0.2.7 Master Plan and Level-Up Extension¶

Current release reconciliation¶

The active Owner-complete state for stable v0.2.7 is 100/100. WP22 final release qualification is PASS_WITH_LIMITATIONS and is closed through R0 and Step 1. Earlier 46/50, 96/100, WP22 PLANNED and 0.2.7a0 values below are retained only as historical planning or pre-closure checkpoints.

Objective and boundary¶

The foundation theme Prismatic Solid Interoperability and Numerical Robustness is historical scope. The active official theme is Reproducible Large-Model Solving and Numerical Trust. The Level-Up extension targets reproducible large-model execution and numerical trust, while preserving the controlled WEDGE6 vertical slice and stronger cross-family contracts and better evidence. It does not authorize new physics by itself. WEDGE15, PYRAMID5, TL HEX8, refined Arc-Length, finite-kinematic J2, friction/finite sliding/mortar and HEX8R qualification are outside the qualified 0.2.7 scope unless a future Owner decision explicitly changes it.

The plan starts from the tagged 0.2.6 baseline e839373b6aef291a93292186d7553ba5cd12af55. WP01 release truth and provenance is PASS; WP02 registry control is PASS; WP03 descriptor and preflight control is PASS; WP04 harness control is PASS; WP05 external oracle preflight is PASS for deck validation only; and WP06 mesh-quality contract control is PASS. WP07 is now PASS for the experimental elemental kernel; WP08 and WP09 are complete with their recorded bounded scope, WP10 has completed its independent bounded modal qualification for the declared consistent-mass route, WP11 records PASS_WITH_LIMITATIONS with its Owner review closed by WP20, and WP12 retains its bounded readiness status. The foundation WP01-WP12 history is preserved; the Level-Up portfolio is CLOSED / ACCEPT_WITH_CONSOLIDATION; WP13 is now PASS on its controlled golden-baseline evidence and WP14 is PASS on its frozen execution contract; WP15 is PASS_WITH_LIMITATIONS, WP16 is PASS after the official PETSc retry, WP17 is PASS_WITH_LIMITATIONS, WP18 is PASS_WITH_LIMITATIONS after the Bronze/Silver ladder, 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 and release-truth cleanup; WP22 is PASS_WITH_LIMITATIONS through the completed R0 release-qualification checkpoint.

Work packages and STOP/GO criteria¶

WP01 - Release truth and provenance¶

Objective: establish one authoritative baseline, version, branch and evidence-SHA vocabulary for 0.2.7.
Dependencies: none.
GO: exact baseline, clean-state record, package/tag separation and reproducible manifest rules are reviewed.
STOP: SHA/version/tag ambiguity, dirty baseline or historical evidence silently reused as current evidence.
Evidence required: baseline manifest, release-state audit and provenance checklist.
Files: qualification/0_2_7/manifest.json, release notes and provenance records.
Status: PASS (release-truth foundation only).

WP02 - Capability registry v2¶

Objective: represent capability maturity at element x analysis x material x route granularity.
Dependencies: WP01.
GO: unique IDs, source sentinels, evidence links, exclusions and machine-readable status validation pass.
STOP: maturity inferred from implementation, duplicate rows, orphan implementation or missing public combination.
Evidence required: registry schema, migration mapping and audit output.
Files: qualification/0_2_7/requirements.json, registry migration records and coverage matrix.
Status: PASS for the registry control; no numerical capability was promoted.

WP03 - Element descriptors and compatibility preflight¶

Objective: describe element topology, DOFs, faces, quadrature, loads and supported analysis/material routes before execution.
Dependencies: WP02.
GO: deterministic preflight accepts supported combinations and rejects unsupported ones without changing default numerical behavior.
STOP: implicit fallback, misleading support claim, public API break or compatibility decision hidden in runner code.
Evidence required: descriptor schema, accept/reject table and negative tests.
Files: descriptor contract, preflight evidence and targeted tests.
Status: PASS for the technical descriptor and preflight control; no numerical capability was promoted.

WP04 - Declarative V&V harness¶

Objective: add a small declarative runner for new cases without a massive refactor of the existing verification modules.
Dependencies: WP01 and WP02.
GO: case schema, oracle adapters, manifest/digest output and deterministic replay pass on a small fixture corpus.
STOP: arbitrary command execution, changed solver route, hidden defaults or loss of old evidence.
Evidence required: schema tests, fixture replay and migration boundary.
Files: additive harness modules, case definitions and tests.
Status: PASS for the additive harness contract and representative migration samples; no numerical source was changed.

WP05 - External oracle preflight¶

Objective: establish whether C3D6/PENTA6 comparisons are actually available and comparable before WEDGE6 implementation.
Dependencies: WP01 and WP02.
GO: pinned tool/version, node order, quadrature, face convention, observable mapping and reproducible deck contract are recorded.
STOP: unavailable tool represented as PASS, incompatible kinematics or untransformable observables.
Evidence required: preflight report, deck templates and availability classification.
Files: 0_2_7_external_oracle_plan.md, the WEDGE6 external review pack, deck metadata and manifests.
Status: PASS for local-only deck preflight; no WEDGE6 implementation or QF/external correlation is claimed.

WP06 - Mesh quality and distortion contract¶

Objective: define quality, Jacobian, orientation, distortion and conditioning diagnostics that apply across element families.
Dependencies: WP02 and WP03.
GO: metrics, invalid cases, reproducible sampling and Owner policy exist.
STOP: universal aspect-ratio cutoff, mesh quality inferred from one scalar or failed cases hidden by a solver retry.
Evidence required: quality schema, controlled mesh set and failure tests.
Files: 0_2_7_mesh_quality_plan.md, metric definitions and reports.
Status: PASS for the additive diagnostic contract; no universal quality threshold or new numerical capability was introduced.

WP07 - WEDGE6 kernel and elemental V&V¶

Objective: implement and verify the reviewed WEDGE6 formulation within the elemental small-strain elastic scope.
Dependencies: WP03, WP05 and WP06.
GO: the reviewed shape functions, integration and geometry certificate are implemented; elemental invariants, stiffness/recovery and deterministic evidence pass while public maturity remains EXPERIMENTAL.
STOP: certified inversion is not rejected, rank or rigid-body behavior is unexpected, or an existing formulation is modified.
Evidence required: implementation, targeted V&V catalog/evidence and experimental maturity decision.
Files: WEDGE6 kernel, formulation contract, quality/preflight integration, targeted tests and wp07_state.json.
Status: PASS (EXPERIMENTAL elemental kernel only).

WP08 - WEDGE6 static vertical slice¶

Objective: add only the smallest static path needed to exercise WEDGE6 import, faces, loads, BCs, reactions and stress/post-processing.
Dependencies: WP07.
GO: static patch/constant-strain, equilibrium, Gmsh, face loads, error handling and maintained example pass.
STOP: incomplete face mapping, silent unsupported input, source-tree-only example or result without provenance.
Evidence required: case manifest, analytical checks, mesh levels and reproducible smoke.
Files: WEDGE6 implementation and its focused tests, only after WP07.
Status: PASS_WITH_LIMITATIONS for independent experimental evidence.

WP09 - WEDGE6 robustness and external V&V¶

Objective: test distortion, orientation, quality, load paths, failure modes and comparable C3D6/PENTA6 results.
Dependencies: WP08.
GO: declared mesh/refinement/quality matrix and independent correlation support the proposed bounded static claim.
STOP: unexplained divergence, threshold weakening or non-comparable external result presented as validation.
Evidence required: internal and external manifests, negative cases and Owner classification.
Files: WEDGE6 campaign records and external correlation pack.
Status: PASS_WITH_LIMITATIONS; Owner review is required and the existing bounded small-strain J2 scope is retained.

Objective: qualify only the modal route that has its own mass, boundary, residual, mode-shape and mesh evidence.
Dependencies: WP08 and WP09.
GO: consistent-mass modal invariants, first-three-mode refinement, deterministic replay and compatible external frequency/MAC evidence are recorded under a bounded Owner policy.
STOP: transferring static maturity to dynamics or hiding rigid modes and mass-convention assumptions.
Evidence required: mass policy, eigenpair residuals, mode checks, first-three-mode refinement, frequency/MAC matching and replay.
Files: modal contract, final cases, final evidence and Owner state.
Status: PASS_WITH_LIMITATIONS; modal maturity is QUALIFIED_BOUNDED only for the declared consistent-mass WEDGE6 scope.

WP11 - Existing capability maturity and J2 gaps¶

Objective: close high-value V&V gaps in existing TET4/TET10/HEX8/HEX20 small-strain J2 without changing the formulation.
Dependencies: WP01, WP04 and WP06.
GO: increment sensitivity, tangent checks, failure handling and cross-family evidence are complete or explicitly bounded.
STOP: finite-kinematic promotion, weakened tolerance or formulation change disguised as V&V work.
Evidence required: predeclared policies, all-family records and Owner decision.
Files: J2 gap-closure contract and evidence manifests.
Status: PASS_WITH_LIMITATIONS; Owner closure is recorded by WP20 and the existing bounded J2 qualification is retained.

WP12 - Large-scale and 1M-DOF readiness¶

Objective: characterize model creation, assembly, solve and memory boundaries without promising universal 1M-DOF support.
Dependencies: WP02, WP04 and WP06.
GO: 100k/300k/500k/1M measurements have hardware, topology and resource context; resource limitation is an explicit result.
STOP: missing resource data, incomparable runs or an unqualified scalability claim.
Evidence required: timing/RAM/nnz/residual manifests and verdict matrix.
Files: 0_2_7_1m_dof_plan.md and benchmark outputs.
Status: PASS_WITH_LIMITATIONS; bounded evidence recorded, Owner review pending.

Historical foundation WP13 - Research and stretch selection¶

Objective: select only high-value research candidates after the core contracts are stable.
Dependencies: applicable closed prerequisites from WP06-WP12.
GO: explicit Owner selection with a small experiment and a bounded classification.
STOP: WEDGE15/PYRAMID5/TL HEX8/finite J2/coupled physics or HEX8R being promoted by adjacency or test count.
Evidence required: choice record, risk review and non-qualification statement where appropriate.
Files: 0_2_7_owner_decision_log.md, research evidence only.
Status: NOT_STARTED.

Historical foundation WP14 - Documentation and release closeout¶

Objective: reconcile registry, docs, package, reproducibility, full regression and Owner release decision.
Dependencies: WP01-WP13 as applicable to the final scope.
GO: clean source, package/install checks, T3 results, safe claims and signed Owner closeout.
STOP: stale claims, missing manifests, numerical regression, package drift or unresolved public capability.
Evidence required: release criteria, final test matrix, archive audit and Owner decision.
Files: release notes, final manifests and public documentation.
Status: NOT_STARTED.

Order rule¶

The architecture and evidence contracts preceded the WP07 numerical proof. WP07 is complete only for the experimental elemental kernel; WP08 must add the static/import/load/post workflow before any user-facing WEDGE6 claim. Stretch work is optional and cannot delay a bounded core unless it changes a public claim.

Level-Up 1 historical extension¶

The active official theme is Reproducible Large-Model Solving and Numerical Trust. The portfolio decision is CLOSED / ACCEPT_WITH_CONSOLIDATION for the scope extension only. WP01-WP12 and their evidence remain preserved; WP13 and WP14 are complete on their controlled records; WP15 is PASS_WITH_LIMITATIONS on its controlled subscale evidence, WP16 is PASS on the official PETSc retry, WP17 is PASS_WITH_LIMITATIONS, WP18 is PASS_WITH_LIMITATIONS after its Bronze/Silver evidence, WP19 is PASS_WITH_LIMITATIONS on bounded adversarial and HEX8 diagnostic evidence, and WP20 is PASS_WITH_LIMITATIONS for the existing bounded J2 scope. WP21 is PASS_WITH_LIMITATIONS after its surgical cleanup evidence; WP22 is PASS_WITH_LIMITATIONS through the completed final release evidence.

WP	Title	Weight	Priority	Status
WP13	Release truth and golden numerical baseline	4%	MUST	`PASS`
WP14	Large-scale execution contract	5%	MUST	`PASS`
WP15	Matrix-Free TET4 V2 / SPD / preconditioning	10%	MUST	`PASS_WITH_LIMITATIONS`
WP16	True 1M DOF qualification	10%	MUST / release blocker	`PASS`
WP17	PETSc/MPI + large sparse path	5%	SHOULD	`PASS_WITH_LIMITATIONS`
WP18	3M DOF ladder Bronze/Silver/Gold	7%	MUST / mandatory	`PASS_WITH_LIMITATIONS`
WP19	Adversarial robustness + HEX8 diagnostic	5%	MUST	`PASS_WITH_LIMITATIONS`
WP20	Residual J2 / external V&V closure	3%	SHOULD	`PASS_WITH_LIMITATIONS`
WP21	Architecture/API/registry surgical cleanup	3%	SHOULD	`PASS_WITH_LIMITATIONS`
WP22	Final Release Qualification	3%	MUST	`PASS_WITH_LIMITATIONS`

Machine-readable criteria are in qualification/0_2_7/level_up_plan.json. WP16 required a true 1M FEM iterative solve with reactions, residual, equilibrium, energy, subscale comparison and two replays. WP18 separates Bronze model/resource preflight, Silver full solve and Gold distributed/restart evidence; Bronze alone authorizes no 3M solve claim. WP18 Silver completed two 3M true-DOF solves on the declared structured TET4 PETSc/MPI route. Gold remains NOT_ATTEMPTED because no restart/checkpoint or distinct second physical case was run; no Gold claim is made.

The following remain 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.

WP19 - Adversarial robustness and HEX8 diagnostic¶

WP19 is PASS_WITH_LIMITATIONS for the 24-case adversarial catalog and the bounded HEX8 diagnostic. The execution source SHA is dc5975b78727d9dca6d0a48b716e60f355b8799f; the lot started at 7f7ffbaf0b3fdda7d3ad31ba95f20a54e4719a53. Ten positive cases passed and 14 predeclared failure cases returned EXPECTED_FAILURE_PASS; replay was deterministic, all failure paths were fail-closed, and no NaN/Inf result was accepted.

The HEX8 study 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, with maximum full-displacement relative error 1.997130986610937e-06. The inherited deck does not request reaction or energy outputs, so those external observables are NOT_COMPARABLE. Agreement between QF and C3D8 while both deviate from the Euler diagnostic supports LOW_ORDER_LIMITATION with secondary MESH_DEPENDENCE; locking is compatible with the observation but not proven. No HEX8R/SRI/B-bar formulation is evaluated or promoted, and no QF-specific bug was found.

The authoritative records are qualification/0_2_7/wp19_state.json, qualification/0_2_7/wp19_cases.json, qualification/0_2_7/wp19_runtime/wp19_robustness_summary.json, qualification/0_2_7/wp19_runtime/wp19_hex8_diagnostic.json and qualification/0_2_7/wp19_runtime/wp19_golden_replay.json. WP20 is closed with its existing bounded J2 scope; previous maturity decisions remain unchanged. WP21 is closed with PASS_WITH_LIMITATIONS, and WP22 is closed through R0 with PASS_WITH_LIMITATIONS.

WP20 - Residual J2 / external V&V closure¶

WP20 is PASS_WITH_LIMITATIONS with Owner decision OWNER_APPROVED_BOUNDED_KEEP_EXISTING_SCOPE. It closes the residual review of the existing small-strain J2 route without changing the FEM formulation or promoting a capability. TET4, TET10, HEX8 and HEX20 remain qualified only in the existing bounded scope: isotropic small-strain J2, radial return and full Newton for nonlinear static use.

The controlled evidence records passing 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. There is no universal structural increment threshold; tangent symmetry and modified Newton behavior remain diagnostics.

The external part 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. Finite-kinematic J2 remains experimental/not qualified. See qualification/0_2_7/wp20_state.json and docs/verification/0_2_7/0_2_7_wp20_j2_closeout.md.

Level-Up 2: active large-model program¶

The previous Level-Up portfolio is now treated as Level-Up 1: its qualification evidence is preserved and its program block is 50/50 CLOSED. The active namespace is 027-LEVEL-UP-2, starting from the qualified baseline 8f08bfb5a6d4dedcd24966f5474e8c12cbfa5bc3. The pre-closure LU2 checkpoint acquired 46/50 through the completed WP01 observatory, bounded WP02 configuration freeze, WP03 Gold Compute, LU2-WP06 execution/recovery closeout, the LU2-WP07 route maturity audit and the LU2-WP08 decision closeout, with LU2-WP04 Bronze and LU2-WP05 Silver now closed, so the pre-closure global progress was 96/100; the current Owner-complete progress is 50/50 and 100/100. These values replace the older non-additive accounting view for active planning without rewriting any historical result.

The machine-readable source of truth is qualification/0_2_7/level_up_2_plan.json, with state and navigation records in level_up_2_state.json and level_up_2_index.json.

LU2 work package	Focus	Weight	Priority	Current status
LU2-WP01	Evidence and Performance Observatory	4%	MUST	`PASS`
LU2-WP02	CPU/MPI/GAMG readiness and configuration freeze	9%	MUST	`PASS_WITH_LIMITATIONS`
LU2-WP03	3M Gold Compute	9%	MUST	`PASS_WITH_LIMITATIONS`
LU2-WP04	5M Bronze	5%	MUST	`PASS`
LU2-WP05	5M Silver	9%	MUST	`PASS`
LU2-WP06	Execution Contract / Recovery / Diagnostics	4%	MUST	`PASS_WITH_LIMITATIONS`
LU2-WP07	Existing Routes Maturity and Targeted V&V	4%	SHOULD	`PASS_WITH_LIMITATIONS`
LU2-WP08	Mixed / WEDGE15 / PYRAMID5 / HEX8 Decisions	2%	SHOULD	`PASS_WITH_LIMITATIONS`
LU2-WP09	Release Truth / Registry / CI / Final Qualification	4%	MUST	`PASS_WITH_LIMITATIONS`

The 3M Gold contract retains the existing Silver case and requires a materially distinct second 3M FEM workload with two replays. The claim is limited to structured TET4 linear-static execution on the recorded machine and MPI configuration; restart is not required for Gold Compute. The 5M Bronze contract requires two deterministic constructions, matching DOF, partition, ownership and digests, distributed operator construction and preconditioner setup under declared resource budgets, but creates no solve claim. The 5M Silver contract requires a complete two-replay solve with the frozen MPI configuration, residual/equilibrium/energy/finiteness checks and no post-result retuning. The MPI V2 contract targets 2, 4 and 8 ranks on one host, requires 3M strong-scaling measurements, and treats unavailable 8-rank execution as PASS_WITH_LIMITATIONS, never as FULL.

C1 (matrix-free TET4 capacity), C2 (GPU foundation) and C3 (10M capacity) are installed as zero-weight conditional gates. C1 remains dormant because the WP04 time-budget overrun was observed during an owner-interrupted run without persisted progress/resource telemetry; C2 requires real GPU evidence; C3 requires 5M Silver first and can never block LU2. New physics, production HEX8R/SRI/B-bar, WEDGE15, PYRAMID5, finite-kinematic J2, TL HEX8, refined Arc-Length, new nonlinear couplings, matrix-free multi-family and general Newmark/harmonic qualification remain deferred to 0.2.8+.

LU2-WP08 decision closeout¶

LU2-WP08 is closed as PASS_WITH_LIMITATIONS from the controlled decision matrix qualification/0_2_7/lu2_wp08_decision_matrix.json. Mixed TET/WEDGE/HEX is PARTIAL technically but has no qualified end-to-end contract and is deferred. WEDGE15 and PYRAMID5 are not supported or active capabilities and remain deferred. The existing HEX8 route remains bounded; WP19 is a diagnostic, not proof of a universal locking correction. HEX8R, SRI and B-bar are research-only decisions, and hourglass control is deferred with reduced integration. No implementation, large benchmark or maturity promotion occurred. The earlier owner-interrupted WP04 attempt is preserved as historical forensic evidence; the corrected WP04 Bronze and WP05 Silver gates are now closed under the unchanged freeze.

LU2-WP02 is recorded in the controlled execution index qualification/0_2_7/wp02_runtime/wp02_evidence_index.json and freeze qualification/0_2_7/wp02_runtime/wp02_config_freeze.json, under the predeclared contract qualification/0_2_7/wp02_execution_contract.json. The same 3M structured TET4 workload passed at 2, 4 and 8 MPI ranks on the recorded Docker host. GAMG and contiguous partitioning were selected from characterized subscale alternatives. Preflight, redistribution, communication and I/O remain explicitly unmeasured because the legacy runner does not expose separate boundaries; no phase is inferred from total time. The claim is consequently bounded to the recorded host, image, input and configuration. The machine-readable closeout is 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 is retained as Workload A, and a materially distinct 3,000,000-DOF structured TET4 workload completed two replays under the exact WP02 freeze. The contract, preflight, Observatory records, replay comparison and descriptive A/B comparison are listed in qualification/0_2_7/wp03_runtime/wp03_evidence_index.json and summarized in docs/verification/0_2_7/0_2_7_wp03_3m_gold_compute.md. The claim is limited to the recorded single-host Docker/PETSc/MPI route; preflight, redistribution, communication and I/O are explicitly not measured, and no universal, multi-node, GPU, mixed-mesh, nonlinear or restart claim is made. LU2-WP04 was then attempted on a real 5,012,640-DOF TET4 workload. Two independent model constructions and the resource preflight passed, but the owner-interrupted container remained CPU-active in frozen AIJ operator assembly without a completion record. The attempt is therefore PASS, with the earlier owner-interrupted attempt retained as historical forensic evidence. LU2-WP05 subsequently completed the two-replay 5M Silver solve under the unchanged freeze. C1 is not confirmed; the historical LU2 checkpoint was 96/100, while the current Owner-complete program is 100/100. LU2-WP09 is closed through the final release qualification checkpoint.

LU2-WP01 is recorded in qualification/0_2_7/observatory_contract.json with the controlled fixture qualification/0_2_7/wp01_observatory_sample.json. The observatory is opt-in, rejects incomplete or non-finite PASS-like evidence, and does not infer a performance regression or improvement from a comparison.

DECISION_GATE_1 = CONTINUE_TO_LEVEL_UP_2 is closed. After LU2-WP09, FINAL_DECISION_GATE permits only an explicit RELEASE or NEW_LEVEL_UP decision; there is no automatic release. The WP02 execution used the controlled 3M evidence run; the setup commit itself ran no heavy benchmark, changed no numerical source and performed no publication action.