Categories are guessed from token name/description keywords — a heuristic, not a curated taxonomy. Tile size reflects token count, color reflects average 1h momentum where we have data.