026-G08 Mesh Extension¶

Purpose¶

This page records supplemental mesh evidence for the existing 026-G08 linear-buckling gate. It does not reopen or replace the historical Owner closeout, and it does not change the official one-percent policy.

Provenance¶

Baseline checkpoint: c9d5ce8d7ce456c5d3fdcc5ff43d0fcebb2c0c4c
Extension execution source: 151662ac4781718a7fbe3d1e527675ec9e513ad4
Historical mesh evidence source: 6589443e1404a2749ac6c0a9b911f00dd9cb8753
Source worktree at execution: clean
Solver version: 0.2.6a0

The machine-readable record is qualification/0_2_6/g08_mesh_extension_evidence.json; the generated evidence report is qualification/0_2_6/g08_mesh_extension_evidence.md.

Controlled extension¶

The historical levels 1, 2, 4 and 8 are reused without recomputation. Levels 16 and 32 are executed for TET10, HEX8 and HEX20. Every added level is replayed twice. The existing classification is retained:

<=1%: CONVERGED_BOUNDED
1-4%: NEAR_CONVERGED_BOUNDED for diagnostics only
>4%: NOT_STABILIZED

Family	Level 16 factor	Level 32 factor	Direct change	Classification
TET10	9.4770217961	9.4847469311	0.081448%	`CONVERGED_BOUNDED`
HEX8	139.0005235350	138.7686228488	0.167113%	`CONVERGED_BOUNDED`
HEX20	8.4096763077	8.3336219585	0.912621%	`CONVERGED_BOUNDED`

All 12 extension observations passed, including deterministic replays. The maximum eigenpair residual remained below the existing 1e-7 PASS policy for the added rows. No solver, formulation or eigensolver code was changed.

HEX20 interpretation¶

HEX20 reaches the direct-change diagnostic band in this two-level extension, but its historical coarse-to-fine series remains strongly mesh-sensitive. This is bounded supplemental evidence, not a universal convergence claim and not a retrospective change to the existing Owner decision.

The retry of the same one-cell C3D20-style external deck remains BLOCKED_EXTERNAL_TOOL. Code_Aster is not comparable for this route, and no independent high-order analytical buckling oracle was identified in the controlled repository evidence.

One exploratory inline run before the controlled harness observed an ARPACK non-convergence at HEX20 level 32. The controlled two-replay harness did not reproduce it; the observation is retained in the JSON record and is excluded from acceptance decisions rather than silently discarded.

Gate impact¶

026-G08 remains PASS_WITH_LIMITATIONS. The historical family decisions remain unchanged: TET4 is QUALIFIED_BOUNDED, TET10 and HEX8 are PASS_WITH_LIMITATIONS, and HEX20 is MORE_EVIDENCE_REQUIRED. No post- buckling, multi-mode, or general physical-validation claim is added.