Little’s Law. N=XR where:
N = average concurrency (# of requests present in the system)
X = average throughput
R = average response time (latency)

Removing automated sorting machines: increases R, since that part of the process takes longer. In the absence of more capacity to handle the displaced requests (e.g. hiring a bunch of people to manually sort), N will stay fixed, so X (throughput) will decrease.

Banning overtime: this reduces N (capacity), because you have fewer carrier-hours available to deliver. R (latency) is not impacted, so, you guessed it: X (throughput) will decrease.

But the offered load X (how many people are sending mail) is not decreasing, and in fact COVID is increasing it. If you don’t add capacity (N), then response time (R) will shoot up: mail delays.

People are still sending mail at a certain rate, but it is taking longer to deliver. Where’s it sitting? In queues/buffers. Backed up. A queue is capacity—places for requests (mail) to sit—with neutral throughout. Queues just add delay.

We know the system is already overloaded; it required overtime, we reduced automation and increased processing time, and we are already seeing the resultant queuing delays.

And now we expect a bunch of extra load where latency is critical (mail-in ballots). Now is the time to *add* capacity and *reduce* latency, and we’re doing the opposite.

Oh, and by the way—remember the queuing delay? The only way to drain the queues is to increase system throughput X so it is higher than the offered load. And then it still takes time to drain. The longer you wait, the more excess throughput you need.