Skip to content

0.2.7 V&V Strategy

Evidence vocabulary

The campaign uses the following non-transitive states:

State Meaning
SUPPORTED An implementation route accepts the combination.
TESTED At least one declared case was executed.
VERIFIED An invariant, analytical oracle or independent reference was checked.
QUALIFIED_BOUNDED An Owner decision accepts a declared scope and exclusions.
EXPERIMENTAL The route has exploratory evidence but no qualified general claim.
NOT_QUALIFIED The route is explicitly excluded from qualified claims.
DEFERRED The decision or evidence is intentionally postponed.

No state is inferred from code presence, test count or a neighboring element.

Evidence layers

  1. Contract: machine-readable requirement, capability row, oracle and policy exist before execution.
  2. Implementation: route, element descriptor, loads, faces and post- processing behavior are tested in isolation.
  3. Verification: analytical/invariant, mesh, distortion, replay and failure-mode evidence are recorded.
  4. External correlation: a comparable independent deck is used when one exists; incompatible or unavailable tools are explicit skips.
  5. Owner decision: the accepted maturity, bounded domain and limitations are recorded separately from raw results.

WP04 provides an additive declarative contract for the case, oracle, execution verdict and evidence record. It does not replace the historical V&V modules; three representative legacy-route fixtures establish the migration boundary.

Capability axes

The registry target is:

element x analysis x material x route x mesh/load convention

An element descriptor does not authorize every analysis. A WEDGE6 static result cannot qualify WEDGE6 modal, Newmark, harmonic, J2 or geometric nonlinearity without their own evidence. Existing TET, HEX, beam, shell and discrete decisions remain independent.

Required evidence families

  • patch and constant-strain behavior where applicable;
  • positive Jacobian, orientation, volume and mesh-quality checks;
  • rigid-body and equilibrium invariants;
  • analytical oracle with declared assumptions;
  • mesh and, where relevant, time/increment/frequency refinement;
  • cross-element comparison only on physically comparable models;
  • failure and adversarial cases with fail-closed behavior;
  • deterministic replay with source and artifact digests;
  • external correlation with formulation/observable compatibility recorded;
  • resource and performance measurements on declared hardware.

WP10 applies these layers independently to WEDGE6 modal evidence. Consistent mass, finite-positive eigenpairs, normalized mode diagnostics, residual, refinement, frequency matching, MAC and replay are separate from the static WP07-WP09 claims. The bounded Code_Aster comparison uses same-mesh physical mode components and does not transfer static maturity or qualify Newmark, harmonic, nonlinear or general dynamic routes.

Scope discipline

The 0.2.7 core proposal does not promote TL HEX8, refined Arc-Length, finite-kinematic J2, friction/finite sliding/mortar, WEDGE15, PYRAMID5 or HEX8R. Stretch results may be useful research evidence but remain outside the bounded release claim unless a separate Owner decision says otherwise.