A model that learned to count by shape — and what it draws when the sequence runs out.
Capitals taught a net that upper- and lowercase are the same letter in two coats. This is its sibling, pointed at a different relationship: not case, but order. It was trained to take a glyph and morph it into the next one in its sequence — 2 after 1, 3 after 2 — as a slide you can scrub.
The edges are the point. Drag a digit forward and watch it settle into its successor. Then try 9. There is no tenth digit, so it has nowhere to land — and what it reaches for instead is the whole demo.