Generative AI can describe a part, render a part, and model a part. CanItBeMade asks the question that decides whether any of that matters: could this actually be manufactured?

Manufacturability is where AI design tools quietly fail. Image models render "injection-molded" enclosures with zero draft and sealed internal voids. 3D generators emit meshes with paper-thin walls and undercuts no mold could release. Language models write specs that skip the one tolerance that matters. None of this shows up in existing benchmarks — models are graded on how designs look, almost never on whether they could leave a factory.

CanItBeMade grades exactly that, across text, image, and 3D, against the three workhorse processes of physical products: injection molding, CNC machining, and FDM printing. The 3D track is scored purely by geometry — draft angles, wall thickness, tool access — so the headline numbers are checked by math, not by another model's opinion.

Why now

The space is moving but narrow: EngDesign (NeurIPS 2025) tests whether AI can design across nine engineering domains, and BenDFM (2026) assesses manufacturability for sheet-metal bending alone. Nothing yet covers broad, multi-process, multi-modal manufacturability — the question a founder, engineer, or factory actually asks.

How scores stay honest

• Published rules: every rubric and threshold is on the methodology sheet.
• Traceable scores: every number maps to a saved output, a rule, and a scoring record.
• Exact labeling: entries name the precise tier and month tested.
• First response only: no retries, no prompt coaching.