Yes, that part of the scenario still falls inside what I explained poorly 
At that point you describe you are right, if you have a very recent reading before any lights were turned on then that is definitely the one you want to use. The comment I was making relates more to the second entry into the room where potentially the last lux reading had been taken before the lights were turned off. What I was suggesting was that in that case, use the reading before the first entry into the room, being the most recent reading without any lights on in the room. Perhaps this might illustrate it better:
Lux Reading A - low light, recorded with no lights on, stored in a variable or device, etc
Person enters the room
Lights turn on - based on motion and low light from Lux Reading A
Lux Reading B - bright light, recorded with lights on
Lux Reading C - bright light, recorded with lights on
Person leaves the room
Lights turn off
Person enters the room
Lights turn on - based on motion and value in variable for last Lux Reading with no lights on, i.e. reading A
If a lux reading D is recorded in between the lights being turned off and the second entry into the room, then the lux reading stored in the variable (A) would be overridden by D and that would be used when assessing whether to turn the lights on the second time, which I expect is what you were questioning in your last comment.
I'm not claiming it is a good solution or even a complete solution, just an idea. It would fall down in a number of areas, including if the natural light had increased in between the person leaving the room and re-entering and we were still relying on Reading A.