How do I make grid labels readable for a vision model?
Draw them at the resolution the model receives, give every cell its own label, and keep the label box high-contrast.
And how small can a click target be?
A 16 pixel checkbox is common, so aim for an error under 8 pixels.