A score of 2/4 is abstract; the part that earned it is not.
These are representative failures — welded lids, flipped axes,
non-manifold meshes, impossible cut paths — that show why a
math-graded hardware benchmark is legible, without ever leaking the
answers.
Each example pairs a render with the grade levels it failed and a plain explanation. The four-level ladder (structural · geometric · physical constraints · DFM) is the same one described in what MakerBench measures; click a task to open its detail page.
No curated examples are available yet.
How examples are chosen. Maintainers start from a candidate (model-produced) or synthetic artifact that visibly fails and capture only public diagnostics — never oracle geometry, private thresholds, or held-out fixtures. The full workflow and safety rules live in docs/FAILURE_GALLERY.md, and the curated data is site/data/failure_gallery.json.