This works great, and it is still how evaluation is done when the stakes are higher: PhD defenses, executive hiring... even getting hired as an entry-level engi

This works great, and it is still how evaluation is done when the stakes are higher: PhD defenses, executive hiring... even getting hired as an entry-level engineer at Google requires about five hours of what is basically oral examination. But society is not willing to pay that kind of price for the earlier levels of evaluation. We want "scalable" systems. Unfortunately those same evaluations are often treated as more sensitive than they really are. For example, if you're comparing two students I'd argue that 3.0 vs 3.5 GPA gives you at least some signal, whereas 3.5 vs 3.6 GPA gives you basically no signal at all (maybe the 3.6 student took easier courses, maybe they were more lucky with cutoffs, etc.). And yet the distinction sometimes matters e.g. to graduate programs. In well-designed systems, the GPA cutoff is set relatively low and more sensitive methods are used to select the best students from the pool. Often this includes an interview with a professor, which is also a form of oral exam.