CAVEAT — 2025 conceptual replication (Cognitive Research: Principles and Implications) — generation effect did NOT reliably transfer to learning from expository text; some experiments showed disadvantage

Caveat: A 2025 conceptual replication (Cognitive Research: Principles and Implications) found the generation effect did NOT reliably transfer to learning from expository text — some experiments even showed a disadvantage. So the lab-robust word-level effect does NOT automatically scale to complex real-world content.

Source: Cognitive Research: Principles and Implications (2025).

Confidence: Verified (limit / scaling).

Why this matters for Candid: Honest scoping for client conversations — the d ≈ 0.40 finding is for word-level / short-input generation, not for "users learn faster if they generate paragraphs about a complex product." Pair with R4 — Where appropriate, make the user GENERATE inputs (not just pick from menus) — the generation effect d≈0.40 is real but ceilings beyond ~900 words and doesn't scale to expository text but keep the scope realistic.