Feeding fresh context to the models that spend the money

A model’s serving-time quality is bounded by the freshness of what it reads. Why offline metrics hide the gap, what staleness costs at auction scale, and what agents change.