026-WP06-G06 Deep V&V and Corpus Expansion¶
This document is the historical replay recorded at source SHA
b9d523725ab2a6856256731e44921adf2837a1f3. The current clean-source replay
is recorded separately in 0_2_6_g06_prequalification_evidence.md.
Status¶
026-WP06-G06 = PASS_WITH_LIMITATIONS
This work package expands the executable 0.2.6 V&V corpus. READY means that
the controlled runner can execute a case; it does not mean that the associated
element, material or analysis capability is qualified. The official gate
026-G06 is the J2 maturity-extension gate. This historical package is
retained as supporting evidence; the later deep pack records the Owner's
bounded closeout decision.
Controlled provenance¶
| Item | Value |
|---|---|
| Source SHA | b9d523725ab2a6856256731e44921adf2837a1f3 |
| Source worktree | clean |
| Captured at | 2026-08-28T21:31:09.160647Z |
| Solver | qf_solver 0.2.6a0 |
| Registry digest | d2e5f587ac2d67c3c875782c8558aa7aac60cec09df9f43614319eff1b99ea59 |
| Campaign manifest SHA-256 | f8847759716bd7007c7e783338d7f0e6d4522fa6b709d39e2b2eec1cef578804 |
| Quantitative manifest SHA-256 | 4b5a7e96ac6c661edfdf3133cb6d12fb1e294b961756f1593b4973ab2bae34e1 |
| Environment | Windows AMD64, CPython 3.13.1, NumPy 2.2.6, SciPy 1.15.2 |
Corpus and execution¶
The registry contains 180 definitions: 140 READY and 40 PLANNED. The G06
batch converted 80 definitions into executable cases and ran all of them:
| Result | Count |
|---|---|
| PASS | 79 |
| EXPECTED_FAILURE | 1 |
| FAIL | 0 |
| BLOCKED | 0 |
The batch contains 20 analytical equilibrium variants across TET4, TET10,
HEX8 and HEX20; 16 structured HEX8 mesh cases at nx=1,2,4,8; 20 HEX8/HEX20
common analysis routes; and 24 robustness routes. The expected failure is an
inverted TET4 returning INVALID_ELEMENT.
Quantitative studies¶
The final clean-source replay also ran 20 independent analytical cases:
TET4, TET10, HEX8 and HEX20 each use a constrained single-free-DOF model and
an independently integrated effective stiffness. All 20 cases passed with a
maximum relative displacement error of 0.0.
The HEX8 bar series was evaluated at four levels (nx=1,2,4,8). The relative
tip-displacement error was respectively 9.000000e-2, 5.777373e-2,
4.412610e-2 and 3.999261e-2; the trend is non-increasing. The observed
successive orders were 0.6395, 0.3888 and 0.1419. This is a bounded
quantitative mesh study for this bar configuration, not a universal
convergence claim. The reference is u_tip = F L / (E A).
The machine-readable quantitative summary and graph are produced by
scripts/run_g06_quantitative.py under the ignored raw result directory
results/vnv_026_g06_quantitative_clean/; the campaign and quantitative
manifest digests are recorded separately above.
Internal verification¶
The existing independent element verifiers passed for TET10, HEX8 and HEX20.
The existing Total-Lagrangian TET4 kernel campaign also passed, including a
finite-difference tangent error of 4.295361499621963e-10. These are internal
verification results and do not constitute external physical validation.
Limits¶
- The controlled runner records executable solver outcomes and provenance; the analytical cases now also run the independent constrained free-DOF oracle described above.
- The 16 mesh cases remain executable corpus entries; the separate four-level study above is the quantitative evidence and is limited to the stated HEX8 bar configuration.
- No Code_Aster or CalculiX run was available in this local campaign.
- No new maturity promotion is made. External correlation, general mesh
convergence and official gate
026-G06remain open evidence items. - No FEM formulation, tolerance or public solver API was changed.
Checks¶
- Framework tests:
7 passed. - G06 controlled campaign:
79 PASS / 1 EXPECTED_FAILURE / 0 FAIL / 0 BLOCKED. - Ruff:
PASS. compileall:PASS.git diff --check:PASS.
Raw result files remain reproducible under results/vnv_026_g06_final_clean/g06/
and results/vnv_026_g06_quantitative_clean/; they are excluded from normal
source history. The compact evidence and manifest digests above are the
archived record.