Use the cue when it solves a real noticing problem
On first exposure, keep the Mark visible long enough to understand what it points to. Connect it to the sentence, the Frame and a short explanation. There is no prize for hiding information before the learner knows what they are supposed to retrieve.
Textual-enhancement research supports treating visual salience as a plausible attention tool, while also showing why the cue itself should not be confused with acquisition. Seeing is a step; it is not the finish line.
Fade in stages instead of switching from full help to none
One useful progression is: full annotated sentence, plain sentence, Frame with slots, meaning or Move only. Each stage removes one source of information while keeping the task recognisable. The learner can stop at the first stage that becomes genuinely difficult and then return to a lighter hint.
This is better than a binary “easy/hard” design. Scaffolding can shrink as memory becomes more reliable, and the system can still offer help without immediately revealing the complete answer.
Make retrieval precede the reveal
Try to reconstruct before reopening the Mark or Frame. Even a short attempt changes the task from rereading into retrieval. Research on retrieval practice comes from broader learning contexts, not from a direct trial of Metkagram, so the correct claim is modest: it gives a rationale for asking memory to work before feedback arrives.
Feedback still matters. After the attempt, compare the reconstruction with the reviewed example and fix the specific part that drifted.
Know when a Mark has finished its job
A Mark can be retired when you can read an unannotated example, identify the useful structure, recall the Frame and produce a fresh sentence without needing the cue first. That does not mean you will never look at annotation again. Difficult examples can reopen the layer as an inspector.
The goal is independence with access to support, not a ritual in which all hints must disappear permanently.