# Animated GLB Diagnostic Preview Bounded CPU animated GLB to MP4, five-frame PNG contact sheet and hash/coverage receipt. Python 3.10–3.13, no GPU or other product purchase. Create a private venv: `python -m venv .venv`, then `.venv\Scripts\python -m pip install -r requirements.txt` (on macOS/Linux use .venv/bin/python). Use that interpreter for all commands. Pinned NumPy/OpenCV dependencies are downloaded at setup only, not during rendering. Their licenses remain their own; wheels are not redistributed in this archive. Run `python examples/make_sample.py`, then `python app.py` and load examples/input.json. Choose a new .json output; neighboring .mp4 and .png files are generated. CLI: `python checker.py examples/input.json new-preview.json`. Open the MP4 or PNG with your usual viewer. Source inputs and existing outputs are never overwritten. Exactly one perspective camera and animated scene required. Embedded GLB <=16MiB; 320x180 or 640x360, 10 or 15fps, <=20s. Indexed triangles, TRS linear/step animation, skinning, morph position targets and embedded PNG/JPEG base-color textures are supported by the diagnostic engine. External buffers/images, required extensions, orthographic cameras, cubic splines and production PBR are unsupported. Every frame is marked diagnostic. This is not a production-quality Blender replacement. Input validation bounds node/accessor/instance collections before renderer allocations, rejects cycles and preflights texture headers before native decoding. Worker deadline 120s; two OpenCV threads; OpenCL off. These are resource budgets, not an OS security sandbox. Use files you are authorized to inspect in a restricted local environment. The engine independently reopens the MP4, verifies dimensions/frame count, checks motion below the changing title, and requires useful on-screen coverage. Static or poorly framed scenes fail rather than emitting a misleading passing receipt. Tests: `python -m unittest discover -s tests -v`; includes actual encode/decode and contact sheet. This proves diagnostic output, not face likeness or final art quality. Inspect output images and motion for your actual scene. Native UI acceptance separate. Dependency references: https://pypi.org/project/numpy/2.2.6/ and https://pypi.org/project/opencv-python-headless/4.12.0.88/.