Archived benchmark note
AI-assisted CAD assembly benchmark
Three AI-assisted CAD runs attempted an approximately 100-component reference assembly under a company-authored scoring harness.
The outputs were incomplete and not build-ready; no physical machine was built. The benchmark tested component placement and constraints under a defined company-authored harness. Component count and a harness score do not establish correctness, engineering adequacy, buildability, safety, or autonomous physical construction.
This route is retained to identify the historical benchmark. Reuse of any result requires the exact benchmark version, inputs, scoring rules, output artifacts, and an independent review appropriate to the intended claim.
For current CAD validation boundaries, see CADCLAW.