That certainly looks like an improvement, but I would imagine that things like poverty and the student's home situation would still play a similar role to nutrition. That is: given two children of equivalent raw "intelligence", if one has a stable home situation with loving parents who stress focus and discipline and the other bounces between homes in the foster care system with no good role models, I would expect the former to score higher on just about any test you could give them.

> the tasks involved are things nobody really studies for

Perhaps that is true today, but I see no reason Goodhart's law would not apply -- if you start basing a student's educational opportunities on how well they score on an RPM test, then a cottage industry of RPM test prep will emerge. There are some tests that are exceedingly hard to train for (e.g. dual n-back comes to mind) but looking at RPM it doesn't seem like something that would take much practice to pick up some reasonable strategies.

Again, I still think testing is very useful, especially for spotting outliers in a similar-enough population and ensuring they get additional resources necessary to excel. I'm just skeptical about social scientists' claims that intelligence can be objectively measured by a multiple-choice test in a way that cannot be gamed by people with enough resources.