Familiarity is real knowledge, but it is not the whole target
Recognition supports reading and listening. Seeing a construction and understanding it quickly is valuable. Trouble begins only when that success is used as evidence that the learner can produce the same relationship independently.
A learner may recognise dozens of elegant hedging phrases and still default to one blunt structure in meetings because the alternatives are not retrievable under production conditions.
Retrieval research explains why the distinction matters
Classic retrieval-practice experiments separate re-exposure from the act of producing an answer from memory and show durable benefits of repeated retrieval under tested conditions. Language tasks are more complex than laboratory recall, but the distinction gives a useful design principle.
If the later task requires access without a model, at least some practice should also remove the model. Metkagram Choices and production prompts do that deliberately.
Use recognition to learn and recall to test access
Inspect examples until the Frame and its constraints make sense. Then hide the wording and retrieve from the Move or from a new situation. If recall fails, return to the example briefly and try again. This avoids the false choice between explanation and active recall; they solve different problems.
Over time, vary the cue as well. The goal is not perfect recall of one prompt but flexible access to the Frame from the situations where it can actually help.