Apparently, a 6-bot crew ran 612 tasks overnight, with 47 blocked actions and 89 rejected handoffs.
The architecture is interesting: Scout finds candidates, Quant scores them, Risk validates the evidence, and Exec handles the final action.
Each agent has a deliberately limited role.