One of the reasons that there is a large element of unpredictability in acceptances is that nobody can tell how many people will apply to any given college. These numbers have been increasing for the most popular school fairly steadily for a few decades, at least.
As @privatebanker writes, at colleges with very low acceptance rates, the unicorns are pretty easy to predict, as are kid with clear hooks. So are the unhooked kids with stats in the college’s bottom 25% or lower. There is, however, a wide swath of kids, of whom the popular colleges fill about half their class.
When 8,000 kids applied to Stanford, kids who were in the “best” 10% of the applicants were fairly likely to be accepted, since there were about 800 of them, and and Stanford was looking to accept 2,000 students. However, when there are over 47,000 students,the “best” 10% make up 4,700, they’re more likely to be rejected than not.
By “best”, I mean “best fit” based on whatever holistic methodology the colleges is using, not by simple stats.
In 2016, about 36,600 students applied to Stanford, versus about 47,500 in 2018. That meant that there were more students who would be at all “ranks” of the applicants, meaning that fewer would be accepted, unless the were the very very top, which would still be relatively few in number. Somebody calculating the chances of a Stanford applicant in 2018, using the acceptances of the class of 2020, could overestimate the likelihood of acceptance for all but the very very top applicants.
Since Stanford no longer publishes the number of applicants, this will make these predictions even more difficult in the future, especially if the number of applicants grows at the rate it is doing now.