The server easily carries 200 active conversations, then goes further

The server must EASILY handle at least 200 concurrent ACTIVE projects and conversations; where it already does, find the real ceiling and raise it. 200 is the first checkpoint, not a target cap (UD-S4); capacity counts every agent (SCALE-AGENTS). Scaling never sacrifices a harness function, feedback or a log (UR-03 applies to scale as to caching), and every limit is an exposed setting (UR-05). The work is the SCALE queue (the scale design’s tasks, interleaved with the RC sequence; the test-drive checkpoints gain a concurrency scenario).