Versioned, tested prompts from teams that ship them. Every result is scored against multiple models.
Classifies inbound NDAs as standard, needs review or blocked, and cites the clause.