These are three selected studies, rather than a review of all relevant evidence or repeats of one AI intervention. The math and coding cases distinguish assisted performance from an immediate assessment without AI; the physics case reports a positive post-test comparison with test-time assistance unspecified. Apply their different assistance packages and allocation units to the contributor/reader practice hypothesis: record the aid each route provides and distinguish completed work from explanation and checking without AI. Active tutoring is not the same intervention as quiet path-following. Measure explanation and checking immediately and later. Separately record total authoring, review and learning effort and later-reader reuse. This motivates testing proposed routes, without establishing 1Cademy’s benefit or long-term scientists’ skill loss.