The 2006 study compared students who reread material against students who were tested on it. Rereaders predicted they would remember more — and after five minutes they did. After a week, the tested students had retained significantly more. Testing, Roediger and Karpicke argued, does not merely measure memory; it creates it.
The 2011 paper extended the finding to successive relearning — spaced retrieval practice at widening intervals — which remains, by effect size, the strongest study strategy ever documented.
Fluera’s retrieval architecture is built on this evidence. The Socratic mode is retrieval under scaffolding. Fog of War is retrieval under occlusion. Spaced repetition is successive relearning. And the core pedagogical refusal — that the AI asks before it answers — exists because Roediger and Karpicke’s data is unambiguous: being quizzed is how memory becomes permanent.