1. Active recall, not re-reading

Cognitive science is clear: the act of retrieving an answer from memory strengthens it far more than reading it a second time. Every study surface in Note Harbor — flashcards, practice tests, the AI tutor — is built around retrieval, not passive consumption.

2. SM-2 spaced repetition

We use the SuperMemo SM-2 algorithm. After you rate a card (Again / Hard / Good / Easy), the next interval is calculated from the previous interval, your rating, and the card's "easiness factor" (EF, starting at 2.5). "Again" resets the card to a 1-day interval; consecutive "Good" ratings roughly double the interval each time (1 → 6 → 15 → ... days). EF adjusts by up to ±0.08 per review and can never go below 1.3, so genuinely hard cards keep coming back until you master them.

3. Grounded AI generation

Every AI answer, flashcard and mindmap in Note Harbor is generated from the sources you uploaded. We inject those sources into the model context and instruct the model to cite them explicitly. When the AI can't find something in your sources, it says so instead of making it up. If you see a hallucination, please email us — we treat those as bugs.

4. Weak-topic detection

Every card carries a topic tag generated at creation time. Every review writes an event with that tag and whether you got it right. The Weak Areas panel aggregates those events by topic, keeps only topics with at least two reviews, and ranks them by wrong-rate then wrong-count. You can trust the ranking because it's a straight percentage of your own past answers.

5. Which AI models we use

Today Note Harbor uses OpenAI's GPT-4o-mini via Emergent's LLM infrastructure for all AI features. We picked it for the best cost/quality balance on long-context grounded generation. When better models ship, we switch.

6. Your data is yours

Export at any time. Delete at any time. See our Privacy Policy for the details.