@nishtdayngesheft: When you move “the plastic in front of the sensor” the metzios is that the switch will always be able to intitiate the opposite event (i.e. vs. unblocking) because not only is “the system … programmed to recognize that the light is on” as @DaasYochid guesses, but in fact the attempt to send/receive a pulse via a pair of transmitter/receiver devices inside the switch is always essentially an instantaneous programmed event sequence.
Thus immediately after initiating such an event (“Attempt a Transmission” in the loshon of patent US7872576) the [switch] system is programmed to infer and record (i.e. into processor memory) the position of the blocker. As @RebbeYid pointed out [last week] this is as if the “single computer controlling everything says, I’m going to shine a light into my own eye, and if this human puts something in the way, I won’t turn on his light…”.
All of the random delays described in the patent, have no effect on the metzios of an individual Transmission event. Instead, the kinds of randomness they introduce include:
[A] How long to wait (i.e. from time warning light turns green) to initiate a Transmission event.
How long to delay after the Transmission event, to decide if the results of that event indicate the switch should be changed.
[C] Pseudo-randomly decide if an indicated switch change event (e.g. turn on/off externally controlled light) should actually take effect now, or if to instead re-start a new cycle (i.e. [A] wait/transmit/etc.)
So while at the macro level the change in the external circuit (i.e. lights on/off) may theoretically be delayed for a very long time (i.e. by random cycle of A/B/C happening to occur numerous times) it would seem IMHO that the changes happening within the switch at the micro level are of at least equal concern, and need to be dealt with in any complete analysis of why the entire mechanism might (or not) be muttar lechatchilah.