As for myself, I have stated that some of the studies I have posted here are lifted from the pages of my super-secret epic tome, "The Chulanomicon".
Sounds silly and somewhat pompous, but it's almost true.
One example might be the study of the impact of winning or losing the first point on the win probs for each post.
That was posted here a ways back. But of course we could take it further into Q's, Ex's, and Tri's.
Let's just say that if you could predict the first point with 100% accuracy, eg post 2 wins, I am showing the Q's 27 and 28 as 52% and 75% more likely, respectively. Meaning, if not heavily favored, could easily overcome the takeout. However, since you aren't going to get 100%, it's just a matter of working out the gains for the various success rates of your predictions.
Get it right 3 out of 4, and you are still in business, just not so spectacularly.
IMO, observational hcap + cherry picking the best cases might get the job done. Of course you can't have too many obfavs, so that's an issue.
One takeaway from this is that although there are potentially 56 head to head match-ups each game, some are more impactful than others. A computer program would expend an identical amount of effort in calculating each of them, but a human might choose to prioritize accordingly.
Another takeaway is that a King of the Hill contest based on predicting the first point for each game might be very revealing.