ANYfeSolver 0.5.0a1__tar.gz

anyfesolver-0.5.0a1/CHANGELOG.md

@@ -0,0 +1,424 @@
+# Changelog
+
+## Unreleased
+
+## 0.5.0 - release candidate
+
+- Require ANYfileio 0.3.3 or later in the 0.3 line for SESTRA shell stress
+  recovery from relabelled descriptors and reported rotation frames.
+- Rename the solver distribution and import to `ANYfeSolver` and `anyfesolver`.
+  The separately installable results runtime is `ANYfeSolver-results` 2.2.0,
+  imported as `anyfesolver_results`. See [the migration guide](MIGRATION.md)
+  for consumer upgrades, supported historical archives and restart files.
+
+- Remove 22,552 historical development/qualification files from the checkout
+  after verified external archival; retain regression inputs, authority records,
+  source, tests and tools. See `docs/DEVELOPMENT_ARCHIVE.md` for recovery.
+- Reconcile the documentation entry points and local development status after
+  the October 27–28 integrations. Retain dated logs and qualification records
+  with their original scope; see `docs/STATUS.md`.
+- Add linear inertia relief and prepared static-batch integration for the
+  admitted free-free beam/shell domain; see `docs/INERTIA_RELIEF.md`.
+- Add neutral-input floating-body frequency response, fixed-step transient
+  wind-thrust feedback and physical pressure-region loading. These remain
+  bounded development routes; see `docs/FLOATING_RESPONSE.md`.
+- Add concentrated rigid mass with centre-of-gravity offset and inertia,
+  including linear mass/gravity and inertia-relief integration; see
+  `docs/concentrated_mass.md`.
+
+- Qualified Q4 plastic shells evaluate layered plasticity at the hybrid assumed
+  strain (`_q4_assumed_strain_plasticity`). The historical compatible-strain
+  route with the constant qualified correction had an indefinite tangent once
+  layers yielded and converged to spurious mechanisms on a clamped plastic plate
+  (+203 mm at n = 8, wrong sign at n = 16). The new route has an exact symmetric
+  tangent and completes at n = 4..32, within 2.2 % of the legacy element at
+  n >= 8 and moving toward CalculiX (-0.7 % at n = 32). Elastic materials without
+  a hardening curve are bit-identical in linear analyses.
+  Hill48/orthotropic plasticity (with or without a curve), generalized sections
+  with hardening, warped facets and active initial fields are refused with
+  `UnsupportedQ4PlasticRouteError` instead of the unstable form. Plastic
+  `layer_strain`/`layer_stress` now hold assumed-strain values. See
+  `docs/Q4_ASSUMED_STRAIN_PLASTICITY_20261023.md`.
+  Accepted exceptions (owner approval 2026-10-23):
+  - Before yield, plastic-capable elements change through the von-Karman cross
+    terms: 1e-10..2e-9 relative for small rotations, and up to about 1e-3 in the
+    centre deflection (about 5e-3 in-plane max |du|) on large-deflection thin
+    plates. There is no growth under refinement, and the values move slightly
+    toward the elastic CalculiX answer.
+  - Single elastic law: qualified Q4 elements without a hardening curve run the
+    same kernel (return map skipped), so von Karman and corotational nonlinear
+    elastic results of planar isotropic qualified Q4 change. Owner waived
+    bit-identity. C1 elastic n = 8/16: von Karman -9.2e-4 / -9.6e-5 (toward
+    CalculiX), corotational +6.7e-4 / +1.5e-4. Mixed curve/no-curve meshes now
+    equal the pure meshes to 1e-9. The compensated direct elastic route and the
+    batch-B elastic fast path are retired for these elements. Linear analyses and
+    non-qualified-Q4 elements are bit-identical.
+  Known limitations:
+  - Historical compatible-law plastic histories restored through restarts, seeds
+    or a mixed persistent store are not detected. Codex rates this a
+    high-severity restart correctness risk; a law-version tag is the proper fix.
+    R4's own unsealed restart states also reach the migration point, so a refusal
+    there was withdrawn. Open HIGH follow-up.
+  Qualification: TARGET 5c34dc2e4, 68/68 PASS with exceptions W1 (2.5 % n8 in-plane allowance), C4 (legacy
+  rule), A8 (inventory), H1 (historical-state limitation); A9 2.0x legacy / 1.25x base. Open follow-ups:
+  public stress recovery review (option A accepted), plastic-law version tag (HIGH), governed manifest
+  successor record, descriptor-shear gate follow-up for S3 impact. Merged to local main 2026-10-23.
+
+- Refresh the private Windows Python 3.12 beam-shell bundle from `6fb907a1c`,
+  including both completed-prefix S355 checkpoint workflows. Verify 77 installed
+  tests, two pre-yield and six fresh-process plastic full/prefix/resume CLIs;
+  check all 727 installed solver files and wheel/sdist contents against the
+  committed source. Preserve the earlier artifact under its original hash.
+  The private version remains `0.4.7+beam.shell.candidate.20261020.s3`; explicit
+  local wheel paths and bundle hashes distinguish the builds. Repository release
+  version, scientific gates and publication status remain unchanged.
+
+- Deliver the source S355 beam-shell completed-prefix checkpoint workflow for
+  direct and interpolated default coupling: 11/6/5 initial increments to
+  -11/60 m via -0.10 m, with unchanged material, iteration controls and tolerances.
+  Bind corotational Q4 checkpoint records at their constitutive reference
+  deformation in emission, validation and restart preflight. Retain existing
+  schema, integrity checks, failed-terminal refusal and `.converged` semantics.
+  Cancellation and rejected trials preserve caller inputs and recovery files.
+  The previously pinned installed artifact remains pre-yield only.
+  Twelve checkpoint and 113 shared/compatibility tests pass. The four remaining
+  default regressions also pass, using a retained completed comparison fixture
+  with fresh production prefixes/resumes and unchanged assertions. The earlier
+  timed-out selection remains preserved; no full-suite or scientific-gate claim.
+
+- Deliver a private Windows Python 3.12 beam-shell pre-yield candidate, version
+  `0.4.7+beam.shell.candidate.20261020.s3`, with wheel, sdist, pinned dependencies
+  and recorded installed verification (65 focused tests and both CLI modes).
+  That artifact omits serialized plastic checkpoints, which were blocked at its
+  source identity; the failed candidate remains preserved. The newer source
+  workflow above is not included in that unchanged artifact.
+  Director continuation and analytical accuracy remain open.
+
+- Add the standalone public example `examples/beam_shell_coupling.py` for the
+  registered pre-yield S355 beam-shell case with direct or interpolated
+  (default) rotating coupling. Its CLI prints a truthful JSON summary
+  (`status`, `stop_reason`, `requested_load_factor`, `achieved_load_factor`,
+  `probe_displacement`, `accepted_increments`, `constraint_postcheck`,
+  `completed`) and exits 0 only when the solve actually completed to the
+  requested load factor within 1e-8 with passing existing constraints; a
+  cancelled solve reports its accepted increments without claiming success.
+  New focused integration tests cover both couplings, an independent
+  true-elastic control, cancellation with empty and non-empty accepted
+  history, and the CLI JSON/exit-status contract. No solver default,
+  tolerance or material change.
+- Correct the beam-shell connection guide's stale completion claim: the
+  default rotating route has a demonstrated completed S355 displacement-control
+  run to the revised -11/60 m endpoint including the public initial-state
+  restart, while the original -0.20 m target, the director-integrated trace
+  (stopped at -1/6 m) and the analytical accuracy gate remain open. The guide
+  now also explains reading achieved displacements from
+  `result.displacements`, `steps[-1].control_value` under displacement control,
+  and that `result.converged` alone is not completion. No default, tolerance
+  or material change.
+- Speed up nonlinear static solves of models that combine qualified Q4 shells
+  with exact S3 V2D triangles. The solver's own cancellation token and the V2D
+  elements now use the constant-time trusted lifecycle guard inside the Newton
+  loop, and every Numba kernel is cached on disk by default (`cache=True` is
+  applied by `anysolver.jit_compiler`; pass `cache=False` to opt out).
+  Displacements, load factors and step histories are bit-identical to 0.4.7.
+  On the representative plate/cylinder case a warm solve drops from 22.5 s to
+  11.7 s and the first solve in a fresh process from 38.5 s to 12.0 s.
+- A nonlinear static solve now stops with an authority error when an S3 V2D
+  element it owns is modified (any ordinary attribute write or deletion) while
+  it runs. Writes before or after a solve behave as before.
+- A cancellation token is trusted only if it is an unmodified
+  `CancellationToken` both before and after its checkpoint code runs; every
+  other token keeps the complete lifecycle scan.
+- Report which guard path a nonlinear static solve used through the
+  `qualified_guard_trusted`, `qualified_guard_complete_scan`,
+  `qualified_guard_untrusted_token` and `qualified_v2d_trusted_unavailable`
+  events in `info["nonlinear_performance"]["solver"]["event_counts"]`.
+- Speed up the exact qualified lifecycle guard for Q4 and S3 shells: class-level
+  facts and one element-independent serialization module guard are read once per
+  validation call instead of once per element. One complete scan drops from about
+  70 to about 33 microseconds per shell; a warm nonlinear solve of a stiffened
+  synthetic model is about 20 % faster and the GUI-style plate/cylinder run about
+  8 %. Instance-level checks, the element-specific validators and S3 V2D elements
+  are unchanged. Semantics: a change to a base class made after the first element
+  of a class was checked is reported by the next call rather than by the remaining
+  elements of the same call (a change to the Q4 or S3 class itself is still
+  reported by them).
+- Keep qualified element equations, physical tolerances and defaults
+  unchanged.
+
+## 0.4.7 - 2026-09-21
+
+- Speed up representative static analysis and retained-load solves through
+  qualified preparation and reuse; the recorded complete-route medians improve
+  3.98x and 19.85x respectively on the declared static workload.
+- Reduce repeated spectral validation work while preserving cancellation and
+  mutation guards. Retained medium-panel buckling improves 11.52% in the
+  seven-pair gate, with identical modes and residuals.
+- Improve nonlinear static work accounting, accepted-force reaction recovery,
+  fixed dead-load projection, and trial-tangent promotion checks. The frozen
+  representative nonlinear gate records a 12.44% median improvement across
+  non-easy cases with passing physical comparisons.
+- Provide an opt-in Armijo residual line search for supported force-controlled
+  beam/shell analyses. Its separate convergence promotion gate was NO-GO: it
+  did not add difficult solves or reduce failed work, so no convergence benefit
+  is claimed; existing defaults remain unchanged.
+- Preserve runtime constraint and nonlinear failure details in the generated-
+  geometry facade when forwarding locally edited models.
+- Keep qualified element equations, physical tolerances, defaults, and
+  coordinated ecosystem dependency bounds unchanged.
+
+## 0.4.6 - 2026-09-16
+
+- Accept the independently published ANYmesher 0.5 line through the exact
+  dependency bounds `ANYmesher>=0.4,<0.6` and `ANYfileio>=0.3.2,<0.4`.
+- Bind the compatibility release to production ANYmaterial 0.2.0,
+  ANYgeometry 0.4.3, ANYmesher 0.5.0, and ANYfileio 0.3.2 artifacts.
+- Preserve all solver mechanics, B3-GE opt-in behavior, legacy B3 default,
+  and qualified Q4/S3 defaults unchanged.
+
+## 0.4.5 - 2026-09-16
+
+- Publish the bounded B3-GE opt-in release after correcting the isolated
+  installed-wheel gate to install the wheel's declared runtime dependencies.
+- Preserve the failed v0.4.4 GitHub publication attempt as process evidence;
+  no v0.4.4 artifacts reached PyPI.
+- Keep legacy B3 as the default and leave qualified Q4/S3 mechanics and
+  defaults unchanged.
+
+## 0.4.4 - 2026-09-16
+
+- Publish the accepted `B3-GE` model-owned geometrically exact beam workflow
+  as an explicit `b3-ge` opt-in. Legacy B3 remains the default for every
+  existing three-node beam route and historical record.
+- Bind the minor release to the accepted G7 production-opt-in evidence and a
+  bounded installed-wheel gate. Qualified Q4/S3 mechanics and defaults are
+  unchanged.
+- Remove unreferenced development notes from the active tree and exclude
+  research plans, raw evidence, tests, and build scripts from release source
+  archives. The complete pre-cleanup tree remains available at the archive tag
+  `archive/anysolver-pre-0.4.4-development-20260916`.
+
+## 0.4.3 - 2026-09-09
+
+- Add the explicitly selected `ge-beam3` straight mixed beam and the reviewed
+  `ge-beam3-native` model-owned straight/regular-curved beam workflows, with
+  declared generalized and physical-fibre sections, authenticated history,
+  recovery, restart, and admitted reference/current-rest spectral analysis.
+- Add the reviewed objective eccentric generalized-beam connection to one
+  owned elastic qualified Q4 or S3 V2D through the native workflow API.
+- Preserve the full development/qualification history and its scoped
+  independent acceptance. This is not qualification of generic mixed FEModel
+  routing, plastic/fibre-shell coupling, finite-velocity rotational dynamics,
+  arbitrary sections, or stable postbuckling paths from rest.
+- Keep existing B2/B3, qualified Q4/S3 mechanics and all defaults unchanged.
+  See `docs/GE_BEAM3_NATIVE_WORKFLOWS.md` for the supported API and limitations.
+- Separate the explicit current GE runtime CI inventory from preserved extended
+  experiments. Historical evidence and the unpublished qualification wheel
+  remain immutable; the versioned release wheel is verified separately.
+
+## 0.4.2 - 2026-09-03
+
+- Canonicalize solver-owned compact qualified-Q4 plastic state once at the
+  accepted Newton finalization boundary. Public sealing remains fail-closed,
+  while restart and subsequent current-tangent replay become byte-exact.
+- Preserve qualified Q4/S3 mechanics, coefficients, recovery, formulation
+  identities, and defaults.
+
+## 0.4.1 - 2026-09-03
+
+- Correct the coordinated ecosystem dependency bands to require ANYmaterial
+  0.2.x, ANYmesher 0.4.x, and ANYfileio 0.3.1 or newer. Version 0.4.0 carried the
+  pre-activation bands and cannot be resolved together with ANYfem 0.4.0.
+- Add a release gate that resolves the built wheel with the exact coordinated
+  sibling releases before trusted publication.
+- Admit only a fixed binary64 last-place envelope when scalar final-state
+  replay validates compact qualified-Q4 plastic states produced by vectorized
+  assembly. This fixes sealing after valid plastic increments while preserving
+  exact origin/configuration hashes, exact rich-state replay, and strict
+  rejection of material history differences. Qualified Q4/S3 mechanics and
+  formulation defaults are unchanged.
+
+## 0.4.0 - 2026-09-03
+
+- Relicense ANYsolver source releases from GPL-3.0-or-later to MPL-2.0.
+  Earlier published versions retain the licence terms that applied to them.
+- Publish a machine-readable direct-dependency licence inventory and preserve
+  third-party notices in source and wheel distributions. The extracted ANY
+  dependencies remain separately licensed and are not bundled or relicensed
+  by ANYsolver.
+- Add release-time checks for repository metadata, notices, dependency
+  inventory, and licence files in built distributions.
+- Document the qualified-S3 admission boundary for automatically imprinted
+  four-way shell intersections. These junctions continue to fail closed rather
+  than bypassing qualification or silently selecting legacy S3.
+- Activate the V6W-qualified native-parity S3 V2D companion as the default
+  three-node shell for new/current-policy models. Historical records without
+  formulation authority remain explicit `legacy-s3`, direct `ShellElement`
+  construction remains legacy, and qualified S3 creation fails closed without
+  an authoritative physical owner normal. Qualified Q4 mechanics are unchanged.
+- Mark the coordinated breaking activation boundary as ANYsolver 0.4.0 and
+  admit the compatible ANYmesher 0.3 line while retaining the strict 0.4 cap.
+
+- Add the explicitly selected `e4-pl-s3`/`qualified-s3` linear MITC3+
+  companion candidate with frozen source identity, two bubble rotations,
+  basis-invariant three-mode PL completion, stable serialization identity,
+  authoritative-normal admission, formulation-native local recovery, and
+  analytic nodal-plus-bubble mass with the frozen Guyan reduction. The modal
+  solver excludes exact zero-inertia drill coordinates without inventing mass,
+  using coordinate-invariant static condensation for bounded systems and a
+  fail-closed swapped descriptor pencil for larger systems. The candidate now
+  also supplies formulation-native total-Lagrangian layered response,
+  stateless generalized-section geometry with analytically integrated initial
+  stress resultants and generalized eigenstrains, contact virtual work,
+  committed recovery without fabricated layer data, physical-director
+  reversal, current-state modal/buckling operators, and solver-owned
+  static/arc-length checkpoints. The original candidate remained opt-in until
+  the independent V6W qualification and this successor activation boundary.
+- Keep qualified E4-PL as the production Q4 default and begin the explicit
+  legacy-Q4 retirement window. Legacy Q4 selectors now emit a dedicated
+  deprecation warning, diagnostics expose the selected formulation and
+  rollback lifecycle, and migration guidance freezes availability through
+  0.4.x with removal no earlier than 0.5.0 after two clean release gates.
+  TRI3, TRI6, Q8, and Q8R remain unchanged.
+
+## 0.3.0 - 2026-08-23
+
+- Preserve the corrected published-2025 S4 reference, nullspace proof, geometry
+  handoff, and qualification scaffold as dormant release evidence. The legacy
+  S4 remains the production default and improved-S4 activation remains
+  unavailable under the recorded restrictions; no gauge constraint, invented
+  stiffness, dispatch route, or activity-policy change is introduced.
+- Widen the declared ANYfileio consumer range to `>=0.1,<0.3`, add pinned
+  source compatibility cells for the 0.1 and 0.2 public contracts, and align
+  the publication dependency gate. Installed-wheel, index-resolver, and
+  publication qualification remain separate release gates.
+- Widen the declared ANYmesher consumer range to `>=0.1,<0.3`, add CI coverage
+  for pinned 0.1.0 and 0.2.1 endpoints, preserve the public neutral panel, mesh,
+  quality, and section contract, retain the strict 0.3 cap, and align CI and
+  publication dependency gates with the declared metadata.
+- Complete the Sol Ultra `performance_2` campaign with qualified static/arc
+  persistent nonlinear state transactions, compiled canonical-curve Hill-48 batches,
+  orthotropic/generalized S4 kernels, revision-cached corotational block
+  transforms, large-selection isotropic S4 recovery, and direct reduced
+  elastic nonlinear-impact assembly. Unsupported formulations, curves, state
+  layouts, impact plastic/fiber history, small recovery selections, affine
+  constraints, and damage/state scopes retain explicit scalar or
+  full-coordinate fallbacks with reason diagnostics.
+- Add conservative opt-in impact tangent/factorization reuse, compact lazy
+  contact records, and revision-guarded incremental damage K/M updates. A zero
+  reuse budget remains the full-Newton oracle; contact, convergence, plastic,
+  damage, deletion, time-step, and line-search changes force a refresh. Damage
+  plan setup is gated by projected future events and retained-memory headroom;
+  invalidation returns to the exact rebuild and never scales point masses.
+- Add the optional caller-owned `AnalysisSession` for bounded K/M/constraint,
+  reduced-matrix, output-row, and factorization reuse across repeated linear,
+  modal, buckling, transient, and capacity work. Linear transient loads are
+  preprojected once and selected history output avoids full-vector
+  reconstruction when recovery does not require it. Explicit solver,
+  assembly, and recovery thread scopes remain independently controlled and are
+  restored after each call.
+- Release 0.2.0 extraction boundary: material behavior now comes from
+  `ANYmaterial`, neutral geometry/meshing from `ANYmesher`, and SESAM/CalculiX
+  parsing and writing from `ANYfileio`. ANYsolver retains solver-specific
+  FEModel adapters, MPC elements, boundary interpretation, external execution,
+  comparisons, and the 0.2.x legacy import facades. Deterministic FE baselines
+  remain unchanged.
+- Add dependency-free generalized section contracts: pre-integrated shell
+  `A/B/D/As` with membrane-bending coupling and optional areal inertia, plus a
+  coupled beam 6x6 stiffness and optional mass-per-length law. Generated
+  geometry accepts inline or named definitions; recovery exposes exact
+  generalized strains/resultants, nonlinear shell acceleration falls back
+  deterministically, and unsupported CalculiX/fiber/plastic mappings fail
+  closed.
+- Add a dependency-free structural material contract and homogeneous
+  orthotropic engineering materials for shells and beams. This includes
+  projected shell material directions/angles, explicit orthotropic beam
+  torsional rigidity, generated-geometry and CalculiX shell interchange, and
+  deterministic fallback from isotropic accelerated kernels to general
+  assembly.
+- Add material-axis Hill-48 shell plasticity with directional-strength
+  hardening, consistent tangent plus numerical oracle/fallback, stored physical
+  stress recovery, labelled Hill equivalent stress/utilization, orthotropic
+  beam-fiber longitudinal yielding, and damage/contact integration. General
+  anisotropy, laminates, ply failure, and shear/torsion plastic interaction
+  remain outside scope.
+- Add an exact constant-stress analytical/CalculiX orthotropic S4 reference
+  deck, perfect-plastic Hill behavior when no hardening curve is supplied, and
+  current-strength Hill utilization with physical-stress recovery provenance.
+
+## 0.1.3 - 2026-07-28
+
+- Complete the generated-geometry runtime bridge: apply configured shear force
+  and torsional moment, pass current-area follower pressure only to supported
+  nonlinear static/arc-length paths, retain the selected corotational
+  kinematics in arc length, and fail closed for incompatible follower paths.
+  Runtime prestress recovery now consumes matching committed nonlinear
+  histories and exposes Gauss-point membrane/bending resultants plus recovery
+  provenance.
+- Correct nonlinear display recovery so plastic-history von Mises values use
+  only the return-mapped shell in-plane or beam-fiber components, while purely
+  elastic nonlinear states retain their full equivalent stress. Preserve
+  combinations with elastically reconstructed transverse shear/torsion as a
+  separately labelled model-scope diagnostic.
+- Add the public `resolve_runtime_analysis()` contract so application GUIs can
+  reflect normalized nonlinear/material/control/kinematics selections without
+  importing private runtime helpers.
+
+## 0.1.2 - 2026-07-28
+
+- Add opt-in executable CalculiX reference validation with isolated case runs,
+  stale-output protection, timeouts, solver version/hash provenance, ASCII
+  FRD/DAT parsing, tolerance-controlled comparisons, and preserved evidence
+  semantics. Deck-only generation remains explicitly `not_executed`; stale
+  non-executed legacy reports are regenerated deterministically, while invalid
+  executed evidence is preserved and rejected for diagnosis.
+- Add current-area follower pressure and its exact, generally nonsymmetric
+  external-load tangent to nonlinear static and arc-length solves. Extend
+  shell initial-stress stiffness to the Mindlin translation/director field
+  using membrane, bending, and second stress moments, and add the qualified
+  thin-ring pressure-buckling gate.
+- Add an opt-in consistent corotational tangent through the full
+  pull-back/frame/rotate-forward chain rule. `auto` retains the lower-cost
+  rotated tangent for ordinary loads and selects the consistent tangent for
+  follower pressure.
+- Unify nonlinear stress recovery around committed shell-layer and beam-fiber
+  histories, matching solution displacements, objective corotational frames,
+  component-level provenance, and explicit elastic fallbacks. Add guarded
+  full-integration Q4/Q8 patch recovery that separates discontinuities and
+  labels its optional surface-stress L2 indicator as non-energy-norm.
+- Add shell membrane/bending stress and prestrain fields plus beam-fiber
+  initial fields, with zero-load equilibration, immutable field provenance,
+  exact load-stage boundary commits, and persistent plastic history through
+  multi-stage displacement control.
+- Replace the default finite-difference plane-stress algorithmic tangent with
+  the analytical consistent derivative of the discrete return map. Retain the
+  representable-step numerical tangent as a qualification oracle and guarded
+  invalid-row fallback, add a safeguarded local consistency solve with
+  fail-closed residual checks, and add constitutive and full-Newton
+  parity/performance evidence.
+## 0.1.1 - 2026-07-25
+
+- Correct the layered-shell membrane/bending coupling tangent and keep the
+  accelerated plastic shell path consistent with the scalar return mapping.
+- Fail closed for curved B3 members, experimental Q8R qualification, unsupported
+  DNV-RP-C208 thickness rows, and path-dependent staged displacement control.
+- Include point masses in reported mass properties, make buckling constraints
+  call-order independent, and correct CalculiX gravity export.
+- Remove quadratic model-construction and beam/shell coupling lookup scaling,
+  safely reset precomputed incoming elements, cache recovery operators and
+  topology signatures, defer optional acceleration imports, and use lower
+  PyPardiso thresholds only for compatible retained patterns.
+- Preserve straight-sided B3 midpoint geometry through adaptive cylinder
+  refinement and reconstruct complete elastic layer states after accelerated
+  nonlinear solves.
+- Add explicit runtime exports, generic generated-geometry API aliases, and
+  package-root arc-length exports.
+- Adopt the pure-solver test files from ANYstructure (triangular shell
+  backend, local patch transition, geometry panel).
+
+## 0.1.0 - 2026-07-22
+
+- Transfer the qualified beam/shell solver from ANYintelligent into its own package.
+- Add the headless runtime meshing and analysis facade formerly embedded in ANYstructure.
+- Publish the Python import namespace as `anysolver` without a legacy `fe_solver` shim.
+- Preserve SESAM interchange, verification, qualification, benchmark, and baseline workflows.

anyfesolver-0.5.0a1/COPYRIGHT

@@ -0,0 +1 @@
+Copyright (c) Audun Nyhus