Figure out what might be meant by "a good employee." This is hard. Maybe factor in tenure, promotions, projects delivered, peer evaluations, etc. Develop of set of measures for an employee's overall all-time job performance.
Now take each candidate's "interview binder" and map it to some set of ratings (coding proficiency, favorite ice cream flavor, etc.). Set thresholds for each rating initially to reach hiring targets and some vague notion of an acceptable employee profile.
Always allow some small percentage of candidates below the thresholds to receive an offer.
Correlate the interview rating dimensions with the set of employee performance measures as employees work through the job.
Adjust the interview thresholds and experiment with adding or removing dimensions as you learn which things in the interview are correlated with "good employees."
Maybe this is how it's done when it's done properly, but it's the "extend offers to a few people below the cut, and see if they fare worse" idea that I like.
Anyone can write a blog post about what they think is a good interview but there's no empirical proof that their interview technique works.