at low flow rates, the debounce might hit unlucky on low-high transition noise and count a fake event?
need to scope or some detailed timestamping on all events to analyze

the debounce period is probably way too long for the physics involved (unless there are false releases while the contact drags?), so dropping it to be less than contact time at highest flow should reliably detect down and up events?

also unclear on the exact properties of the wiringPI event queueing, if any. Maybe need to up scheduling priority? But the interrupt handler supposedly already is high priority?