One at a time

The mistake

A virtual waiting room is placed in front of the shop to cap concurrency, but the admission limit is set to a single customer at a time rather than to a rate the backend can actually sustain. Because throughput is the limit divided by the average session length, one seat at roughly four minutes per visit caps the whole shop at about fifteen customers an hour, so the queue grows by thousands while the servers sit idle. The system never fails, which hides the fact that the protective limit is now the bottleneck — availability without any usable capacity.