Encounter count is informative but incomplete
Meta-analytic work on incidental vocabulary learning finds a positive relationship between repeated encounters and learning. That is useful evidence against the fantasy that one exposure is enough for everything. It still does not give a universal threshold that transfers unchanged across words, Frames, learners and tasks.
A familiar transparent phrase may need less deliberate attention than a subtle register contrast that repeatedly causes errors.
Ten identical repetitions are not the same as ten useful encounters
Seeing the same completed sentence ten times can become effortless recognition. Distributing encounters over time, varying the sentence and occasionally requiring retrieval changes the work being done. Spacing research shows why timing matters; meaning-focused formulaic-sequence research shows why context matters too.
For a Frame, the useful question becomes: what should this next encounter make the learner do that the previous one did not?
Stop counting when the learning decision becomes stable
A practical system can use performance rather than a magical number. If a learner can retrieve the Frame after a delay, distinguish it from nearby options and use it in a changed context, another identical exposure may have low value. If the same boundary keeps failing, the Frame deserves another return or a better Contrast.
This is more demanding than a fixed repetition schedule, but it aligns repetition with the actual skill being built rather than with a counter.