All valid points, and there are testing suites that do exactly what you say that are widely used (this one is pretty standard: http://www.nwea.org/products-services/computer-based-adaptiv...), but they aren't the assessment that's measured by state departments. There are probably lots of reasons for this, (cynical: they were outlobbied by Pearson, which has million dollar contracts with many states to write their state tests.), but nonetheless they exist.

The other problem is that this is a problem at which we should be throwing our best data scientists, and in many cases we're lucky if those involved have an intro stats course under their belt. Evaluating student performance is no trivial task (ESPECIALLY when trying to create some accurate measure of reading level), but we're not putting the brainpower behind it to figure out. And as much as I love it, I don't think Khan Academy will be solving those problems, but I'd love to be wrong.