The evidence, in full
Here's the research behind our method, grouped so you can see what each source shows. Most of it is mechanism — why stories and characters help ideas stick. We're honest that we haven't yet run our own controlled study of test-score gains. References are listed in their original academic form; email educators@sparkanvil.org for PDFs or per-app curriculum mapping.
Frameworks & mechanisms
- Design constraintNaul, E. & Liu, M. (2020). Why story matters: a review of narrative in serious games. Journal of Educational Computing Research, 58(3), 687–707. — the four-feature framework (distributed narrative · intrinsic integration · empathetic characters · adaptive storytelling).
- Design constraintHabgood, M. P. J. & Ainsworth, S. E. (2011). Motivating children to learn effectively: exploring the value of intrinsic integration in educational games. Journal of the Learning Sciences, 20(2), 169–206.
- MechanismBruner, J. S. (1986). Actual Minds, Possible Worlds. Harvard University Press. — the narrative vs. logical-categorical modes of thought.
- MechanismGreenspan, S. I. & Wieder, S. (1998/2006). The developmental (DIR/Floortime) model — abstract reasoning is rooted in emotional experience ("thinking begins in feeling").
- MechanismSfard, A. (1998). On two metaphors for learning and the dangers of choosing just one. Educational Researcher, 27(2), 4–13. — the participation metaphor.
- MechanismGreen, M. C. & Brock, T. C. (2000). The role of transportation in the persuasiveness of public narratives. Journal of Personality and Social Psychology, 79(5), 701–721.
- MechanismHirsh-Pasek, K., Zosh, J. M., Golinkoff, R. M., Gray, J. H., Robb, M. B. & Kaufman, J. (2015). Putting education in "educational" apps. Psychological Science in the Public Interest, 16(1), 3–34. — the four pillars of active learning (active · engaged · meaningful · socially interactive).
- Market-proofOn-screen pedagogical agents (2021–2025): a decade of meta-analyses (e.g. Schroeder, Adesope & Gilbert; Castro-Alonso et al.) find small but reliable gains in retention, transfer and motivation when the character has clear emotion, gesture and a defined teaching role.
- Design constraintMcNeil, N. M., Jordan, N. C., Viegut, A. A. & Ansari, D. (2025). What the science of learning teaches us about arithmetic fluency. Psychological Science in the Public Interest, 26(1), 10–57. — recommends retrieval practice and time-limited practice once a child is accurate. We do not time practice: that is our calm design law, a choice we make on purpose, not a finding that timing harms. A child fills our Fact Field by naming a way to a fact, never by speed.
The AI-mentor evidence
- MechanismBastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. (2025). Generative AI can harm learning. PNAS. — students given answer-giving ChatGPT did worse on an unaided exam; a hints-only version erased the harm. Our mentor is built on the second model.
- MechanismChase, C. C., Chin, D. B., Oppezzo, M. A. & Schwartz, D. L. (2009). Teachable agents and the protégé effect. Journal of Science Education and Technology, 18, 334–352. (Stanford / Vanderbilt) — kids who teach a character work harder and learn more; the weakest gain the most.
Proven in the market
- Market-proofStory Time Chess (2018–present). A character-driven method that independently converges on the same four features and is used in schools and homes worldwide — evidence the approach works at scale (we apply it across a whole library, not one game).
Notice the balance: the evidence is strong on mechanism (why the method works) and honest about outcomes (we haven't measured our own test-score gains yet). That's deliberate. We'd rather show you the reasoning than promise a number we can't back up.