Class recognition
The input string is split into tokens on whitespace. Each token is matched against the utility grammar: a base such as p-, m-, w-, bg-, text-, rounded-, flex, or grid, optionally followed by a scale step like 4, 500, or lg. Recognized patterns resolve to one or more declarations, with step values from built-in tables: spacing is 0.25rem times the step, and named steps pick from the size table. Tokens it does not recognize produce no declarations.