About railmon
What this system does, in plain terms.
A point machine is the electric motor that flips a railway switch. Every time it throws a switch, it draws a burst of electricity — high while working, then dropping to zero once the switch locks. A healthy flip has a very consistent shape. A faulty one looks different: the motor never really starts, never finishes, or the current does something jagged. railmon reads that electrical fingerprint on every throw and flags the ones that look wrong — because a switch that didn't lock properly is a derailment hazard.
Most detectors say only "good" or "bad." railmon has a third answer: not sure — a human should check. That's deliberate. In a safety system, a confident wrong answer is the dangerous one, so when it's uncertain it raises its hand instead of guessing. The number that matters most is faults missed silently — the goal is to keep that at zero.
The detection is the real algorithm — not a mockup. What's simulated is the test track: known-answer throws (across all four machines, with a realistic spread of faults) so you can watch it get them right or wrong. On real track data it gives a verdict but won't grade itself, because out there nobody knows the right answer in advance.
On the Accuracy Test page you can see the tradeoff: turning sensitivity up catches more subtle faults but sends more throws to a human to double-check. That's a business decision — how much caution versus how much review workload — and it's yours to set.