QF Solver 0.2.5a0 gate matrix

The matrix records the final controlled status after the Owner scope decision dated 2026-08-28. Gate status values are restricted to PASS, OPEN, BLOCKED and NOT_IN_RELEASE_SCOPE. Evidence labels such as PASS_INTERNAL or PASS_EXTERNAL_CORRELATION_BOUNDED describe the supporting proof, but do not replace the gate status. Only evidence generated on the recorded SHA can close a gate. NOT_IN_RELEASE_SCOPE is a closed governance status only when the controlled Owner decision records the explicit exclusion; it is never evidence of a functional PASS.

G01 is closed by controlled internal and Code_Aster evidence under the stable g01_latest and g01_code_aster_latest output paths. Both final manifests record the same clean candidate source revision and artifact digests. The Owner-approved treatment of mesh/load-step trends and rollback reference differences is bounded and does not create a universal convergence claim or lower any requirement. G02 is closed for the Owner-approved bounded elastic Total-Lagrangian scope defined below. The qualified numerical source remains distinct from the documentary Owner decision. G04 and G06 retain provisional observations and are excluded from the qualified 0.2.5a0 claim by the explicit Owner scope revision; neither is a functional PASS. G03 is closed for the bounded TET4 linear-buckling scope defined below. G08 is closed by the controlled replay recorded below. G09 has a separate controlled failure campaign and is closed below; this does not close any dependent functional gate.

Gate	Name	Mandatory closure criteria	Dependencies	Status
025-G00	Baseline and architecture frozen	exact 0.2.4 SHA; provenance reconciled; tests/coverage/V&V/performance/API baselines; Owner audit approval	none	PASS
025-G01	0.2.4 nonlinear V&V debt closed	multi-element four-family J2; mesh/load-step/cyclic/energy/rollback/tangent evidence; bounded external curves; Owner-approved acceptance treatment	G00	PASS
025-G02	Geometric nonlinear core verified	approved measures; TET4/HEX8 objectivity, tangent, energy, mesh and external evidence; common Full Newton	G01	PASS
025-G03	Linear buckling verified	sparse eigenpath; Euler + nontrivial benchmark; factor/mode convergence and external correlation	G02	PASS
025-G04	Arc-length verified	one sparse method; branch, limit point, restart, cutback and external response evidence	G02	NOT_IN_RELEASE_SCOPE
025-G05	Frictionless contact verified	common residual/tangent/state/Newton; finite sliding, recontact, penetration sensitivity, rollback, external correlation	G02	PASS
025-G06	Coupled nonlinear core verified	approved J2 finite-kinematics model; J2+geometry and geometry+contact; limit recovery, tangent, energy, mesh and external evidence	G01, G02, G05	NOT_IN_RELEASE_SCOPE
025-G07	Frictional contact verified	only after Owner promotion; objective stick/slip, state, dissipation and external evidence	G05, G06, Owner GO	NOT_IN_RELEASE_SCOPE
025-G08	Performance characterized	reproducible cost/memory profiles for all mandatory paths; HEX20 explained; numerical non-regression after optimization	relevant functional gates	PASS
025-G09	Failure modes verified	complete mandatory failure matrix; structured reasons; no false convergence; exact rollback	relevant functional gates	PASS
025-G10	External correlation bounded	all mandatory external matrix cells complete for the Owner-approved qualified scope; excluded claims remain explicitly excluded	G01-G03, G05; approved G04/G06 exclusions	PASS
025-G11	Full regression	complete 0.2.4 + accepted 0.2.5 tests, coverage policy, docs, V&V, build and smoke pass on candidate SHA	replay evidence; functional release gates remain reported separately	PASS
025-G12	Aggregate release readiness and Owner closure	aggregate review of gate states, final-SHA evidence, qualification/README/changelog/metadata consistency, limitations, artifacts and explicit Owner release decision	G00-G11 under approved final scope	PASS

OWNER_SCOPE_REVISION = APPROVED SCOPE_CHANGE = YES CONTRACT_LOWERED = NO (no requirement, tolerance or historical result was relaxed; the qualified release scope was explicitly narrowed)

Final Owner scope disposition

The controlled Owner decision is recorded in 0_2_5_owner_scope_revision.md. It accepts the 0.2.4 replay limitation and narrows the qualified 0.2.5a0 release claims without rewriting the original requirements. The original plan classified G04 and G06 as MUST; their historical evidence and blockers remain visible below and in the V&V pack.

Gate	Gate status in this final matrix	Evidence/claim classification	Release treatment
025-G00	PASS	PASS_WITH_HISTORICAL_LIMITATION	accepted by Owner; the exact 0.2.4 replay remains recorded as 1440 passed / 1 failed / 32 skipped / 187 deselected, with two historical RQ-G08 governance findings
025-G04	NOT_IN_RELEASE_SCOPE	EXPERIMENTAL / NOT_QUALIFIED	no arc-length qualified claim; published compatible reference and four-level closure remain future work
025-G06	NOT_IN_RELEASE_SCOPE	CODE_COMPLETE / EXPERIMENTAL / QUALIFICATION_DEFERRED	no finite-kinematic J2 or coupled qualified claim; external gaps remain future work
025-G10	PASS	bounded external correlation	evaluated only on the remaining qualified MUST cells; G04/G06 cells are excluded, not passed
025-G12	PASS	aggregate readiness	applies to the final qualified scope and does not authorize tag, push or PyPI publication

CalculiX remains a SHOULD/supporting comparison. A positive or negative CalculiX result is not promoted to MUST and does not by itself block G10.

Gate closure record

Every gate record includes:

• gate ID/status and exact SHA;
• requirements and V&V rows covered;
• command/environment and artifact links/digests;
• measured values and frozen thresholds;
• skips/exclusions and justification;
• known limits and residual risks;
• reviewer, date and Owner decision where required.

Historical G10 sweep before Owner scope revision

Before the explicit Owner scope revision, 025-G10 was BLOCKED by the still-open mandatory external cells associated with 025-G04 and 025-G06. The external matrix is classified as follows:

Capability row	Current classification	Release consequence
J2 multi-element	PASS_BOUNDED	Two-cell Code_Aster history is supporting bounded evidence; it is not a complete G10 closure for every claimed path
Large deformation	PASS_BOUNDED	Bounded TET4/HEX8 elastic evidence only
Linear buckling	PASS_BOUNDED	Euler and bounded Code_Aster TET4 evidence; CalculiX remains a non-blocking SHOULD with recorded deviations/limits
Arc-length	OPEN	G04 remains open: the required independent reproducible reference and four-level branch study are not closed
Frictionless contact	PASS_BOUNDED	Bounded G05 contact contract only; no general external surface-to-surface claim
J2 + geometry	OPEN	G06 MUST: finite-kinematic J2 is not an approved model and the Code_Aster comparison is incomplete/non-comparable for the required scope
Geometry + contact	OPEN	G06 MUST: the finite-kinematic comparison has a mapped reaction deviation and no equivalent finite-sliding external history
Triple coupling	NOT_REQUIRED	SHOULD not promoted to a 0.2.5 release requirement
Friction	NOT_REQUIRED	COULD and G07 remains NOT_IN_RELEASE_SCOPE

CalculiX is not promoted from SHOULD to MUST. Its available positive or negative results remain supporting evidence and do not replace the Code_Aster MUST cells. No requirement or tolerance is lowered by this decision.

Controlled closure: 025-G09

025-G09 is closed by the controlled campaign at results/vnv_0_2_5/g09_latest/summary.json and its evidence_manifest.json. The manifest records the exact source SHA, dirty=false, command, tool versions and SHA-256 digest of the report. The campaign contains 22 intentional failure cases and reports 22 passed / 0 failed, with converged=false and a structured diagnostic for every case. The campaign is an internal failure-contract qualification; it does not promote contact, arc-length, buckling or coupled capabilities to qualified status and does not close their functional gates.

Controlled closure: 025-G01

025-G01 is PASS. The internal manifest at results/vnv_0_2_5/g01_latest/evidence_manifest.json records PASS_INTERNAL_J2, dirty=false, the exact candidate source revision and digests for the constitutive, tangent, transaction, four-family, mesh/load-step, cyclic, energy and rollback evidence. The Code_Aster manifest at results/vnv_0_2_5/g01_code_aster_latest/evidence_manifest.json records PASS_EXTERNAL_CORRELATION, dirty=false and 64/64 comparable checks for TET4, TET10, HEX8 and HEX20.

The closure uses the Owner-approved bounded classifications in the J2 report: mesh/PEEQ, load-step sensitivity and rollback numerical comparison are ACCEPTED_BOUNDED_OBSERVATION; the rollback transaction invariant is THRESHOLD_JUSTIFIED. CONTRACT LOWERED = NO. This closure changes no other gate.

Controlled closure: 025-G08

025-G08 is PASS for bounded performance characterization. The final controlled replay is archived under results/vnv_0_2_5/g08_latest/. Its manifest records the exact clean source SHA, dirty=false, commands, runtime versions, environment data and SHA-256 digests for the raw samples, aggregate summary and report.

The mandatory load-control profile contains 12 PASS samples: three repeats for each of TET4, TET10, HEX8 and HEX20. The same clean-SHA campaign also contains one-repeat characterization smokes for the geometric-static, arc-length, contact and coupled paths. These smokes document shared-driver costs; they are not claims that those functional gates are closed.

The aggregate report quantifies total, element-kernel, assembly and sparse solve costs, Python allocation peaks, RSS observations, Newton iterations and residuals. It explicitly identifies the HEX20 kernel as the dominant measured cost and records the corresponding kernel/assembly and kernel/total ratios. The repeated numerical fields and convergence status are stable across the three load-control repeats. No speedup, multi-million-DOF scalability or memory-efficiency claim is inferred from this local bounded campaign. The closure therefore satisfies the existing G08 contract without changing any functional gate or acceptance threshold.

Controlled closure: 025-G02

The dedicated pack at results/vnv_0_2_5/g02_latest/ records the exact clean source SHA, dirty=false, reproducible command, environment and SHA-256 digests. It contains:

• rigid translation, 0.7 rad rigid rotation and combined translation/rotation objectivity checks for TET4, TET10, HEX8 and HEX20;
• sparse internal-force tangent finite-difference checks for all four families;
• bounded TET4/HEX8 large-rotation paths above 0.5 rad with positive det(F);
• four-level coarse/medium/fine/refined pre-limit mesh observations for TET4 and HEX8;
• four-family small-strain-limit recovery;
• twelve-step pinned Code_Aster 18.1 TET4/HEX8 histories for displacement, reactions, SIEF_ELGA stress and EPSI_ELGA strain.

The source pack intentionally remains OPEN: it faithfully records the state before the Owner decision and is not rewritten after the numerical campaign. The Owner decision record 0_2_5_g02_owner_review.md accepts the coherent four-level pre-limit trend as APPROVED_BOUNDED_REFINEMENT, without inventing a universal mesh-convergence band. The accompanying review manifest distinguishes the qualified source SHA from the documentary Owner-evidence SHA.

025-G02 is therefore PASS only for the following bounded scope:

• elastic Total-Lagrangian finite-deformation statics using the common Full Newton path;
• TET4 and HEX8;
• the recorded pre-limit, positive-det(F) load domain and the associated objectivity, tangent, energy, mesh-trend and Code_Aster evidence.

The load-control line-search failure at the physical stability boundary is retained as a limitation and is not recast as a defect or a PASS. This closure does not qualify total_lagrangian_j2, finite-kinematic plasticity, TET10/HEX20 finite-kinematic behavior, post-limit response, buckling, arc-length, contact or coupled paths; G03-G06 remain unchanged. The Code_Aster result is bounded numerical code-to-code correlation, not physical validation.

Controlled closure: 025-G03

025-G03 is PASS for a bounded first tangent-instability scope. The final internal Euler campaign is archived under results/vnv_0_2_5/g03_euler_final/; it contains four structured TET4 levels (16x4x4, 24x6x6, 32x8x8, 40x10x10) with sparse initial-stress geometric stiffness, positive precritical det(F), relative brackets below 5e-3, and a finest-level Euler error of 5.89 %. The successive critical load changes are recorded as 15.63 %, 6.57 % and 3.31 %; this is a bounded refinement trend, not a universal mesh-convergence claim.

The nontrivial external probe is archived under results/vnv_0_2_5/g03_final/ and is linked to qualified source SHA 85c75d06955976251dd54ad782f57f1eb5a7f8f4. QF Solver computes a critical factor of 221.54828247814925 against 221.774 from the pinned Code_Aster 18.1.0 execution, a relative difference of 1.018e-3. The best modal MAC is 0.9999999989229131 and the QF critical-mode residual is 1.72e-15.

This closure follows a demonstrated numerical correction: the buckling path now assembles only the sparse initial-stress geometric contribution instead of treating a full tangent difference as Kg. The exact TET4 unit test and the targeted buckling suite are green. The current Docker image could not replay Code_Aster in this environment because its launcher lacks mpi4py; the archived Code_Aster deck, image digest and mode output remain preserved as the external execution evidence. CalculiX remains a SHOULD cell: its bounded TET4 campaign is retained as supporting evidence and its blocked high-order probe is retained as negative evidence, but neither is required to close G03.

The qualified scope is limited to sparse linear buckling / first tangent instability with the supported total-Lagrangian solid path and the recorded TET4 external probe. TET10, HEX8 and HEX20 remain internal/research for this gate; post-buckling, imperfection-sensitive collapse and general stability prediction remain outside the claim. The Owner decision is APPROVED with mesh decision APPROVED_BOUNDED_REFINEMENT; CONTRACT LOWERED = NO.

Historical Owner audit: 025-G04 was open before scope revision

The controlled audit is recorded in 0_2_5_g04_owner_review.md, with the strict model/path comparison in 0_2_5_g04_external_branch_diagnostic.md and the targeted pack under results/vnv_0_2_5/g04_latest/. The common-driver QF Solver path records one signed load-factor turn, exact restart suffixes and a controlled cutback/retry, but remains PASS_INTERNAL_RESEARCH on a two-element TET4 model. No four-level arc-length mesh study is available.

The historical monotone Code_Aster result has been resolved as a deck configuration mismatch: it had the opposite physical force direction and used mean crown displacement rather than the QF apex control DOF. The corrected pinned Code_Aster 18.1 Docker replay uses downward FZ=-1/3, APEX/DZ, and a matched continuation window; it produces one turn and agrees with the QF equilibrium branch by a peak-normalized load-factor difference of 1.3052e-05. This is bounded code-to-code diagnostic evidence, not physical validation or a gate closure.

At the time of this audit, 025-G04 was OPEN, CONTRACT LOWERED = NO, and no arc-length production claim was promoted. The remaining functional blockers were the missing published FEM branch reference and the required four-level mesh study. The final Owner decision does not erase those blockers; it excludes G04 from the 0.2.5a0 qualified claim.

Historical controlled G06 targeted evidence

The final controlled G06 evidence is archived under results/vnv_0_2_5/g06_latest/. Its summary.json, report.md and evidence_manifest.json record qualified source SHA 8df4b4ac32e9416e89fe342871aab6e75cdd245c, dirty=true at the current documentary capture, and artifact digests. The dirty flag reflects the uncommitted documentation updates; no solver source changed. The replay covers the existing pairwise/triple composition checks, the four-family J2 plus Total-Lagrangian path, the four-family updated penalty-contact composition, a three-level (1/2/4) J2 plus geometry mesh study, an independent global coupled tangent finite-difference check, and a three-level geometry/contact mesh replay. All targeted internal paths converged and remain classified PASS_INTERNAL_RESEARCH; the independent tangent checks are PASS_INTERNAL.

The original complete step-level history capture is archived as results/vnv_0_2_5/g06_latest/coupled_histories.json with original capture SHA a56db0863835ee16485adf5c9d30954c2f425ecb. The aggregate manifest records that only documentation changed between that clean capture and the current qualified SHA; no solver source changed.

This evidence did not close 025-G06. The independent coupled tangent FD and the bounded geometry/contact mesh replay are now closed as internal sub-proofs, but they do not replace the external MUST. A separate pinned Code_Aster GREEN_LAGRANGE replay reaches the end of the bounded history for TET4, HEX8 and HEX20; TET10 stops at the first load point after the configured Newton limit. The measured QF/Code_Aster deviations are archived for convention review, not promoted to unconditional external PASS. The historical GDEF_LOG replay remains archived separately, including its HEX20 non-convergence.

The Code_Aster CONTINUE frictionless surface oracle also executes and is recorded as bounded supporting evidence. Its discrete support springs require DEFORMATION='PETIT'; a direct GREEN_LAGRANGE probe is rejected by Code_Aster for that reason. It is therefore not a comparable finite-kinematic 3D solid geometry/contact full-history correlation and does not close the G06 MUST. A separate deformable TET4 GREEN_LAGRANGE probe reaches ten load points and agrees in displacement/gap after contact activation, but its mapped contact reaction differs by up to 76.7% and is not an identical multiplier observable; it remains open comparison evidence. The finite-kinematic J2 path remains experimental/research; no qualified coupled or physical-validation claim is made. In the compared QF static route, contact is solved by the historical exact active-set multiplier solver; the penalty formulation belongs to the separate bounded research composition. Neither route establishes an equivalent finite-sliding external correlation. CalculiX remains supporting SHOULD evidence and cannot replace the missing Code_Aster MUST cells. CONTRACT LOWERED = NO.

Controlled closure: 025-G05

025-G05 is PASS for the bounded contract in 025-REQ-018 through 025-REQ-021. The controlled evidence pack is archived under results/vnv_0_2_5/g05_latest/ and records source SHA a3ab8de707ffc88fc5e39e4f999eb872c9223b73 with dirty=false.

The targeted contact suite reports 82 passed / 2 skipped. The internal pack covers sparse common-driver assembly, fixed-active tangent FD (approximately 6e-9), open/close/recontact, updated normals, two-face and three-facet traversal, facet-transition rollback, and penalty values from 1e2 through 1e6. The Code_Aster 18.1.0 Docker campaign compares ten-point histories for the bounded corner, faceted-ramp and deformable TET4 cases; the 768-element and 9,984-element replays both return PASS_EXTERNAL_CORRELATION. The 768- element case retains an explicit transition observation: the second contact activates at a different first load sample and the displacement-curve difference is 4.33998 %; the active-branch gap check remains below 4.1e-16 m. The 9,984-element confirmation removes that activation mismatch for the recorded ten-point history.

The qualified claim is deliberately limited to frictionless penalty contact from a slave node/patch to an explicitly supplied triangulated master surface, with opt-in updated search and bounded finite-sliding projection. This is not a mortar or segment-to-segment formulation and does not qualify unrestricted surface-to-surface contact, self-contact, impact, friction or general large sliding. The finite-sliding and updated-normal paths are internally qualified; the external correlation is a bounded compatible normal-contact correlation. CalculiX remains supporting evidence only and is not a G05 closure condition.

CONTRACT_LOWERED = NO. At capture time, G05 closure did not close G06, G10, G11 or G12. The final Owner scope revision is recorded separately and does not promote the G06 evidence.

STOP/GO policy

An OPEN or BLOCKED prerequisite forbids dependent implementation. Optional independent branches may continue only when their full dependency chain is closed. NOT_IN_RELEASE_SCOPE is acceptable for predeclared optional work or for a formally approved final-scope exclusion. It must never be presented as a functional PASS or appear as an unqualified README/package capability claim.