On a related note, we have a bunch of replication failures in education for selection effects reasons. It turns if you have a highly motivated staff and engaged parents - pretty much every flavor of educational approach has a positive impact. When you try the same thing with an overworked and demotivated staff, unengaged parents, and with non-selective student populations that have behavior issues or other concerns ... most methods fall apart. And some of the approaches might even work, presuming similar conditions.

Getting policy right under adversarial conditions is really hard - even harder than the already hard problem of identifying and testing good policy.