QF Solver 0.2.5a0 controlled known limitations and final scope¶
These are default release boundaries. They may be narrowed by evidence and may be expanded only through an Owner-reviewed scope/gate revision.
Final Owner scope revision — 2026-08-28¶
OWNER_SCOPE_REVISION = APPROVED and SCOPE_CHANGE = YES are recorded in
0_2_5_owner_scope_revision.md. The qualified release scope includes only the
bounded PASS gates G01, G02, G03, G05, G08, G09 and G11, plus the aggregate
governance closure G10/G12. The original G04 and G06 requirements remain
preserved for traceability but are excluded from the qualified 0.2.5a0 claim:
- G04 is
EXPERIMENTAL / NOT_QUALIFIED; its arc-length evidence and blockers remain visible and are deferred. - G06 is
CODE_COMPLETE / EXPERIMENTAL / QUALIFICATION_DEFERRED; finite- kinematic J2 and coupled external correlations remain unqualified. - G07 is
NOT_IN_RELEASE_SCOPE.
This is an explicit scope change, not a silent lowering of requirements:
CONTRACT_LOWERED = NO. No physical-validation, general finite-strain,
general arc-length or triple-coupling claim is made.
- J2 remains the existing isotropic-hardening small-strain model until an explicit finite-kinematic coupling model is approved and verified.
- No general finite-strain plasticity, anisotropic/kinematic hardening, damage, fracture, creep, viscoplasticity or cyclic-material qualification is claimed.
- TET4 and HEX8 are the MUST geometric-nonlinear element scope. TET10 and HEX20 remain SHOULD until their own evidence closes.
- Linear buckling is an idealized bifurcation analysis, not an imperfection-aware collapse prediction.
- Linear buckling is qualified only as a bounded first tangent-instability analysis: the Euler TET4 total-Lagrangian evidence uses four structured levels and the external Code_Aster probe uses a constrained five-TET4 block. TET10/HEX8/HEX20 external buckling, imperfection-sensitive collapse and post-buckling remain unqualified.
- Arc-length includes a reduced scalar shallow-arch branch-following check, but
remains bounded to the verified benchmark/problem class; no FEM snap-through,
post-buckling, arbitrary branch switching or general snap-back behavior is
implied. The global solver's signed target and
max_stepsturning window are opt-in research controls and do not change the default target-load path. - The finite-kinematic J2 arc-length path has bounded internal research evidence
on homogeneous TET4, TET10, HEX8 and HEX20 paths up to signed load factor
0.5, using an opt-in adaptive radius. It remains a monotone target-load continuation contract; it does not qualify FEM snap-through, snap-back, post-buckling, arbitrary branch switching or external response correlation. - Frictionless contact is the mandatory contact scope. Friction remains optional and cannot be advertised unless G07 is promoted and closed.
- The common contact path supports a bounded slave-node patch against a triangulated master surface, updated facet search, current-configuration normals and penalty enforcement. The patch expands to independent node-to-faceted-surface contributions; it is not a mortar or segment-to- segment formulation. G05 is closed only for this bounded contract: the finite-sliding, updated-normal and facet-transition paths are internally qualified, while Code_Aster provides bounded compatible normal-contact histories. This does not qualify general continuum surface-to-surface contact, unrestricted finite sliding or self-contact.
- Self-contact, cohesive, thermal and advanced mortar contact are excluded.
- Triple J2+geometry+contact coupling is SHOULD, not a release prerequisite.
- External numerical correlation is not physical validation.
- The regular two-cell Code_Aster J2 campaign is convention-matched and reports
PASS_EXTERNAL_CORRELATIONfor TET4/TET10/HEX8/HEX20 when the explicit QF TET10code_aster_5nonlinear quadrature is selected. The legacy four-point Hammer rule remains the default for existing models and linear paths, so this bounded evidence does not by itself close G01/G10. - The release is not an HPC expansion and makes no new million-DOF nonlinear claim. PETSc/SLEPc remain optional backends.
- Performance claims are limited to recorded hardware/software/mesh histories.
- Full Newton is the qualified global method. Modified Newton remains outside production scope unless separately gated.
- A capability whose qualification is open, deferred or excluded remains experimental/research regardless of whether code or tests for it exist.
Governance roadmap after 0.2.5¶
0.2.5: finish the current release without enlarging its functional scope. The final gate matrix marks G04 and G06NOT_IN_RELEASE_SCOPEunder the approved scope revision; their qualification records remain open/deferred. G06 isCODE_COMPLETE / EXPERIMENTAL, and no finite-strain J2 or coupled finite-kinematic claim is made.0.2.6: focus on maturity, V&V, robustness, benchmarks and scalability/performance. Corrections are allowed only when a defect is demonstrated; no major new physics is planned.0.2.7: target an approved finite-strain J2 formulation, coherent stress/strain measures, constitutive tangent and state-transaction evidence, independent V&V, Code_Aster correlation and G06 requalification. Friction remains a separate possible work package and is not committed here.
Coverage provenance note¶
The apparent change from approximately 88.37 % to 63.64 % was a
measurement-environment discrepancy, not a numerical regression. The valid
candidate run used the unit/integration command with --cov=solveur,
--cov-branch, the configured exclusions (not benchmark and not large and
not evidence), and an explicit PYTHONPATH=src; it selected 1550 tests,
measured 17867 source statements and reported 88.341 %. The invalid run
imported solveur from the neighbouring QF_solver_public checkout because
the source path was not explicit; it measured a different tree (565 files
and 58861 statements) and reported 63.64 %. Future release evidence must
record the resolved solveur.__file__ and the candidate source_sha.