Yes, that used to be kind of what the Delay Compensation did … it was a way to record at low latency in a project that had high latency plugins.
So, 20 years later …
There is enough memory and CPU available for the project to play ahead. Imagine an invisible cursor that has already played your song and loaded that into memory. When you click play, it looks like it is playing what is under the cursor, when in reality it is streaming stuff it wrote into memory awhile ago. To you it looks like it is just now playing ( the project cursor ), but to the project it is already playing 20ms or whatever ahead of you… you just don’t hear it yet.
So, you click the monitor button … it simply subtracts the extra buffers and adds to the stream at the correct location. I’m oversimplifying a bit I’m sure, but it isn’t too far off of this. It also does take system resources to do the “look ahead” processing.
Instead of Constrain Delay removing the plugins from the path, the system now just reads the output into memory before YOUR cursor gets to the point you need to hear it.