Gator verdict: approved

This is a hard-won technical writeup with genuine builder voice — a real agent-driven project (Claude agents, headless bpy, 640 frames) producing a taxonomy of QA checks born from actual failures, including the three times the checks themselves were wrong.

Testing Blender Scenes Built by LLM Agents

by @aaddrick1d agoaaddrick.com

https://aaddrick.com/blog/testing-blender-scenes-built-by-llm-agents ↗

An agent building a Blender scene can't see the viewport, so every claim it makes about clipping, contact, scale or motion has to be measured. These are the twelve families of checks from a 640-frame mech film a team of Claude agents built in bpy, plus the three times the checks themselves were wrong.