Ah yes, I see you are correct.
I gave up on seeing the “both valid” condition to quickly while debugging.
If I dump a stream of write lines to the debug window I do indeed see a mix of events fired where both, left or right eyes are showing as valid when both eyes are open and being seen by the sensor.
So yes, I would agree that the data stream does seem a bit glitchy right now and not reliable enough on its own to sense when either eye is closed or when both eyes are closed.
I do have code that toggles a bound dependency property to indicate when a user is present or not but I did not end up using the UserPresenceChanged event for my blink detection logic. My current logic detects both individual left/right eye blinks as well as both eye blinks. The messy aspect of it is that it required the dispatch timer firing in the background to timeout my eye open variables which introduces an additional performance hit that would have been nice to avoid.
So I do have something that is working, I just wanted to provide some feedback based on what I was experiencing with the alpha SDK. Now that you have clarified that the stream is a mix of both eye and individual eye events, my feedback would be that it would be nice if the events more reliably reported back on both eyes each time. However, I am sure there are lots of factors involved and I appreciate that some of those factors may make it more challenging to do so.