Measure what parsing prose costs

The two paths are identical except for the output contract. The prose path gets a sentence and has to turn it into one of six labels; the typed path asks a jevlocal Choice whose criteria are the six teams, so its answer is a label by construction. The decision that matters most is what "naming a team" means in prose (aliases, negation, teams that don't exist), because that is where the parse fails. Settling those cases on concrete sentences before you write the parser is what makes the failure modes explicit.