Your has so much good information that I hope I don't miss anything important out!

First of all, many of your arguments reveal my ignorance on this concept of EILD. I may be biased towards thinking that a good deal of volition would "solve" most of the problem with successive encounters with the cue during the day. The fact - which you explain perfectly - is that despite this, the problem of (de)sensitization persists:

- Stimulus generalization: every light can be considered the ND (NovaDreamer). This means you perform reality checks like a mad man the entire day. But at some point, extinction will occur.
- Stimulus discrimination: it cannot occur at all. Since the stimulus is not specific at all (and add the fact that you rely on it's mysterious and unpredictable incorporation into the dream), you cannot risk training yourself to distinguish different types of lights. Even if the light is extremely bright, that can lead to a self-justification (like the report you showed)...

Still, (and this is something that I feel it's relevant throughout most of this reply), I may here sound arrogant or optimistic (definitively dwelling on confirmation bias I guess ), but there are reasons to wonder if the users are maximizing their chances with the instrument. Two examples:

- Majority of reports include people using it all night. Imo, this is a complete waste for several reasons: bright light affects your circadian clock, it might wake you up, and even if you achieve lucidity within the first sleep cycles, it's duration won't be big enough to compensate for the possible negative effects. Besides, couldn't that be the cause for reports of desensitization? How many flashes per 10 minutes are we talking about? Can you calibrate it?
- In this meta-analysis, it's reported a low-methodological quality of MILD studies. Regardless of the use of MILD and EILD being presented as one of the best methods for induction, did LaBerge really tested MILD or even EILD with several samples regarding their levels of prospective memory? That would be relevant.

However I think there is a way around this that would help - and thats to make the light a stronger dreamsign. I never tried this at the time of using the novadreamer 20 yrs ago (and I've moved onto vibration as a cue now) but looking back at others experiences I think the key to using a light based signal is to make it VERY VERY BRIGHT.
Great catch !

That would indeed allow for a stimulus discrimination: if the bright lights always pull your attention, that would divert much of the tiring process of looking out for them. Meaning you wouldn't spend the whole day trying to catch a "light". You would just need to develop a response regarding every light that produces negative responses (well, I guess this would still cause some problems during the night).
One thing that is relevant: I keep trying to remind myself to get that study, but it's present in the book The Mind at Night. It basically mentions that interrupting REM pressures your brain to experience even more of it, mentioning some significant changes that could be compared to a person that sleeps 1 or 2 more hours (don't know exact estimates because I don't have the book on me, and the text itself wasn't exactly explicit in the figures). Still, waking up wouldn't be the worst situation in the world, especially considering the opportunities that would create for a WILD (especially DEILD): people like VagalTone rely on this to achieve many of their lucids.

Interestingly in the last 3 quotes (in all but Lynne Levitan from the lucidity institute) none of them got lucid from those bright lights - they just kept on dreaming - but at least afterwards they recognized them as a pretty obvious dreamsign.
Now imagine what happens once your response towards that specific dreamsign becomes automatic ^^

You should only expect to see about 10-20% of the cues even if your dream recall is good
Once again, I wonder what they based those statistics on. What's interesting is that they seem to talk about 2 different things: seeing the lights (regardless of whether you become lucid or not) and recalling the lights are different things, especially assuming the amount of dream content that is lost especially in the earlier stages of the night. But the rules apply for everyone (sadly ): the huge chunk of REM is in the last hours of the night, so it's not like they can have failed taking that into consideration...Still, I can't help but feel skeptic towards those numbers, but it's unrealistic to expect a huge number of dream incorporation when that's most likely not the case.

after this my brain seems to ignore it going off, so even within a short space of time the level of threshold required to get the stimuli incorporated into the dream can alter. I try and use it one night on one night off to limit habituation/desensitization. I like using vibration as it is a pretty unique dreamsign to recognize, and it doesn't occur that naturally in the real world (but it can be simulated for training/conditioning purposes pretty easily).
Interesting. I hope you don't mind the question (just trying to see the overall image), but in LaBerge book he said that light cues were the most effective ones according to their studies. Do you agree and have some preferred taste for vibration cues, or you think that depends a lot on the person? Since I got no experience with EILD devices, one can only ask