Steven, thanks for the comment. Frankly, my concern is systematic emotion measurement and analysis at scale -- what I termed "emotion AI" -- rather than the sources, and like you (I believe), I am interested how emotion models might be turned to predictive use, which is roughly equivalent to your "situations that lead to an emotional response." I will say that as a technologist, I subscribe to attitudes that say you should aim for adequate and useful and then iteratively refine -- or abandon and restart if necessary -- in preference to a requirement for complete theoretical grounding from the start.

Simon, I won't dispute that "Most of emotional experience is actually 'hidden and non conscious'." Yet "nothing is hidden that will not be made manifest, nor is anything secret that will not be known and come to light." So my attitude is, measure what you can, and see what you can make of what you measure. Our descriptions will be representations rather than absolutes (cf Wittgenstein on color). Their correctness will resist proof. But they may be useful nonetheless, particularly as we develop and move to the robust methods that we agree are needed.