Kernel design notes

The detailed notes on how the CUDA kernels are laid out — which kernel does what, the tolerance allowed for each stage against the NumPy reference, and why each choice of precision and memory layout was made — live beside the code they describe, in src/vidmag/cuda/DESIGN.md.

They are kept there rather than copied here on purpose. They are read while changing the kernels, they refer to the files around them, and a second copy in this site would be a second thing to keep current — which in practice means one of them would quietly stop being true.

Other internal documents

Document	What it covers
docs/dev/PLAN.md	The plan this restructure follows, its decisions and their reasons
docs/dev/packaging-notes.md	Findings recorded while making the project installable, including defects found
docs/dev/gpu-runner.md	How the graphics-processor test machine is set up
.claude/rules/development-practices.md	The development rules this project is built under