Built on what works.
SetFlow isn't designed on intuition. Every Tori interaction, every flashcard deck, every practice exam, every Socratic prompt is grounded in five well-replicated findings from the last hundred years of learning research.
Spacing effect
~140 yrs of evidence
Testing effect
~70 yrs of evidence
Interleaving
~50 yrs of evidence
Bloom's taxonomy
1956 of evidence
Cognitive load
~35 yrs of evidence
Hermann Ebbinghaus discovered the forgetting curve in 1885. We design against it.
Ebbinghaus showed that without review, learners forget roughly half of new material within 24 hours and ~75% within a week. The decay is exponential. A century of replication studies (Bahrick 1979, Cepeda et al. 2006) confirms the shape of the curve and shows that spaced repetition — reviewing material at expanding intervals — flattens it dramatically.
SetFlow's flashcard system uses the SM-2 algorithm (Wozniak 1990), the same scheduler that powers Anki. Cards you almost-remember reappear sooner; cards you nail get pushed further out. The next review is always the one your brain is about to drop.
In SetFlow
Every flashcard deck the student creates — or that Tori generates from uploaded notes — is automatically scheduled. The student doesn't pick review dates; the algorithm does, against each individual card's forgetting curve.
The forgetting curve
Retention over 30 days, with and without spaced review
The testing effect
Retention one week later by study method
Adapted from Roediger & Karpicke (2006). Self-quizzing more than doubles retention vs. re-reading the same passage four times.
Practising recall beats re-reading. Every controlled trial says so.
Roediger & Karpicke (2006) ran the canonical experiment: students who studied a passage once and then quizzed themselves on it remembered ~80% of the material a week later. Students who re-read the same passage four times remembered just ~40%. Karpicke & Blunt (2011) replicated this for science concepts and concept maps.
The mechanism is retrieval-induced consolidation: the act of pulling information out of memory strengthens the neural pathway more than passive re-exposure ever does. Quizzing isn't how you check what was learned. Quizzing is how the learning happens.
In SetFlow
Tori generates retrieval-style practice questions from uploaded notes — not the multiple-choice trivia you see in most LMS quizzes, but short-answer and explain-this-concept prompts that force genuine recall. Mock exams pull from the student's own material so every test is also a study session.
Knowing the answer is the bottom of the pyramid. Tori climbs.
Benjamin Bloom's 1956 framework (revised by Anderson & Krathwohl in 2001) ranks cognitive skills from remembering at the bottom to creating at the top. Genuine understanding lives at analysing, evaluating, and creating — not at remembering a definition.
Most AI tutors stop at remembering: they tell you the answer. Tori's student persona uses the Socratic method (Paul & Elder 2007) to walk you up the pyramid one rung at a time. It asks what you tried, where the gap is, then asks the question that makes the gap visible.
In SetFlow
Tori is hard-coded to refuse to write a student's assignment. When you ask a question, it asks one back. When you give it a draft, it tells you where the reasoning is thin — not what to write. The persona itself is the pedagogy.
Bloom's taxonomy of cognitive skills
Mapped to SetFlow surfaces
Anderson & Krathwohl (2001). Higher-order skills (top of the pyramid) require prerequisites at every layer below.
Interleaving vs. blocked practice
Math-skill accuracy during practice, then on a delayed test
Blocked practice
Interleaved practice
Adapted from Rohrer & Pashler (2007). Interleaved learners feel less confident during practice — and remember nearly twice as much a week later.
Mixed practice feels harder. It works better.
Rohrer & Pashler (2007) compared two ways to learn four math concepts: drill one to mastery, then move on (blocked practice) — or rotate between all four every day (interleaved practice). The interleaved group reported feeling less confident during practice. They scored ~76% on the test; the blocked group scored ~38%. The pattern replicates across math, surgery training, and motor skills (Dunlosky et al. 2013).
Why: blocked practice lets the learner pattern-match the problem to the chapter they're in. Interleaved practice forces them to identify which technique applies — which is what real exams and real life require.
In SetFlow
Mock exams Tori generates pull from across uploaded materials, not a single chapter. The exam-prep persona for SAT / ACT / AP rotates problem types deliberately so the student practises identifying, not just executing.
Working memory is small. Build the worked example, not the wall of text.
John Sweller's cognitive load theory (1988, refined through 2010s) starts from a finite working-memory budget — Miller's famous seven plus or minus two chunks. A learner who has to hold a paragraph of text and a problem and a worked example all at once will fail on all three. Splitting the load — the worked-example effect, the redundancy effect, the modality effect — reliably outperforms density.
The implication for SetFlow is concrete: Tori's answers don't come back as walls of prose. They come back as one explained step, one follow-up question, and the diagram or formula needed to take the next step. The student's working memory does not get blown out.
In SetFlow
Lesson generation, exam-pack PDFs, and the writing assistant all chunk content into named, scannable beats — short sentences, headings every couple of paragraphs, diagrams for spatial concepts. The document is the pedagogy.
Working-memory budget
How a learner's attention partitions during study
Intrinsic load
The inherent difficulty of the material itself.
Germane load
Effort that builds long-term schemas. Maximise.
Extraneous load
Wasted effort on layout, formatting, ambiguity. Minimise.
Sweller, van Merriënboer & Paas (2019). SetFlow's lesson and Tori response formats are deliberately structured to shrink extraneous load and free up working memory for the germane kind.
Primary references
- Ebbinghaus, H. (1885). Über das Gedächtnis. Untersuchungen zur experimentellen Psychologie.
- Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
- Karpicke, J. D., & Blunt, J. R. (2011). Retrieval practice produces more learning than elaborative studying with concept mapping. Science, 331(6018), 772–775.
- Anderson, L. W., & Krathwohl, D. R. (Eds.). (2001). A taxonomy for learning, teaching, and assessing: A revision of Bloom's taxonomy of educational objectives.
- Rohrer, D., & Pashler, H. (2007). Increasing retention without increasing study time. Current Directions in Psychological Science, 16(4), 183–186.
- Sweller, J., van Merriënboer, J. J. G., & Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. Educational Psychology Review, 31, 261–292.
- Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students' learning with effective learning techniques. Psychological Science in the Public Interest, 14(1), 4–58.
Every claim above is independently citable. We're happy to send a printed bibliography to journalists or curriculum leads on request — email [email protected].