Earlier quoted context omitted.
Regarding 2), you do it via Value Added Modeling. You build a statistical predictor of student performance (based on standardized test scores) then measure Performance = Avg(Actual - predicted).
Statistical forward looking models proposed by a guy who works in finance? How could that ever go wrong :) EDIT: More seriously, if it's already impossible to measure, I fail to see how more complicated metrics would do anything but do an even worse job of measuring. There might be exceptions to this rule in a few cases but it's generally the way things work.
If you have low quality students, who are only expected to score at the 25% level, and you boost them up to 35%, you win. If you have high quality students expected to reach 75%, but the only achieve 65%, you lose. Do you believe it's really that hard to predict that Asian kids in 2 parent $250k/year+ households will score higher than black kids in 1 grandparent $10-20k/year households?
If you truly believe the effects of education are immeasurable, perhaps we should stop wasting money on it. We don't spend $859 billion on homeopathy for much the same reason.