THE SCIENCE OF NOT
LYING TO YOURSELF.

01 // THE GENERATION EFFECT

Self-generated answers are retained far better than answers a learner merely reads or recognizes. Flipping a flashcard is a reading event. Answering a prompt from scratch is a generation event.

Reference: Slamecka & Graf, 1978. "The generation effect: Delineation of a phenomenon." Journal of Experimental Psychology.
[DOI: 10.1037/0278-7393.4.6.592]

02 // THE ILLUSION OF COMPETENCE & CALIBRATION

Learners are demonstrably bad at knowing what they know. The same cue-then-answer structure that makes flashcards easy to build also creates false confidence. This is why Memblit measures the calibration delta between what you predict you know and what you actually know.

Reference: Koriat & Bjork, 2005. "Illusions of competence in monitoring one's knowledge during study." Journal of Experimental Psychology.
[DOI: 10.1037/0278-7393.31.2.187]

Reference: Koriat, Lichtenstein & Fischhoff, 1980. "Reasons for confidence." Journal of Experimental Psychology.
[DOI: 10.1037/0278-7393.6.2.107]

03 // SELF-EXPLANATION

Prompting a learner to generate their own explanation produces massive learning gains (g = 0.67), but if a system (like an AI tutor) hands the learner the explanation, the benefit collapses. We force you to explain it to us.

Reference: Bisra et al., 2018. "Inducing Self-Explanation: a Meta-Analysis." Educational Psychology Review.
[DOI: 10.1007/s10648-018-9434-x]

04 // INTERLEAVING AS DISCRIMINATION TRAINING

Mixing content isn't enough. Interleaving works because it forces the learner to select the correct approach among confusable alternatives. Memblit clusters commonly confused concepts and forces discrimination between them.

Reference: Rohrer et al., 2014. "Teaching spaced and interleaved mathematics." Journal of Educational Psychology.
[DOI: 10.1037/edu0000001]

05 // SLEEP CONSOLIDATION

The largest single consolidation event in memory formation happens during the first night of NREM sleep after learning. Memblit schedules around your sleep consolidation windows, not just elapsed calendar days.

Reference: Murre & Dros, 2015. "Replication and Analysis of Ebbinghaus' Forgetting Curve." PLoS One.
[DOI: 10.1371/journal.pone.0120644]

Reference: Hu et al., 2020. "Targeted memory reactivation during sleep." Psychological Bulletin.
[DOI: 10.1037/bul0000230]

06 // THE AI CRUTCH EFFECT

Unrestricted AI tutoring improves practice scores but can lower unaided exam performance because students rely on it as a crutch. Memblit is designed to prevent this by never handing you the answer directly.

Reference: Axios / Wharton Study, 2024. "AI Tutors in Education."
[Link: axios.com/2024...]

07 // GRADING CONSISTENCY

AI grading of free-text answers reaches 65–85% agreement with human markers, depending on rubric complexity. That ceiling is real, and pretending otherwise is how products ship inconsistent scores. Memblit's response: every submission is graded twice at temperature zero. If the two scores diverge by more than 8 points, a third pass breaks the tie, and the disagreement is logged as a first-class metric (grading_consistency_rate) tracked from day one.

Engineering commitment: Double-grading at temp 0, third-pass on >8pt divergence, grading_disagreement_flag tracked per submission. This directly prevents the inconsistent-scoring failure mode documented in competitor user reviews.

08 // THE KNOWLEDGE ONTOLOGY

Traditional study tools track items in flat lists. Memblit structures every concept a student has produced into a graph: concept_units as nodes, confusable_pairs as edges, with calibration history and consolidation state attached to each. While calibration training has strong theoretical and adjacent empirical grounding, Memblit is actively testing this exact architectural application at scale. This is an architecture moat — it compounds with usage data and cannot be replicated by uploading the same material to another app.

Architecture note: The distinction between architecture moats (compound with data) and content moats (copyable with contractor-hours) is the central design principle. Memblit's differentiation rests on the former, never the latter.

09 // YOUR CONCEPT GRAPH

Every node is a concept you've proven (or failed to prove). Every edge links concepts your brain keeps confusing. Green nodes are well-calibrated. Pink nodes are where you're overconfident. No competitor combines a concept-relationship graph with calibration scoring and confusable-pair discrimination — because none of them track the relationships between what you know and what you think you know.