DatumGuard v0.2.0 — Independent CAD Assurance
DatumGuard v0.2.0 is a public Engineering Design Assurance release for contract-first CAD generation and independent artifact remeasurement.
What ships
- Product case study: problem → method → PASS/FAIL evidence → limits
- Architecture: wall topology, openings, grids, columns, rooms, and R2013 DXF approval
- Plant / semiconductor piping: route connectivity, orthogonality, inline components, support spacing, and clearance
- Mechanical / ship plate: holes, slots, cutouts, edge distance, ligament, and overlap
- Limited 3D solid: mounting plate, angle bracket, and flange STEP geometry through isolated OpenCascade processes
- Artifact Lab: immutable DXF, STEP, and IFC audit plus revision comparison
- MCP + FastAPI: shared deterministic application services and structured evidence envelopes
Assurance model
Contract → serialized DXF/STEP → independent reopen → remeasurement → verified-only export
Writer memory is not used as verifier evidence. Official bundles remain unavailable when required measurements or constraints fail.
Release evidence
- CI: 256 pytest, 24 Playwright, type/lint/build, both container builds
- Push security: pip audit and CodeQL Python and JavaScript/TypeScript
- PR dependency review
- Production smoke: six page sentinels, API capability contract, Architecture and Artifact canaries, Solid fail-closed, CORS
- Vercel Production and Render API deployments both point to
40dd132 - Attached API and web container CycloneDX 1.6 SBOMs were generated by the release CI run
Explicit limits
- The public Render Free API keeps Solid execution disabled and returns
503 DG_CAPABILITY_DISABLED; verified local/CI evidence remains available. - Artifact Lab findings are informational and never create a fabrication approval.
- DatumGuard does not certify structural safety, code compliance, fluid/thermal/fatigue behavior, materials, welding, or manufacturability.
- The planned 100 golden-contract + 50 language-input benchmark is not yet complete.
Merged work
Evidence correction: Superseded by v0.2.1. The v0.2.0 source tag remains immutable, but its Case Study displayed the older 19-test baseline and its first Production smoke ran before the recorded Render deployment completed. v0.2.1 corrects the public count and adds a post-Render exact revision smoke.