Skip to content

User Guide Native Openfoam Inspection

github-actions[bot] edited this page Jul 10, 2026 · 1 revision

Native OpenFOAM inspection

CaeReflex 2.0.0a2 adds an isolated, read-only OpenFOAM reader for deep and forensic profiles.

caereflex inspect CASE_ROOT \
  --adapter openfoam \
  --profile deep \
  --out case.json \
  --agent-context agent_context.json

The resulting case records metadata.inspection_execution, metadata.native_openfoam, parser attempts, diagnostics and lazy array references.

Supported native evidence

For bounded ASCII cases, the reader can decode:

  • constant/polyMesh/points;
  • constant/polyMesh/faces;
  • constant/polyMesh/owner;
  • constant/polyMesh/neighbour;
  • constant/polyMesh/boundary;
  • scalar, vector, spherical-tensor, symmetric-tensor and tensor internal fields;
  • uniform and nonuniform internalField values;
  • available numerical time directories;
  • field dimensions and class metadata.

Mesh coordinates, connectivity, ownership and field values are stored as content-addressed artefacts and exposed through ArrayRef handles.

caereflex arrays describe ARRAY_ID --json
caereflex arrays sample ARRAY_ID --count 100 --json
caereflex arrays reduce ARRAY_ID --operation min --json

Fallback behaviour

Binary files, malformed counted lists, includes, code streams and unsupported grammar are not executed or guessed. CaeReflex records an ordered failed parser attempt, identifies the fallback and states what information was lost.

Relevant diagnostics include:

  • CRX-OPENFOAM-NATIVE-FALLBACK-001;
  • CRX-OPENFOAM-FIELD-FALLBACK-001.

A successful worker status can still contain field-level fallback diagnostics. Review the attempt ledger before treating a case as fully decoded.

Important limits

The native reader does not:

  • run OpenFOAM solvers or utilities;
  • expand #include or #codeStream directives;
  • load OpenFOAM shared libraries;
  • infer missing coordinate units;
  • prove topology validity;
  • validate boundary-condition suitability;
  • prove convergence, mesh independence or physical accuracy.

The current core reader targets bounded ASCII evidence. Binary and decomposed parallel cases remain future work unless an optional backend is added later.

Clone this wiki locally