One of the most replicated findings in learning science is also one of the least acted on. Retrieving information from memory strengthens the memory more than re-encountering the same information does. Reading your notes a fourth time feels productive and mostly is not. Trying to say the thing without looking is uncomfortable and works.
This is usually filed under study technique. It has an assessment implication that gets missed.
The measurement instrument is also an intervention
If retrieval produces learning, then every act of assessment is doing two things at once, and the format determines the ratio. A multiple choice question asks for recognition, which is a weak form of retrieval. An essay written over two weeks with the sources open is barely retrieval at all: it is synthesis with the material present.
A conversation about your own work is retrieval under mild pressure, with a follow-up that goes one layer deeper than you prepared for. On the learning-science account, that is close to the strongest version of the thing.
The uncomfortable implication is that the assessment formats that are easiest to scale are the ones that teach the least.
Why the follow-up matters
There is a second mechanism worth naming. A follow-up question forces articulation of reasoning that was tacit. Students frequently discover, mid-sentence, that they cannot say why they chose a method, and the discovery is the useful part. That is a generation effect: producing an explanation yourself does more than reading a correct one.
This is also why a scripted oral exam is a weaker instrument than an adaptive one. If the questions are fixed and known, students prepare answers, and preparing an answer is closer to rehearsal than retrieval.
What we take from this
We are an assessment company, so our interest here is obvious and worth declaring. But the argument does not depend on us.
- If you already run oral checks, the follow-ups are where the learning is. Depth beats coverage.
- If you cannot run them at scale, even one per term is not nothing, and the effect literature suggests spacing them matters more than length.
- If a student walks out having realised they could not explain their own choice, the check has done something a mark never does.
The claim we are careful not to make is that this replaces teaching. Retrieval strengthens what is there. It does not put it there.
Our fuller position on the evidence is at research.