0.2.7 Gate Matrix¶
Current release reconciliation¶
The active Owner-complete state for stable v0.2.7 is 100/100. WP22 is
closed as PASS_WITH_LIMITATIONS through R0 and Step 1. The 46/50,
96/100, WP22 PLANNED and 0.2.7a0 values retained below are historical
planning or pre-closure checkpoints only.
This matrix is a controlled status record. WP01 is PASS for release truth, WP02 is
PASS for the combination-level registry control, and WP03 is PASS for the
technical descriptor and fail-closed preflight control, and WP04 is PASS for
the additive V&V harness control. WP05 is PASS for the external deck
preflight only; it issues no WEDGE6 QF result or capability. WP06 is PASS
for the additive mesh-quality diagnostic contract; it introduces no new
numerical capability or universal quality cutoff. The WP07 row is technical
kernel evidence only: WEDGE6 remains EXPERIMENTAL and is not publicly
qualified. WP08 adds the bounded static vertical slice without changing that
maturity. WP10 separately records consistent-mass modal evidence and bounded
Code_Aster frequency/MAC correlation; its modal combination is
QUALIFIED_BOUNDED only within the declared scope.
WP12 records bounded large-scale readiness evidence for the existing structured
TET4 route; its Owner decision remains pending and its resource limits are not
success claims. WP11 records a bounded maturity extension for existing
small-strain J2 evidence across TET4, TET10, HEX8 and HEX20; it keeps the
qualified scope and does not create a universal increment-independence claim.
WP18 records a PASS_WITH_LIMITATIONS 3M Bronze/Silver ladder under the frozen
WP14 contract; Gold remains unattempted and no Gold claim is made. WP20 closes
the Owner review of the existing bounded small-strain J2 scope with
PASS_WITH_LIMITATIONS; WP21 is now PASS_WITH_LIMITATIONS for surgical cleanup
and WP22 is closed as PASS_WITH_LIMITATIONS through the final Owner release
action.
| Gate | Work package | Purpose | Dependencies | GO evidence | STOP conditions | Status |
|---|---|---|---|---|---|---|
027-G01 |
WP01 | Release truth and provenance | None | exact baseline, clean-state and version records | SHA, tag or version mismatch | PASS |
027-G02 |
WP02 | Element x analysis x material x route registry | G01 | schema, uniqueness and implementation cross-check | inferred maturity or orphan capability | PASS |
027-G03 |
WP03 | Compatibility descriptors and preflight | G02 | deterministic accept/reject matrix | default-path or API behavior change | PASS |
027-G04 |
WP04 | Declarative V&V harness | G01-G02 | additive runner contract and replay proof | numerical route refactor or hidden command execution | PASS |
027-G05 |
WP05 | C3D6/PENTA6 oracle preflight | G01-G02 | comparable deck contract and availability record | non-comparable oracle presented as PASS | PASS |
027-G06 |
WP06 | Mesh quality/distortion contract | G02-G03 | metrics, invalid cases, deterministic classification and preflight diagnostic | universal aspect-ratio cutoff | PASS |
027-G07 |
WP07 | WEDGE6 kernel and elemental V&V | G03, G05, G06, T1-R | reviewed contracts, six-node kernel, stiffness/recovery checks, exact quality certificate and deterministic evidence | certified inversion not rejected; rank or rigid modes unexpected; affine patch or quadrature check fails; existing element formulation changed | PASS |
027-G08 |
WP08 | WEDGE6 static vertical slice | G07 | 15-case Gmsh/import/load/assembly/equilibrium/post evidence with deterministic replay | missing face/load, silent unsupported path or unexpected failure | PASS |
027-G09 |
WP09 | WEDGE6 robustness and external V&V | G08 | mesh/distortion/adversarial evidence with explicit external availability and comparability outcomes | unexplained divergence, weakened threshold or false external PASS | PASS_WITH_LIMITATIONS |
027-G10 |
WP10 | WEDGE6 modal evidence | G08-G09 | mass, modes, residual, mesh, replay and bounded external frequency evidence | static evidence transferred without modal proof; hidden mass convention; unexplained non-finite modes | PASS_WITH_LIMITATIONS |
027-G11 |
WP11 | Existing maturity and J2 gaps | G01, G04, G06 | all-family J2/increment/tangent evidence | formulation change or relaxed acceptance | PASS_WITH_LIMITATIONS |
027-G12 |
WP12 | Large-scale readiness | G02, G04, G06 | declared size/resource measurements, bounded backend/resource verdicts and deterministic replay | universal 1M claim or uncharacterized failure | PASS_WITH_LIMITATIONS |
027-G13 |
WP13 | Research/stretch selection | relevant closed prerequisites | explicit Owner scope selection | scope creep or transitive promotion | NOT_STARTED |
027-G14 |
WP14 | Release closeout | G01-G13 as applicable | docs, full regression, package and Owner review | stale claim, provenance gap or regression | NOT_STARTED |
027-G07 required a design review before implementation. The external oracle
preflight and formulation review preceded the kernel, and the current PASS is
limited to the elemental technical route. The descriptor reports
EXPERIMENTAL_ROUTE; WP08 provides a bounded static user workflow, and WP09
adds controlled robustness evidence without creating a public qualification.
CalculiX C3D6 is explicitly non-formulation-compatible with the QF production
quadrature. WP09-R repairs the headless Code_Aster path and records one
bounded affine PENTA6 correlation through a derived image pinned to the base
image digest. WP10 modal evidence is recorded separately for its own claims.
The consistent mass, finite-positive modes, residual, orthogonality, replay and
zero-density fail-closed checks pass. The four-level 4/8/16/32 refinement
passes for the first three modes under the predeclared 1e-2 final-change
rule; modes four to six remain diagnostic. Code_Aster 18.1.0/PENTA6 provides
bounded same-mesh frequency and MAC evidence over four cases, with 24/24
matched modes. COMB-WEDGE6-modal is QUALIFIED_BOUNDED only for the
declared consistent-mass scope; no general dynamics claim is made. G13 may remain open
or deferred without blocking a bounded core release when research routes are
explicitly excluded.
027-G12 is PASS_WITH_LIMITATIONS for the declared structured TET4
linear-static readiness scope. Matrix-free CG completed 100k through 750k
target levels (actual maximum 750141 DOF); 300k assembly-only evidence reached
311469 DOF; SciPy direct was resource-limited at 107811 DOF; higher SciPy
levels were rejected by the configured 200000-DOF guard; and the 1M
matrix-free attempt was time-limited after 600 seconds. These results are
hardware- and topology-specific and do not qualify a universal 1M solve.
The evidence is recorded in
qualification/0_2_7/wp12_scaling_evidence.json,
qualification/0_2_7/wp12_assembly_probe_300k.json and
0_2_7_large_scale_readiness.md; 027-OD-005 remains
PROPOSED_OWNER_REVIEW.
Official Level-Up portfolio¶
027-LEVEL-UP is CLOSED / ACCEPT_WITH_CONSOLIDATION: the scope was
accepted, and the executed WP16 and WP18 results are recorded separately
below.
WP13 is PASS on its controlled golden-baseline evidence. WP14 is PASS on
its frozen execution contract; WP15 is PASS_WITH_LIMITATIONS on controlled
subscale evidence; WP16 is PASS after the official PETSc retry; WP17 is
PASS_WITH_LIMITATIONS after its pinned PETSc/MPI closure; WP18 is
PASS_WITH_LIMITATIONS after the Bronze/Silver ladder; and WP19 is
PASS_WITH_LIMITATIONS on bounded adversarial and HEX8 diagnostic evidence.
WP20 is PASS_WITH_LIMITATIONS for the existing bounded small-strain J2
scope. WP21 is PASS_WITH_LIMITATIONS for surgical compatibility, public
audit and release-truth cleanup; WP22 is PASS_WITH_LIMITATIONS through the
final Owner release action. The active theme is Reproducible Large-Model Solving and
Numerical Trust.
| Level-Up gate | Work package | Status | Weight | Required interpretation |
|---|---|---|---|---|
LUP-027-G13 |
WP13 release truth and golden baseline | PASS |
4% | distinct SHA roles and golden baseline |
LUP-027-G14 |
WP14 large-scale execution contract | PASS |
5% | frozen topology, resource and execution contract; no solve claim |
LUP-027-G15 |
WP15 matrix-free TET4 V2 | PASS_WITH_LIMITATIONS |
10% | SPD, preconditioning and assembled subscale; no WP16/1M claim |
LUP-027-G16 |
WP16 true 1M DOF qualification | PASS |
10% | two PETSc CG/GAMG replays satisfy frozen residual, equilibrium, energy and SPD criteria |
LUP-027-G17 |
WP17 PETSc/MPI and large sparse path | PASS_WITH_LIMITATIONS |
5% | two-replay 1M and 3M Silver evidence on the pinned route; host availability and AIJ memory remain limitations |
LUP-027-G18 |
WP18 3M Bronze/Silver/Gold ladder | PASS_WITH_LIMITATIONS |
7% | Bronze preflight and two-replay Silver solve; Gold unattempted |
LUP-027-G19 |
WP19 adversarial robustness and HEX8 diagnostic | PASS_WITH_LIMITATIONS |
5% | 24-case fail-closed corpus and bounded HEX8/C3D8 diagnostic; no formulation promotion |
LUP-027-G20 |
WP20 residual J2 and external V&V closure | PASS_WITH_LIMITATIONS |
3% | existing four-family small-strain J2 scope retained; no universal structural increment claim |
LUP-027-G21 |
WP21 surgical architecture/API/registry cleanup | PASS_WITH_LIMITATIONS |
3% | legacy compatibility preserved; broader redesign deferred |
LUP-027-G22 |
WP22 final release qualification | PASS_WITH_LIMITATIONS |
3% | final regression/package/Owner checkpoint |
Machine-readable criteria are in qualification/0_2_7/level_up_plan.json and
qualification/0_2_7/requirements.json. WP16 is satisfied and WP18 records
the mandatory Silver result. Bronze creates no solve claim, Gold remains
unattempted, and no Level-Up record promotes a capability.
WP18 evidence is recorded in
qualification/0_2_7/wp18_runtime/wp18_summary.json and
qualification/0_2_7/wp18_state.json. The 3M Silver route completed twice
at 3,000,000 true DOF with residual, equilibrium, energy, finite-output and
SPD checks passing. The claim is limited to the structured TET4 linear-static
model and the pinned two-rank PETSc/MPI environment; AIJ setup and peak RSS
remain material limitations.
027-G19 adversarial robustness and HEX8 diagnostic¶
WP19 evidence is recorded in qualification/0_2_7/wp19_state.json and the
linked runtime artifacts. The 24 predeclared cases produced 10 positive
PASS results and 14 EXPECTED_FAILURE_PASS results, with deterministic
replay, fail-closed failure paths and no accepted NaN/Inf result. The failure
classes include invalid geometry and ordering, duplicate nodes, singular or
inconsistent constraints, invalid material/load data, malformed Gmsh and
missing physical groups.
The HEX8 diagnostic covers three axial-refinement, three slenderness and
three transverse-resolution rows. Six same-mesh CalculiX 2.20/C3D8
displacement comparisons pass the existing one-percent diagnostic threshold;
the maximum full-displacement relative error is 1.997130986610937e-06.
External reactions and energy are NOT_COMPARABLE because the inherited deck
requests displacement only. QF and C3D8 agree while both deviate from the
Euler diagnostic, supporting a LOW_ORDER_LIMITATION classification with
secondary MESH_DEPENDENCE; locking remains compatible but unproven. No
HEX8R/SRI/B-bar formulation is evaluated or promoted.
027-G11 bounded J2 maturity extension¶
WP11 records PASS_WITH_LIMITATIONS for the existing small-strain J2 route,
with its Owner review closed by WP20 as
OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE. The
controlled catalog and evidence cover elastic prediction, first yield, radial
return, internal variables, unloading/reloading, simple cycling, tangent finite
differences, connected multi-element paths, energy, rollback and explicit
failure modes on TET4, TET10, HEX8 and HEX20. Full Newton is the accepted
route; modified Newton non-convergence is recorded as diagnostic.
The increment study uses a declared load path and 1/2/4 subdivisions per
branch. It records family-specific sensitivity without introducing a new
universal acceptance threshold. Algorithmic tangent symmetry is diagnostic
only, finite-kinematic J2 remains experimental/not qualified, and no new
external structural campaign is claimed. Evidence is in
qualification/0_2_7/wp11_j2_evidence.json and its source SHA is
94461602dfd1782be57c20e1801a0d5d8e262ef1.
LUP-027-G20 residual J2 and external V&V closure¶
WP20 is PASS_WITH_LIMITATIONS with Owner decision
OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE. It retains the existing
QUALIFIED_BOUNDED MAT-J2-SMALL scope for small-strain isotropic J2 with
radial return and full Newton on TET4, TET10, HEX8 and HEX20. No family is
promoted or demoted, and finite-kinematic J2 remains experimental/not
qualified.
The evidence closes the residual review items: return mapping, yield
detection, unload/reload, simple cycling, tangent finite differences,
increment characterization, rollback, energy, cross-family consistency,
failure modes, no NaN/Inf and deterministic replay. The tangent FD maximum is
2.120472111937634E-10 against the existing 1E-6 limit. Increment results
remain characterization only; no universal structural increment threshold or
tangent-symmetry qualification is claimed. Modified Newton non-convergence
remains diagnostic.
External V&V is PARTIAL_REUSED_CONTROLLED_EVIDENCE: the controlled
Code_Aster 18.1.0 constitutive evidence from G06 is reused for all four
families. No new structural J2 campaign is claimed, and no post-result
tolerance retuning occurred. 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 then closed its
surgical compatibility and release-truth scope; WP22 is closed through the
final Owner release checkpoint.
Level-Up 2 gate namespace¶
The prior Level-Up is closed as Level-Up 1 at 50/50. The active namespace is
027-LEVEL-UP-2, based on the qualified snapshot
8f08bfb5a6d4dedcd24966f5474e8c12cbfa5bc3. The pre-closure LU2 checkpoint
was at 46/50 after the WP01
observatory, bounded WP02 configuration freeze, completed WP03 Gold Compute
and LU2-WP06 execution/recovery, LU2-WP07 route maturity audit and LU2-WP08
scope closeout, with the 5M Bronze and Silver gates now closed, and the global
program was therefore 96/100; the current Owner-complete program is 50/50
and 100/100.
The active
contracts and state are
authoritative in qualification/0_2_7/level_up_2_plan.json and
qualification/0_2_7/level_up_2_state.json.
| Gate | Work package | Weight | Priority | Current status | PASS boundary |
|---|---|---|---|---|---|
LU2-027-G01 |
LU2-WP01 observatory | 4% | MUST | PASS |
phase-separated, digest-backed performance evidence |
LU2-027-G02 |
LU2-WP02 configuration freeze | 9% | MUST | PASS_WITH_LIMITATIONS |
explicit CPU/MPI/PETSc/KSP/PC and fail-closed backend |
LU2-027-G03 |
LU2-WP03 3M Gold Compute | 9% | MUST | PASS_WITH_LIMITATIONS |
distinct 3M FEM workload, two replays, bounded claim |
LU2-027-G04 |
LU2-WP04 5M Bronze | 5% | MUST | PASS |
real 5M construction, global readiness, finalization and evidence |
LU2-027-G05 |
LU2-WP05 5M Silver | 9% | MUST | PASS |
complete two-replay solve under frozen MPI and resource criteria |
LU2-027-G06 |
LU2-WP06 execution/recovery | 4% | MUST | PASS_WITH_LIMITATIONS |
explicit, fail-closed outcomes and state integrity; recovery bounded to existing checkpoint routes |
LU2-027-G07 |
LU2-WP07 existing routes | 4% | SHOULD | PASS_WITH_LIMITATIONS |
existing route maturity retained at bounded combination scope; no transitive promotion |
LU2-027-G08 |
LU2-WP08 route decisions | 2% | SHOULD | PASS_WITH_LIMITATIONS |
deferred/experimental routes remain unqualified; existing HEX8 remains bounded |
LU2-027-G09 |
LU2-WP09 final qualification | 4% | MUST | PASS_WITH_LIMITATIONS |
explicit Owner RELEASE or NEW_LEVEL_UP decision |
LU2-WP02 is closed with limitations by the controlled 3M MPI evidence index
and configuration freeze. The route covers 2, 4 and 8 ranks on one recorded
Docker host, with GAMG selected over the characterized HYPRE and graph
partition alternatives. The legacy runner does not expose separate preflight,
redistribution, communication or I/O timings; these fields remain explicitly
unmeasured and are never inferred. See
docs/verification/0_2_7/0_2_7_wp02_configuration_freeze.md and
qualification/0_2_7/wp02_state.json.
LU2-WP03 is closed as PASS_WITH_LIMITATIONS with 3M_GOLD_COMPUTE = PASS.
The existing WP18 Silver case remains the controlled Workload A, while a
materially distinct 3,000,000-DOF structured TET4 workload completed two
replays under the unchanged freeze. The records are
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; exact metrics
and the bounded claim are documented in
docs/verification/0_2_7/0_2_7_wp03_3m_gold_compute.md. Preflight,
redistribution, communication and I/O remain explicitly NOT_MEASURED, and
the evidence makes no universal, multi-node, GPU, mixed-mesh, nonlinear or
restart claim. LU2-WP04 was attempted with a real 5,012,640-DOF TET4
workload. The two independent constructions and preflight passed, but the
owner-interrupted container remained CPU-active in frozen AIJ assembly without
a completion record. That first attempt is preserved as historical forensic
evidence. The corrected Bronze run completed with global readiness and the
subsequent Silver gate completed two replays under the unchanged freeze. C1 is
not confirmed and LU2-WP09 is closed through the final release qualification
checkpoint.
Conditional gates C1 (matrix-free TET4 capacity), C2 (GPU foundation) and
C3 (10M capacity exploration) are installed with zero weight. C1 remains
DORMANT because the 2x time budget overrun was observed during an
owner-interrupted run without persisted progress/resource telemetry; S1 now
installs optional telemetry for the supervised retry; C2 requires real
hardware evidence, and C3 requires 5M Silver first; none can
block LU2.
Bronze never
creates a 3M solve claim, Gold does not require restart, and no post-result
retuning or silent fallback is permitted.
LU2-027-G08 route-decision closeout¶
LU2-WP08 is PASS_WITH_LIMITATIONS. The controlled decision matrix is
qualification/0_2_7/lu2_wp08_decision_matrix.json
and its state record is
qualification/0_2_7/lu2_wp08_state.json.
No new element, formulation or mixed-mesh framework was implemented.
Mixed TET/WEDGE/HEX is PARTIAL at technical infrastructure level but has no
qualified end-to-end route and is deferred. WEDGE15 and PYRAMID5 are
NOT_SUPPORTED and deferred; negative fixtures do not constitute
implementation. Existing HEX8 remains SUPPORTED_WITH_LIMITATIONS within
the bounded WP19 diagnostic scope. HEX8R, SRI and B-bar are
RESEARCH_ONLY, while hourglass control is deferred with any future reduced
integration route. These decisions do not change the active capability
registry or any earlier Owner decision.
F5 packaging gate¶
F5 is PASS_WITH_LIMITATIONS. The controlled record is
qualification/0_2_7/f5_packaging_compatibility_audit.json, with the review
view at docs/verification/0_2_7/0_2_7_f5_packaging_compatibility_audit.md.
The qf-solver / qf_solver 0.2.7a0 wheel and sdist build, pass twine
check, and install from outside the repository in the directly tested
Windows Python 3.12/3.13 environments. A P1 missing-runtime-resource defect
was fixed and guarded by packaging tests. The core API and CLI remain usable
without optional PETSc/MPI/HDF5 dependencies; unavailable optional routes
fail with explicit infrastructure diagnostics.
The current limitation is compatibility evidence, not a hidden release claim:
Linux is CI-declared rather than locally exercised, Python 3.10 is not locally
available, Python 3.11 and macOS are unverified, and release-vv is a
controlled legacy evidence tool rather than a 0.2.7 qualification result.
No tag, publication, numerical change or maturity promotion occurred.
F6 numerical and performance regression audit¶
F6 is PASS_WITH_LIMITATIONS. The controlled record is
qualification/0_2_7/f6_numerical_performance_regression_audit.json and the
review document is
docs/verification/0_2_7/0_2_7_f6_numerical_performance_regression_audit.md.
The current release candidate retains zero P0 and zero P1 regression findings.
Targeted numerical, route, MPI/PETSc and performance sanity checks passed.
The full suite result is 2147 passed, 3 failed, 184 skipped, 2 warnings;
the three failures reproduce F4 and remain outside the officially supported
bounded release matrix.
The active 1M, 3M and 5M replay evidence and bounded C3 10M record remain applicable. No numerical source, tolerance, active baseline or maturity changed, so no heavy replay was required. F6 is a technical pre-R0 gate only; it makes no publication or release decision and does not start R0.
R0 release readiness¶
R0 is PASS_WITH_LIMITATIONS. The controlled record is
qualification/0_2_7/r0_release_readiness.json and the review document is
0_2_7_r0_release_readiness.md. The release tree is clean after a foreign
0.2.1 documentation edit was preserved in targeted stash
a9e64d5e03ebadfcd1292ad3c5b04460761c10aa; no foreign work is included in
the release candidate.
The closeout checks pass, F1-F6 remain closed, the active qualification score
historical R0 accounting is 96/100; current Owner-complete accounting is
100/100, and no numerical source, active baseline, historical evidence or
maturity changed. The 5M Silver evidence remains applicable and the 10M C3
evidence remains bounded. Wheel/sdist artifacts pass distribution checks and
F5 clean-install evidence remains valid.
The generic publication readiness script intentionally remains NOT_READY
before the tag and reports a reachable-history manual-review item. R0 records
this as an Owner action, not as an undisclosed source or numerical failure.
No tag, GitHub release or PyPI publication is performed by this gate.