It is fairly accurate. I would say within a couple of MPH. Wheel slip would throw it off, but it is not noticed.
If I remember correctly, my sensor was magnetic too, but I removed it and made a simple switch for it. To trip the sensor, I put a bump (cog?) on the axle. It hits a brass strip which makes contact with a screw point. That way it hits the contact once per revolution of the wheel / axle.
The switch closure effectively measures the RPM of the wheel. The electronics in the bicycle speedometer calculates the speed.
QSI sound decoders have an audio “talk back” that will tell you how fast the engine is moving. Some Atlas and BLI engines use these decoders. In comparing different engines I suspect the speeds could be a few scale MPH off, but it seems relatively accurate.
For the bike speedometer, both the sensor and the instrument/display could be mounted on a car with the sensor situated so that it can run close to small magnets hidden in items at discrete and defined distances from each other along the right of way. Just as a rotating bike wheel makes the magnet on the spoke move past the sensor at various intervals, the one constant in either case will be the length of time between magnets or the circumference of the bike wheel. As the wheel turns faster, the interval shortens, and it would work the same with lineside magnets. The instrument can be programmed for the distance/circumference due to their varying circumferences on bikes, and to your lineside details as well.
The problem will be in finding a scale conversion because the HO scale movement past the magnets will be much longer between sensings, unlike the quickly revolving wheel on a bike.
It’s not distorted; it’s just isn’t loud enough to be heard over the extraneous steamer sounds. I’ll try my Atlas HH600/660 tonight and see if that one is any better.