0.2.7 Owner Decision Log

This log preserves the foundation decisions and records the Terra-authorized WP07 technical implementation separately from public qualification.

Decision ID	Topic	Proposed state	Decision	Evidence/SHA	Date
027-OD-001	WEDGE6 formulation and scope	PROPOSED_OWNER_REVIEW	pending	WP05/WP06/WP07	-
027-OD-002	mesh-quality policies	PROPOSED_OWNER_REVIEW	pending	WP06	-
027-OD-003	external C3D6/PENTA6 comparability	PROPOSED_OWNER_REVIEW	pending	WP05	-
027-OD-004	J2 gap-closure policies	OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE	KEEP_QUALIFIED_BOUNDED_WITH_LIMITATIONS	WP20, 94461602dfd1782be57c20e1801a0d5d8e262ef1	2026-09-01
027-OD-005	1M-DOF verdict and public boundary	PROPOSED_OWNER_REVIEW	pending	WP12	-
027-OD-006	stretch/research selection	PROPOSED_OWNER_REVIEW	pending	WP13	-
027-OD-007	final release scope	PROPOSED_OWNER_REVIEW	pending	WP14	-
027-OD-008	WEDGE6 elemental kernel	TERRA_GO	PASS_TECHNICAL_EXPERIMENTAL_ONLY	T1-R4, WP07 evidence	2026-08-31
027-OD-009	WEDGE6 static vertical slice	WP08_REVIEWED	PASS_TECHNICAL_EXPERIMENTAL_ONLY	WP08 state/evidence, 8040909d6d65f740e1daf858ce572d250a87b39a	2026-08-31
027-OD-010	residual J2 and external V&V closure	OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE	PASS_WITH_LIMITATIONS	qualification/0_2_7/wp20_state.json, 94461602dfd1782be57c20e1801a0d5d8e262ef1	2026-09-01
027-OD-011	WP21 architecture/API/registry and release-truth cleanup	OWNER_REVIEWED_BOUNDED	PASS_WITH_LIMITATIONS	qualification/0_2_7/wp21_state.json, qualification/0_2_7/wp21_final_release_truth.json, 0f565dc9669763751a75f13b02004bde18af571c	2026-09-01

WP01 is a release-engineering foundation control rather than a numerical Owner qualification decision. Its status is PASS because the SHA roles, actual 0.2.6 publication state, artifact classes and tag/version guard are recorded in the WP01 machine-readable records. This does not approve any 0.2.7 capability.

WP02 is a registry-engineering foundation control rather than a new capability qualification decision. Its status is PASS because the v2 source of truth preserves all 33 public legacy identifiers, exposes 44 combination records, keeps historical statuses out of the active vocabulary, and passes deterministic schema, migration and generated-view checks. Inherited QUALIFIED_BOUNDED states remain bounded 0.2.6 scope; this work package introduces no new evidence or promotion.

WP03 is a compatibility-engineering foundation control rather than a new capability qualification decision. Its descriptors describe technical routing for existing element families, while the v2 registry remains the source of maturity. The preflight is fail-closed and reports supported, experimental, non-qualified and unsupported routes distinctly. Targeted descriptor, workflow and analysis tests pass; no numerical source or Owner maturity decision was changed.

WP05 is an external-oracle engineering control rather than a WEDGE6 qualification decision. The pinned Docker runs validated one minimal CalculiX C3D6 deck and one Code_Aster PENTA6 deck locally. Both oracle routes are AVAILABLE_LOCAL_ONLY; no QF WEDGE6 implementation, comparison verdict or public capability was created. The eight future benchmarks remain PLANNED, and the comparability tolerance remains OWNER_REVIEW_REQUIRED.

The WP05 preflight is PASS with the local-only limitation recorded in the machine-readable contract and review pack. This does not authorize WP07 or change any existing 0.2.6 decision.

WP04 is an additive V&V-engineering control rather than a capability qualification decision. Its declarative case/oracle schema, runner verdicts, canonical evidence serialization and replay mismatch checks pass on three representative fixtures. Historical V&V runners remain supported; no numerical source or maturity decision was changed.

WP06 is an additive mesh-quality control rather than a numerical capability qualification. The common contract covers TET4, TET10, HEX8 and HEX20 with deterministic VALID, VALID_WITH_WARNING and INVALID classifications, and invalid geometry is surfaced by preflight as MESH_GEOMETRY_INVALID. No universal aspect-ratio or conditioning threshold was approved; inherited family diagnostics retain their legacy provenance. Rigid-transform, dimensionless-scale, inversion, duplicate-node and serialization checks pass. The WEDGE6 quality contract is prepared only as an inactive future schema and WEDGE6_IMPLEMENTED remains NO. This status records engineering evidence and does not promote any capability.

An entry may be changed to an approved state only with a decision owner, exact evidence SHA, scope, limitations and date. No decision here reopens G07 or changes any 0.2.6 maturity classification.

T1-R is a remediation checkpoint, not an Owner qualification decision. It records the inactive WEDGE6 formulation, quadrature candidate and richer verification reference, asymmetric node/face fixture, prism Jacobian validity controls, stiffness/mass V&V obligations and external primary observables. Its targeted tests pass and no WEDGE6 capability or kernel is exposed. The checkpoint is ready for Terra/Owner re-review; WP07 remains NOT_STARTED and kernel authorization remains OWNER_REVIEW_REQUIRED.

T1-R3 closes the two review findings that remained open. The inactive quality contract now specifies a scale-aware certified minimum: at fixed triangular coordinates det(J) is quadratic in t, so endpoints and every interior stationary point are checked at each triangular vertex. Quadrature, face and interior samples are diagnostics only. The Terra adversarial prism is rejected by this certificate, with no automatic repair and no WEDGE6 kernel exposure.

The WEDGE6 external contract has one unique primary-observable declaration, strict duplicate-key JSON loading, an affine same-mesh 1e-6 candidate marked OWNER_REVIEW_REQUIRED, and mandatory case-specific Owner approval for non-affine, distorted or refinement cases. Post-observation tolerance retuning is forbidden. T1-R3 is ready for Terra re-review; WP07 remains NOT_STARTED and kernel authorization remains OWNER_REVIEW_REQUIRED.

WP07 technical implementation decision

Terra High's T1-R4 review authorized implementation after all six original pre-WP07 blockers were closed. WP07 is PASS for the six-node, 18-DOF small-strain homogeneous-isotropic elastic elemental kernel and its targeted V&V. Evidence covers shape identities, affine constant strain, exact Jacobian certificate, stiffness symmetry, rank 12 with six rigid-body modes, prescribed load-state tests, distorted geometry, energy and production/reference quadrature replay.

This is not a public qualification. The registry maturity for COMB-WEDGE6-linear_static is EXPERIMENTAL, and preflight returns EXPERIMENTAL_ROUTE. Gmsh import, face loads, user-level assembly and reactions, modal/dynamic routes, nonlinear materials, contact and external correlation were deferred to WP08 and later gates. Existing TET/HEX formulations and their evidence are unchanged.

WP08 static vertical-slice decision

WP08 is PASS for the bounded six-node Gmsh Prism 6 static workflow. The 15-case controlled evidence contains 14 PASS results and one EXPECTED_FAILURE_PASS for fail-closed rejection of inverted geometry, with no unexpected failures. It covers canonical TRI3/QUAD4 face mapping, nodal and distributed loads, sparse static assembly, reactions, force and moment equilibrium, constraints and displacement/strain/stress/energy recovery.

This decision does not promote WEDGE6: COMB-WEDGE6-linear_static remains EXPERIMENTAL and public qualification is DEFERRED. Modal/Newmark/harmonic, nonlinear, J2, TL, contact, robustness and external-correlation claims remain outside this gate and require their own evidence.

WP09 robustness and external V&V decision

WP09 is PASS_WITH_LIMITATIONS for the declared WEDGE6 static robustness campaign. The controlled evidence contains 22 cases: 18 internal PASS results and four EXPECTED_FAILURE_PASS results for inverted geometry, wrong node ordering, malformed Gmsh input and singular boundary conditions. Aspect, skew, near-degenerate, rigid-transform, scale-invariance, multi-element, refinement and deterministic replay checks are recorded without changing the kernel or existing element numerics.

The external results remain explicitly non-qualifying. CalculiX C3D6 completed the affine deck locally, but its two-point integration route is not formulation-compatible with QF WEDGE6 production TRI3_X_GAUSS2 integration; the displacement and energy differences are retained as diagnostics, not as a failure or a pass. Code_Aster PENTA6 is now executed through the WP09-R derived headless image: mpi4py was present in the pinned image's Spack view but was not exposed by the stock profile, and --no-mpi does not bypass that import. One affine same-mesh correlation passes for displacement, total reaction and strain energy under the predeclared 1e-6 candidate, which remains OWNER_REVIEW_REQUIRED. No pressure, refinement or distorted external correlation is claimed. WEDGE6 remains EXPERIMENTAL and public qualification remains DEFERRED; WP10 may proceed under its separate modal contract.

WP09-FINAL external qualification closure

WP09-FINAL is recorded as PASS_WITH_LIMITATIONS for a bounded external Code_Aster PENTA6 campaign. The run is sourced from 4b2fcdc9ed51821b05b52851912be3ebbe764b14 and contains 12 deterministic cases: affine tension/compression, shear, bending, TRI3 pressure, QUAD4 pressure, prescribed displacement, a conforming multi-element mesh, a declared affine-skew prism and refinement levels 1, 2 and 4. The headless derived image is Code_Aster 18.1.0 with one launcher process and no GUI.

Displacement, total reaction and strain energy are the primary observables. All 12 cases pass the predeclared comparisons; maximum relative errors are 2.93e-15, 3.60e-15 and 2.48e-15. The 1e-6 affine candidate and the 1e-5 refinement/distortion candidates were fixed before execution and remain OWNER_REVIEW_REQUIRED; no post-result retuning occurred. CalculiX C3D6 is NOT_COMPARABLE under the controlled integration convention and supplies no qualification verdict.

This closes the WP09 external evidence gap without promoting WEDGE6. The public maturity remains EXPERIMENTAL and public qualification remains DEFERRED. The evidence is bounded to the declared homogeneous-isotropic, small-strain, linear-static WEDGE6/PENTA6 slice and does not qualify modal, dynamic, nonlinear, J2, TL, contact, stress-extrapolation or general physical validation routes. The prior 22-case internal WP09 corpus remains preserved.

WP10-FINAL modal qualification decision

WP10-FINAL is PASS within a bounded scope, while gate 027-G10 remains PASS_WITH_LIMITATIONS. The active Owner decision is OWNER_APPROVED_BOUNDED_FIRST_THREE_MODES for homogeneous isotropic small-strain WEDGE6 with consistent translational mass. The final policy was fixed before replay: normalized eigenpair residual <= 1e-7, same-mesh frequency error <= 1e-2, MAC >= 0.99, near-degenerate subspace matching for relative gaps <= 1e-5, and final adjacent refinement change <= 1e-2 for the first three frequencies. Post-result retuning is forbidden.

The four-level sequence 4/8/16/32 passes the refinement rule for the first three modes; modes four to six remain diagnostic because the fourth final change is 1.075439%. Code_Aster 18.1.0/PENTA6 headless comparisons pass on four same-mesh cases (axial single, bending multi-prism, distorted valid prism and multi-WEDGE): all 24/24 modes match, the maximum frequency error is 1.927e-13, and the minimum MAC is 0.9999999999999991. The evidence is tied to source SHA 9d79dc8b306e6cc65f2f4ae2e77e00f676182b84, the final case catalog and the pinned Code_Aster image.

COMB-WEDGE6-modal is now QUALIFIED_BOUNDED only for this declared route and scope. Static evidence is not transferred automatically. Lumped mass, Newmark, harmonic, nonlinear, J2, TL, contact and general dynamics remain outside WP10. The previous 16-case WP10 records remain preserved as historical evidence; the active final records are the wp10_final_* artifacts.

WP12 large-scale readiness checkpoint

WP12 is recorded as PASS_WITH_LIMITATIONS for a bounded readiness study of the existing generated structured TET4 linear-static route. The campaign completed matrix-free CG through 750141 DOF, recorded a separate 311469-DOF assembly-only probe, and classified the 1M attempt as RESOURCE_LIMITED_TIME. SciPy CG/direct limits and PETSc/SLEPc unavailability are explicit. The optimization caches existing grouped connectivity and flattened DOF indices in the matrix-free operator; targeted numerical equivalence passes and no FEM formulation changed. This is evidence for 027-OD-005, not an Owner closeout: the decision remains PROPOSED_OWNER_REVIEW, and no universal 1M claim is made.

WP11 - Existing small-strain J2 maturity

WP11 is recorded as PASS_WITH_LIMITATIONS; WP20 closes Owner review of the bounded maturity record. The existing MAT-J2-SMALL qualification is kept for TET4, TET10, HEX8 and HEX20; no capability is promoted or demoted. The controlled evidence covers material state transitions, radial return, finite difference tangent checks, unload/reload and cycling, multi-element response, energy, rollback, increment sensitivity, full-Newton behavior and explicit failure modes.

The source under test is 94461602dfd1782be57c20e1801a0d5d8e262ef1, from a WP11 start at 4d0ee14f4aa61b9337874a991263a93b4f9a8c73. The maximum tangent FD error is 2.1204721119376345e-10 against the existing 1e-6 limit. Increment partition results are characterization only: no universal structural independence threshold is introduced. Tangent symmetry remains a diagnostic, modified Newton remains diagnostic, finite-kinematic J2 remains experimental/not qualified, and no new external structural run is claimed.

WP15 - Matrix-free TET4 V2 / SPD / preconditioning

Date: 2026-09-01. Execution source snapshot: 6e20bc53d175e5b4eac37a1e76f13266998ce074; baseline snapshot: 2a12f04479eb085137c1d586c99bcf191e702ccc.

Owner decision: PASS_WITH_LIMITATIONS. The structured homogeneous TET4 matrix-free route matches the assembled subscale action, displacement and energy within the frozen WP14 limits on 81, 375, 2,187 and 14,739 DOF. The deterministic SPD checks, residual, equilibrium and energy evidence pass.

Private operator workspaces are retained because they reduce the measured Python allocation peak by 15.4% at 14,739 DOF. No general solve speed-up is claimed: timing remains hardware/run dependent and was neutral-to-slower in the recorded before/after sweep. Nodal block-Jacobi remains the selected WP14-compatible preconditioner; diagonal-Jacobi is characterized but not promoted from this subscale evidence. The scatter bincount remains a future optimization target. WP15 does not qualify 1M DOF; WP16 remains the official release-blocking qualification gate.

027-LEVEL-UP - official extended scope decision

Date: 2026-09-01. Source snapshot: 72888ea63241a4445f6600aa6c2b882401f85ef1.

Owner decision: CLOSED / ACCEPT_WITH_CONSOLIDATION for the portfolio named Reproducible Large-Model Solving and Numerical Trust. The decision accepts WP13-WP22 as the official next scope. WP13 is complete on its controlled golden-baseline evidence and WP14 is complete on its frozen execution contract; WP15 is now PASS_WITH_LIMITATIONS and WP16-WP22 remain open. The decision does not promote a capability or rewrite WP01-WP12 evidence. The WP13 record is in qualification/0_2_7/wp13_state.json.

WP14 - Large-scale execution contract

WP14 is PASS for contract evidence only. The authoritative record is qualification/0_2_7/wp14_execution_contract.json, with state in qualification/0_2_7/wp14_state.json. It freezes the structured TET4 model at 1,029,000 true DOF, four assembled/matrix-free subscale cases, the captured Windows/Python/NumPy/SciPy/OpenBLAS profile, explicit CG and SPD conditions, predeclared residual/equilibrium/energy tolerances, replay requirements, resource safety rules and the 3M Bronze/Silver/Gold ladder.

No 1M or 3M solve was run in WP14, no solve claim is created, no numerical formulation changed, and WP15 is the next execution package.

WP16 is a release blocker and requires a true reproducible 1M-DOF iterative FEM solve with loads, BC, reactions, residual, equilibrium, energy, two replays and full resource/provenance capture. WP18 is mandatory and separates 3M Bronze model/preflight, Silver full solve and Gold PETSc/MPI distributed restart/second-case evidence. Bronze alone authorizes no 3M solve claim.

Deferred to 0.2.8+: mixed TET/WEDGE/HEX, WEDGE15, PYRAMID5, production HEX8R/SRI/B-bar, finite-kinematic J2, TL HEX8, refined Arc-Length, new nonlinear couplings, matrix-free multi-family and general Newmark/harmonic qualification. Machine-readable criteria are in qualification/0_2_7/level_up_plan.json.

WP19 - Adversarial robustness and HEX8 diagnostic

Date: 2026-09-01. Lot start SHA: 7f7ffbaf0b3fdda7d3ad31ba95f20a54e4719a53. Execution source SHA: dc5975b78727d9dca6d0a48b716e60f355b8799f.

Owner decision: PASS_WITH_LIMITATIONS. The 24 predeclared adversarial cases produce 10 positive PASS and 14 EXPECTED_FAILURE_PASS outcomes. Replay is deterministic, failure handling is fail-closed, no NaN/Inf result is accepted, and no QF-specific bug is found.

The HEX8 diagnostic covers axial refinement, slenderness and transverse resolution. Six same-mesh CalculiX 2.20/C3D8 displacement comparisons pass the existing diagnostic threshold with maximum full-displacement relative error 1.997130986610937e-06. The result is classified as LOW_ORDER_LIMITATION with secondary MESH_DEPENDENCE; locking is compatible with the observation but not proven. Reactions and energy are NOT_COMPARABLE in the inherited displacement-only deck. No HEX8R/SRI/B-bar formulation is promoted, no numerical formulation changes, and WP20 closes the J2 review with bounded limitations; WP21 is next.

WP17-FINAL - PETSc/MPI large solver path closure

Date: 2026-09-01. Review start SHA: cd4da89797ff37e0b142d0705aa4fae0972a7065.

Owner decision: PASS_WITH_LIMITATIONS. The pinned PETSc 3.25.1 / MPICH 5.0.1 CG+GAMG route is explicit and reproducible. The controlled evidence contains two passing 1,029,000-DOF replays, a passing same-configuration subscale comparison and two passing 3,000,000-DOF WP18 Silver replays. The WP14 acceptance tolerance remains 1e-8; the supplemental internal PETSc target 1e-10 was predeclared and is not an acceptance-policy change.

PETSc remains optional at runtime and unavailable host selection remains fail-closed. AIJ memory is approximately 3.52 GiB at 1M DOF and 10.08 GB at 3M DOF on the pinned two-rank route. The observed 1M speedup of about 7.4x against the old matrix-free run is case- and environment-specific. No public default backend, general HPC/GPU claim, formulation or existing numerical route is changed. The machine-readable closeout is qualification/0_2_7/wp17_final_state.json.

WP20 - Residual J2 and external V&V closure

Date: 2026-09-01. Review start SHA: 26a734d1656c1c824c27f4708a8783abfddde17c. Evidence source SHA: 94461602dfd1782be57c20e1801a0d5d8e262ef1.

Owner decision: PASS_WITH_LIMITATIONS with OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE. TET4, TET10, HEX8 and HEX20 remain KEEP within the existing QUALIFIED_BOUNDED small-strain J2 scope. The scope is isotropic small-strain J2 with radial return and full Newton for nonlinear static use. No family is promoted or demoted, and finite-kinematic J2 remains experimental/not qualified.

The evidence closes return mapping, yield detection, unload/reload, simple cycling, tangent finite differences, increment characterization, rollback, energy, cross-family consistency, explicit failure modes, no NaN/Inf and deterministic replay. The maximum tangent FD error is 2.120472111937634E-10 against the existing 1E-6 limit. No universal structural increment threshold is claimed; tangent symmetry and modified Newton behavior remain diagnostics.

External V&V is PARTIAL_REUSED_CONTROLLED_EVIDENCE: Code_Aster 18.1.0 constitutive evidence from G06 is reused for all four families. No new structural external campaign or post-result tolerance retuning is claimed. The authoritative records are qualification/0_2_7/wp20_state.json and docs/verification/0_2_7/0_2_7_wp20_j2_closeout.md.

WP21 - Architecture, API and registry surgical cleanup

Date: 2026-09-01. Review start SHA: a24c02e2c95edf374d2e2357c6445afc772bc000. Cleanup source SHA: 0f565dc9669763751a75f13b02004bde18af571c.

Owner decision: PASS_WITH_LIMITATIONS. The active package and runtime identity are aligned on 0.2.7a0; the public audit is clean; legacy solveur and qf_solver imports and the documented launchers remain covered; the registry retains all 33 public capability anchors and 46 combination records. The WP13 golden replay records eight positive passes and one expected-failure pass with no mismatch.

This decision is surgical and does not approve a broad architecture redesign, change any FEM formulation, promote WEDGE6 static, promote finite-kinematic J2, or alter any earlier Owner maturity decision. WP22 remains the separate final release action. The controlled records are qualification/0_2_7/wp21_state.json, qualification/0_2_7/wp21_final_release_truth.json, qualification/0_2_7/wp21_public_document_audit.json and qualification/0_2_7/golden/wp21_replay_evidence.json.

Level-Up 2 setup decision

Decision ID	Topic	Decision	Evidence/SHA	Date
027-OD-012	Level-Up 2 governance installation	CONTINUE_TO_LEVEL_UP_2; LU2 OPEN at 0/50, LU1 CLOSED at 50/50	qualification/0_2_7/level_up_2_plan.json, level_up_2_state.json, 8f08bfb5a6d4dedcd24966f5474e8c12cbfa5bc3	2026-09-02

The decision installs the theme Reproducible Large-Model Performance and Solver Maturity and the nine LU2 work packages with a total weight of 50 %. It does not execute a benchmark, alter a numerical formulation, promote a capability or change the immutable LU1 evidence. C1, C2 and C3 are dormant zero-weight conditional gates. The final gate after LU2-WP09 is installed with the explicit alternatives RELEASE and NEW_LEVEL_UP; publication is never automatic.

LU2-WP01 observatory closeout

Date: 2026-09-02. Execution source SHA: e1703b5bc00e9cf2eb92e7e346783c9764201808.

Owner decision: PASS for the evidence and performance observatory contract. The additive API records phase-separated timings, iterations, matvecs, residual/equilibrium/energy, rank-aware resource metrics, environment and artifact digests. PASS-like evidence requires a committed clean source, input and result SHA-256 digests, a non-empty command and declared environment. Comparisons are descriptive only and never infer a regression or improvement. The legacy benchmark path remains unchanged; missing input provenance becomes NOT_COMPARABLE. The controlled record is qualification/0_2_7/wp01_observatory_sample.json, with contract qualification/0_2_7/observatory_contract.json and implementation src/solveur/verification/observatory.py. No heavy benchmark, full regression, numerical formulation change or maturity promotion occurred. LU2-WP02 was the next work package.

LU2-WP02 configuration freeze closeout

Date: 2026-09-02. Execution source SHA: 3cb817c9391ef7998c5950d3071c8d9ce1be5dd8.

Owner decision: PASS_WITH_LIMITATIONS for the recorded CPU/MPI/PETSc route. The existing 3M structured TET4 FEM workload completed at 2, 4 and 8 MPI ranks with PASS invariants and unchanged WP14 tolerances. Two replays were recorded at 2 and 8 ranks. The selected frozen configuration is contiguous partitioning, AIJ, CG and GAMG in Docker image qf-solver-large@sha256:d6a1718001fc36772906d1a9505637bbd0a4b7e1d8ccc9afdbcb6f67b7ff6d0e, with freeze ID LU2-WP02-FREEZE-bfd1975b012453a3.

HYPRE/BoomerAMG and graph partitioning were characterized on the recorded subscale but were not selected. Preflight, redistribution, communication and I/O are explicitly unmeasured because the legacy runner does not expose separate boundaries; they are not inferred from total time. The evidence is therefore bounded to the recorded structured TET4 input, host, container, rank counts and frozen configuration. No solver/formulation, tolerance or public maturity decision changed. The controlled records are qualification/0_2_7/wp02_execution_contract.json, qualification/0_2_7/wp02_runtime/wp02_evidence_index.json, qualification/0_2_7/wp02_runtime/wp02_config_freeze.json and qualification/0_2_7/wp02_state.json. LU2-WP03 is ready.

LU2-WP03 3M Gold Compute closeout

Date: 2026-09-02. Execution source SHA: 0a6b573485cb39d07b5e179aecd654af41bbc8e7.

Owner decision: PASS_WITH_LIMITATIONS, with 3M_GOLD_COMPUTE = PASS. The existing WP18 Silver result remains the controlled Workload A. A materially distinct real 3,000,000-DOF structured TET4 linear-static workload with a 2.0 m x 0.75 m x 1.25 m block geometry completed two PASS replays under the unchanged WP02 freeze: PETSc 3.25.1, MPICH 5.0.1, 8 ranks, contiguous partitioning, AIJ, CG and GAMG. Both runs passed the frozen residual, equilibrium, energy, finite-output, no-NaN/Inf and SPD checks; each used 1,046 iterations. The replay maximum numerical relative delta was 6.06e-13.

The claim is bounded to structured TET4 homogeneous isotropic linear-static FEM on the pinned single-host Docker/PETSc/MPI configuration. Preflight, redistribution, communication and I/O remain explicitly NOT_MEASURED because the runner does not expose those phase boundaries. Workload A/B timings are descriptive only because their geometries differ; no speedup or universal 3M, multi-node, GPU, mixed-mesh, nonlinear or restart claim is made. The controlled records are qualification/0_2_7/wp03_execution_contract.json, qualification/0_2_7/lu2_wp03_state.json, qualification/0_2_7/wp03_runtime/wp03_evidence_index.json, qualification/0_2_7/wp03_runtime/wp03_summary.json and qualification/0_2_7/wp03_runtime/wp03_replay_comparison.json. LU2-WP04 is the next work package.

LU2-WP08 scope decision closeout

Date: 2026-09-02. Decision source SHA: 8ef34e345f970879548a4dfdce4ac5ba32c11bda.

Owner decision: PASS_WITH_LIMITATIONS. The controlled matrix is qualification/0_2_7/lu2_wp08_decision_matrix.json. Mixed TET/WEDGE/HEX is partial at infrastructure level but has no qualified end-to-end route and is deferred. WEDGE15 and PYRAMID5 are not supported or active capabilities and are deferred. The existing HEX8 route remains bounded by WP19; HEX8R, SRI and B-bar are research-only, and hourglass control is deferred with any future reduced-integration route. No element, formulation, public maturity or large-model benchmark changed. The WP04 supervised retry remains the next operational action.

LU2-WP06 execution and recovery closeout

Date: 2026-09-02. Decision source SHA: 4771a23af8cee4549460b9e84edb9228c3a9f60d.

Owner decision: PASS_WITH_LIMITATIONS. The additive execution lifecycle, stable diagnostic taxonomy and fail-closed checkpoint boundary are installed and covered by focused tests. Existing nonlinear static, arc-length and Newmark checkpoint routes retain the recovery claim; linear, buckling, harmonic, geometric nonlinear and large distributed routes remain explicitly non-recoverable under this contract. No universal execution graph, timeout recovery, distributed recovery or fault-tolerance claim is made. The controlled records are qualification/0_2_7/wp06_execution_contract.json, qualification/0_2_7/lu2_wp06_state.json, docs/verification/0_2_7/0_2_7_wp06_execution_contract.md and tests/unit/test_execution_contract.py. LU2-WP07 is ready as an independent targeted-V&V work package; the separately pending WP04 supervised retry is unchanged.

LU2-WP07 existing route maturity closeout

Date: 2026-09-02. Review source SHA: 65b9d2b168ac7be1df1c0c1cd2d58e8286d4af00.

Owner decision: PASS_WITH_LIMITATIONS. The machine-readable maturity matrix is qualification/0_2_7/lu2_wp07_maturity_matrix.json, with state in qualification/0_2_7/lu2_wp07_state.json. Linear static, modal, buckling, harmonic, Newmark, small-strain J2 and frictionless contact retain their existing bounded scopes. Nonlinear static, Arc-Length and static WEDGE6 remain experimental; HEX8 buckling, finite-kinematic J2, friction, coupled routes and deferred candidates remain outside qualified claims. WP06, WP19 and WP20 evidence was reused, no new external case was run, and no maturity promotion or demotion was made. LU2 progress is now 32/50 and 82/100 globally. The LU2-WP04 supervised retry remains unchanged.

F5 packaging and compatibility audit

Date: 2026-09-04. Audit start SHA: 124c61f6492eee351a34e3542d198a13c00c2874.

Owner decision recorded by the controlled F5 audit is PASS_WITH_LIMITATIONS. The candidate 0.2.7a0 wheel and sdist build and install cleanly in the directly verified Windows Python 3.12/3.13 environments, and the public API, CLI and runtime resources work from outside the source tree. The audit found and fixed one P1: two qualification matrices required by the public release-vv entry point were omitted from the sdist manifest and wheel data files. No numerical source, historical evidence or maturity was changed.

This decision is bounded. Linux is covered by the existing CI matrix but was not locally artifact-tested; Python 3.10 is CI-covered, Python 3.11 and macOS were not directly verified; optional PETSc/MPI/HDF5 routes still require their declared environments. The installed release-vv command returns a controlled report/status for its bundled legacy 0.2.1 context and this is not a 0.2.7 release qualification claim. Tagging, publication and F6 remain separate Owner actions. The controlled record is qualification/0_2_7/f5_packaging_compatibility_audit.json.

F6 numerical and performance regression audit

Date: 2026-09-04. Audit start SHA: f6cfde036f5866c15e688bce70be5ed21b493ff1.

The controlled F6 audit closes as PASS_WITH_LIMITATIONS. Targeted element, solver, BC/load/material, post-processing, MPI/PETSc, baseline-integrity and performance-sanity checks passed. The full suite completed with 2147 passed, 3 failed, 184 skipped, 2 warnings; the three visible failures are exactly the pre-existing F4 failures in experimental or stale nonlinear paths. No P0 or P1 remains for the supported 0.2.7 scope.

The existing 1M, 3M and 5M replay evidence, representative AIJ preallocation record and bounded C3 10M record remain applicable. No relevant numerical source change occurred during F3-F6, so no new heavy 5M/10M replay was required. No source, tolerance, active baseline or maturity was changed.

The performance interpretation remains bounded to each recorded workload, host, PETSc/MPI environment, backend, preconditioner and configuration. No universal speedup or hardware-independent scaling claim is authorized. F6 is ready for the separate Owner R0 decision; R0, publication, tagging and push remain deferred. The controlled record is qualification/0_2_7/f6_numerical_performance_regression_audit.json.

R0 release readiness closeout

Date: 2026-09-04. R0 starts from F6 closeout SHA 5323a0996f214a3203f06b5bb468843b57c25270 and records PASS_WITH_LIMITATIONS. One modification to the old 0.2.1 hygiene report was found during coactivity inspection. It was classified as foreign and preserved in targeted stash a9e64d5e03ebadfcd1292ad3c5b04460761c10aa; it is not included in the release tree or the R0 commit.

The candidate tree is clean, F1-F6 are closed, active qualification accounting remains 96/100, and closeout smoke/guard tests pass. The 5M Silver result and bounded C3 10M result remain applicable. No P0/P1 release blocker was found, and no numerical source, active baseline, historical evidence or maturity was changed.

The package remains 0.2.7a0. The wheel and sdist pass distribution checks and F5 clean-install evidence remains valid. R0 deliberately does not create the v0.2.7a0 tag, GitHub release or PyPI publication. The generic readiness tool therefore continues to report the expected pre-publication tag failure and a reachable-history manual-review item. The Owner recommendation is RELEASE_WITH_KNOWN_LIMITATIONS, subject to the explicit manual review before irreversible publication actions.