A corpus is evidence, not a finished lesson
Concordance lines can show whether a phrase occurs, what words cluster around it and how usage varies. That evidence is powerful, but raw examples do not automatically identify which distinction a learner needs, how broad a Frame should be or which neighbouring pattern deserves comparison.
Formulaic-language research has used corpus-informed approaches precisely because repeated multiword structure can be investigated empirically. The pedagogical step still requires selection and interpretation.
Metkagram adds curation and stable identities
A reviewed Frame gets a stable identifier, language identity, constraints, examples and links to Moves or Contrasts. That allows the same object to appear in a lesson, API, AI tutor or later evaluation. The goal is not to replace corpus evidence but to make a subset of it usable as a maintained teaching system.
This distinction also makes provenance clearer. A corpus can support the observation that a structure is used in particular contexts; an editor still owns the decision to publish a certain abstraction as a Metkagram Frame.
The strongest workflow moves in both directions
Corpus evidence can challenge an existing Frame: perhaps the examples are too narrow, the register label is wrong or a supposed pattern is rare outside one phrase. A curated library can in turn identify questions worth checking in corpus data, such as whether two Frames really differ in context or frequency.
For the learner, the result should remain simple. They see reviewed examples and decisions rather than a wall of concordance lines, while the project can still use broader language evidence behind the editorial process.