You might find this interesting: https://www.theatlantic.com/technology/archive/2015/11/why-d...

It's actually partially true. The NYCT signalling system is more like a sandwich. At the bottom they have a legacy system called AWS. It is built using a logic of relays that will raise or lower mechanical arms on the tracks. When the arm is raised it will collide with a valve on the train that, once opened will apply the brakes on the train forcibly. The logic driving this is from the beginning of the 20th century and seriously pretty clever. The guys at the time even timed the relays to raise arms according to the recommended maximum speed for driving the train. On top of this, computer based systems have been deployed, especially CBTC, starting in the Canarsie line. So, frankly, the statement made is quite inaccurate. The system works in the sense that it allows higher potential capacity. The real problem is that with a legacy sandwich so deep the failure rate is terrible, especially when age of material and obsolescence kicks in. Now imagine fixing a technical debt that spans nearly 2 centuries, with limited maintenance time, and life critical functions buried inside. This is the problem.

Reminds me of an article I read on some guys that updated the controls for a sewage tunnel that runs under the Hudson river. The more they investigated the dynamics of a 1000 foot long, 10 diameter pipe full of poo, the more terrified they became. And more impressed with the long dead engineers that originally solved the problem in the 1920's.

Original controls were all mechanical.

What Adam (?) said, but now in video form: https://www.youtube.com/watch?v=Mjx3S3UjmnA

Look up mechanical signal interlocking. It's pretty interesting.