DatumGuard v0.3.0 — FrameGuard Verified Screening

DatumGuard v0.3.0 adds FrameGuard, a fail-closed structural frame screening workflow that connects engineering requirements, explicit CAD provenance, deterministic analysis, independent serialized-DXF verification, and uncertainty-gated research inference. It also publishes the OpenBIM research workspace while keeping its hosted execution disabled until the external validation and licensing gates are complete.

What changed

• Added /frame with verified and missing-brace presets, exact-result evidence, independent DXF remeasurement, and gated downloads.
• Added a deterministic NumPy 2D Euler-Bernoulli frame solver with displacement, reaction, and member-force screening results.
• Added Rhino 8 / Grasshopper exchange adapters that require explicit units and datum; no coordinate guess is accepted.
• Added R2013/mm DXF generation with contract/entity XDATA and a separate ezdxf reopen-and-remeasure verifier at a 0.001 mm comparison grid.
• Kept official CAD output fail-closed: exact screening failure or DXF verification failure leaves dxf_base64 unavailable.
• Added genuine OpenSeesPy parity evidence (6/6 cases) and a real PyG topology-holdout benchmark (90 cases). Surrogate output is non-authoritative and uncertainty/OOD cases become REVIEW_REQUIRED.
• Published /openbim as a research UI with frozen IFC4/IDS/BCF evidence. Production Render keeps OpenBIM/BCF execution disabled.
• Updated the public Case Study, privacy boundary, operations handoff, deployment smoke, and rollback documentation.

Assurance model

Contract → exact screening → serialized CAD → independent reopen → remeasurement → verified-only download

A successful CAD command is not accuracy evidence. FrameGuard is an early screening aid and never issues structural safety approval or code certification.

Immutable release evidence

• Source/tag: v0.3.0 → 71b4aeb2e56bb910a4848b21345343602c5099a3
• PR #20 — OpenBIM Evidence Guard
• PR #22 — FrameGuard CAD and structural AI assurance
• PR #23 — v0.3.0 production evidence
• Main CI 29196188166: 376 pytest passed, 6 optional-runtime skipped, 35 real-API Playwright passed, 15-page build, OpenBIM interoperability, both container builds, CycloneDX SBOM, fixed-critical scans
• Security 29196188182: dependency audit and CodeQL Python + JavaScript/TypeScript passed
• Frame research 29194964854: genuine OpenSeesPy parity and PyG training/portable-inference smoke
• Render-success strict smoke 29196357533: exact release SHA, route sentinels, capabilities, Architecture/FrameGuard/Artifact Lab canaries, Solid fail-closed, CORS
• Vercel Production deployment: 5413292136
• Render API deployment: 5413286383 / dep-d99q796k1jcs73fglel0
• Public health: version=0.3.0, release_sha=71b4aeb2e56bb910a4848b21345343602c5099a3

Attached SBOMs

• datumguard-api.cdx.json — CycloneDX, SHA-256 ba537b1b10380df511ebed9e116214dcad31f1b422277d0f1222180fb86c992a
• datumguard-web.cdx.json — CycloneDX, SHA-256 a9f48ccecc1803170e34aa358e9f2a22e84ea5f54b8b42128232270e229dfca9

Explicit limits

• This release screens a limited, synthetic 2D linear-elastic pipe-rack family. It does not certify structural safety, nonlinear/buckling/seismic/fatigue behavior, materials, welds, connections, foundations, building codes, or manufacturability.
• The PyG model is research-only and never overrides the exact solver or CAD verifier.
• The live Rhino/Grasshopper/Cordyceps smoke was unavailable for this release because those desktop applications were not running; neutral exchange scripts, unit/datum validation, tests, CI, and the hosted path were verified independently.
• Production keeps DATUMGUARD_ENABLE_OPENBIM=false, DATUMGUARD_ENABLE_BCF=false, and Solid execution disabled. /openbim and /solid expose UI/schema/evidence, not hosted approval capability.
• External graphical IFC viewer validation, final third-party license review, real-site generalization, external DXF upload entity allowlisting, and the planned 100+50 benchmark remain follow-up gates.

Rollback

v0.2.1 remains the documented full-stack rollback baseline. Rolling back removes FrameGuard and the OpenBIM public research route; incident communication must state that capability change explicitly.

DatumGuard v0.2.1 — Verified Deployment Evidence

DatumGuard v0.2.1 is the self-consistent public Engineering Design Assurance patch release. It corrects the displayed evidence and proves the deployed API revision after Render finishes, without moving the existing v0.2.0 tag.

What changed

• Corrected the Product Case Study from the historical 19-test value to the verified 256 pytest + 24 Playwright baseline.
• Removed the duplicate Privacy title and aligned every public route's canonical, Open Graph, and Twitter metadata.
• Bumped the API and web packages to 0.2.1.
• Added a validated release_sha to API health/readiness; only a 40-character hexadecimal revision is exposed.
• Split deployment verification into a Vercel compatibility smoke and a Render-success strict smoke, avoiding the checksPass circular dependency.
• Corrected the security evidence: push runs cover pip-audit and both CodeQL languages; dependency review is evidenced by the pull-request run.
• Kept the planned 100 golden-contract + 50 language-input benchmark explicitly incomplete.

Assurance model

Contract → serialized DXF/STEP → independent reopen → remeasurement → verified-only export

Writer memory is not approval evidence. Required verification failures keep the official bundle unavailable.

Immutable release evidence

• Source: v0.2.1 → 0e871c4a93872415f7151a99050b102f273986dd
• PR #14 — Prepare the self-consistent v0.2.1 release
• Main CI — 256 pytest, 24 Playwright, type/lint/build, both container builds, SBOM, fixed-critical scans
• Push security — pip-audit and CodeQL for Python and JavaScript/TypeScript
• PR security — dependency review plus the same audit/CodeQL gates
• Vercel compatibility smoke
• Render-success strict smoke: version=0.2.1, release_sha=0e871c4a93872415f7151a99050b102f273986dd, Architecture and Artifact canaries pass, Solid fails closed, CORS passes
• Vercel Production deployment: 5410294611
• Render API deployment: 5410321107

Attached SBOMs

• datumguard-api.cdx.json — CycloneDX 1.6, SHA-256 ed49a6515d5fd4c4ab8bd324e66701d05b24cf6df7295c1f24ae3bef656e9c06
• datumguard-web.cdx.json — CycloneDX 1.6, SHA-256 8bc6f7c0cdb05dd79ad5fa0fe5d4d3386bf1c0cbca689cb418e55132ac2dd9c4

Explicit limits

• The public Render Free API keeps Solid execution disabled and returns 503 DG_CAPABILITY_DISABLED; verified local/CI STEP 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+50 benchmark is not complete.
• release_sha attests the Render Git revision; signed OCI/SLSA provenance remains future supply-chain hardening.

The immutable v0.2.0 tag remains the documented fallback; v0.2.1 supersedes its public Case Study and deployment-evidence claims.

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

• #12 Publish the DatumGuard CAD assurance case study
• #13 Fix the case study social preview metadata

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.