I worked in this industry for many years. The railroads were always hyper-focused on increasing throughput. I left the industry about ten years ago, but back then derailments were considered the most significant risk to average network speed. At the time, average network speed was roughly correlated with profit. The rule of thumb was a 1 mph increase in average network speed was worth about $100MM in profit. That was 15 years ago.

There were a lot of systems in place to monitor rolling stock:

- Wheel Impact Detectors

- Hotbox detectors

- Acoustic bearing detectors

- Truck performance detectors

To name just a few. There were also efforts to monitor the railway infrastructure. The things I remember:

- Rail stress management (rails need to be under the right amount of stress, which of course varies with temperature)

- Top of rail friction management

- Rail profile management (the name eludes me, but the idea is you want the interface between the rail and wheel to meet certain parameters)

I worked on the rolling stock side measuring wheel impacts, overloads, imbalances, and a handful of more esoteric metrics. One of the outputs of these measurements was a train consist. For each of our locations we were able to build up the consist of the entire train (which was a fun CS problem in itself).

I stared at a lot of consists over the years. In North America I never saw anything longer than about 100 cars and 2-4 locos. However, in Northwest Australia they routinely ran 300 car trains meeting the description in this article. But, the reason they could get away with that is they were running a straight shot from the heart of the Pilbara to one of the port towns on the north west shoulder (Karratha and Port Hedland).

I need to check in with my old colleagues and see if things have changed. It wouldn't surprise me if train lengths have gotten longer, but I would be surprised if this correlated with a large increase in derailments, as that would have a tremendous impact on average network speed and thus profit.

As someone mentioned elsewhere on this thread, there are a lot of single track corridors. It's bad enough when one train has to sidetrack. It's really bad when a train takes out the whole corridor. These aren't packet switched networks. It's not easy to reroute. And it's really expensive and difficult to lay new rail.