, and documentation lanes.

Highlights

• Hardened release readiness checks so optimization prelaunch/admission gates require explicit tracked rows, artifact paths, blocker semantics, and minimum sample counts.
• Hardened nonlinear parallelization artifact checks so production-speedup gate rows cannot drift from their source profiler/scaling artifacts.
• Kept nonlinear domain decomposition scoped as diagnostic/profiler evidence unless fresh source artifacts support a production speedup claim.
• Clarified documentation around scoped quasilinear model-development results, selected QA optimized-equilibrium nonlinear audits, and deferred W7-X zonal/TEM lanes.
• Refreshed release-readiness metadata for version 1.6.3.
• Verified the lightweight package artifacts: wheel is about 500 KB and sdist is about 1.2 MB.

Validation

• Current-head CI for commit 084be5a completed successfully: 59 successful jobs, 1 expected skip.
• Local release checks passed: release readiness, version/tag/PyPI checks, repository-size manifest, Sphinx docs build, package build, and twine check.
• PyPI publish workflow completed successfully and uploaded both wheel and sdist with digital attestations.

Install

pip install spectraxgk==1.6.3

PyPI: https://pypi.org/project/spectraxgk/1.6.3/

SPECTRAX-GK v1.6.2

SPECTRAX-GK v1.6.2

This patch release makes the PyPI artifacts lightweight while keeping the repository-hosted documentation/validation atlas intact.

Highlights

• Excludes docs/_static from source distributions, so pip install spectraxgk no longer downloads the docs figure/validation atlas.
• Keeps docs images in Git for README and documentation rendering.
• Moves the runtime and memory comparison panel near the top of the README, immediately after the highlights and before the current claim scope.
• Bumps package metadata and runtime __version__ to 1.6.2.

Package Size

• Wheel: about 0.41 MiB.
• Source distribution: about 0.99 MiB.
• Both artifacts contain 0 docs/_static entries and 0 NetCDF/HDF5/NPZ payloads.

Validation

• Local checks passed: docs build, repository size manifest, release artifact manifest, release-version check, package build, twine check, and focused release/CLI tests.
• PyPI now reports spectraxgk 1.6.2 as the latest version.