0.2.7 External Oracle Preflight¶
Purpose¶
This plan determines whether an independent C3D6/PENTA6 comparison is
available before WEDGE6 implementation. WP05 is PASS for the controlled
preflight recorded in the review pack, not
evidence of a WEDGE6 result. The machine-readable contract, node/face mapping,
case catalog and deck-validation evidence are the controlled records.
Required comparability record¶
For each candidate tool and deck, record:
- solver name, version, container/image digest or executable fingerprint;
- element name, node order, local axes, shape-function convention and integration rule;
- geometry, units, material, thickness/section and mesh hash;
- BCs, load application, load sign and reference/load-factor convention;
- observable definition and sampling location;
- stress/strain measure and any conversion used;
- command, input deck, output files, exit status and artifact digests.
The comparison classification is one of PASS_EXTERNAL,
PASS_WITH_LIMITATIONS, NOT_FORMULATION_COMPATIBLE,
SKIPPED_UNAVAILABLE or FAIL_EXTERNAL. SKIPPED_UNAVAILABLE and
NOT_FORMULATION_COMPATIBLE are never PASS.
C3D6 / PENTA6 preflight¶
| Check | Required result before WP07 GO |
|---|---|
| Availability | executable/container and version are reproducible, or explicit unavailable record |
| Topology | C3D6/PENTA6 node order and face map are controlled |
| Kinematics | displacement, strain and stress measures are equivalent or transformable |
| Integration | quadrature and sampling are documented |
| Loads/faces | nodal, face and reaction conventions are identical |
| Mesh | identical mesh or a controlled physical-equivalence map exists |
| Outputs | displacement, reaction and selected internal observable are extractable |
| Replay | same deck and command reproduce the recorded output |
The current host has pinned Docker images for both tools. They are classified
AVAILABLE_LOCAL_ONLY: the affine decks were replayed locally, but neither
tool is a CI dependency and no QF WEDGE6 correlation was issued. If an image
or executable is unavailable in a future environment, retain
SKIPPED_UNAVAILABLE and keep external correlation as a limitation. Do not
install a heavyweight stack solely to turn a planned row green.
Planned observables¶
The first WEDGE6 external study should prefer displacement, total reaction and load-displacement curves. Stress recovery is included only where the external sampling and measure are comparable. Pointwise stress values at a singularity are not primary evidence.
Stop conditions¶
Stop the comparison and request Owner review if the kinematics, load path, observable or node map is ambiguous; if an external failure cannot be reproduced; or if a new tolerance would be needed after seeing the results.