Find out what you only think you know.
Paste your notes. Answer five questions from memory, saying how sure you are before each result. Recall First shows you the gap between what you felt you knew and what you could actually recall, and tells you what to study first.
2 of 5
You were Certain on 3 answers — and got 2 of them wrong.
What does the hippocampus do during sleep, according to your notes?
You wroteIt stores short-term memories.
CorrectIt replays the day's patterns and transfers them to the neocortex for long-term storage.
Summaries feel like studying. They aren't.
Ask an AI to summarise your notes and you get a smooth paragraph that reads as if you already know it. That smoothness is the problem: fluency of reading gets mistaken for knowledge, and people stop studying too early.
Recall First never explains first. You answer, you commit to a confidence, and only then do you see the result. The thing it measures is the one most study tools ignore: whether your confidence is earned.
How it works
-
1
Paste your notes
Lecture notes, a textbook section, your own summary. About a page is ideal. Mix in a second topic if you like.
-
2
Answer from memory
Five short-answer questions written only from your text. The notes are hidden. Type or speak your answer, then say how sure you are: Guessing, Unsure, Fairly sure, or Certain.
-
3
Get an honest grade
Right, partly right, or wrong, with a line of feedback and the correct answer quoted from your notes. Paraphrases count. Half-answers get half credit.
-
4
See the gap
Your felt confidence next to your real score, the answers you were sure about but got wrong, and a second chance at exactly those.
Points for honesty, not just answers.
Every answer earns calibration XP. Certain and right is worth the most. Certain and wrong costs the most. "I don't know" costs nothing, because honesty is never the mistake. Keep a daily streak, watch your gap shrink.
Built around one idea: is your confidence earned?
Catch the illusion
- Calibration scale. Felt vs knew on one line, the gap highlighted.
- Certainty-based marking. Sure and right earns most; sure and wrong costs most. You see what honesty would have scored.
- Fluency traps. Answered fast, Certain, and wrong: the feeling of knowing without the knowing.
- Spot the false sentence. An AI summary of your notes with one planted error. It reads just as smoothly.
Close the gap
- Fix these now. A new question on each confident miss, from a different angle. High-confidence errors are the easiest to correct once seen.
- Teach the gap. Write the correction in your own words; it's checked against your notes.
- Hints, not answers. Three hints, weakest first. A right answer after a hint counts half.
- Blind-spot map. Which concepts in your notes were right, missed, sure-but-wrong, or never tested.
Make it stick
- Guess first. Quiz before you read, read, quiz again. Pretesting primes what matters.
- Come back to it. Export what you missed for Anki, or add 1, 3 and 7-day reminders to your calendar.
- Your trend. Felt and knew per run, kept only in your browser.
- Share the shape. A small card with your boxes and dots, none of your notes.
Based on research on retrieval practice
Recall First doesn't claim to raise your grades. It applies findings that are well replicated, and it is honest about what it measures.
- Pan & Rickard, 2018Meta-analysis of 122 experiments: practice testing beats rereading, and the benefit transfers to new questions.
- Rowland, 2014Recall tests help more than recognition tests. Typing an answer beats picking one.
- Barenberg & Dutke, 2018After retrieval practice, students' confidence judgments become more accurate and less biased.
- Bastani et al., 2025Students with free access to an AI that gave answers did worse once it was removed; hints instead of answers reduced the harm.
Small, open, and private by design
One Node server, one AI model, no database. Your notes travel in the request and nowhere else; the only thing kept is your felt-versus-knew score, in your own browser. The AI key lives on the server, never in the page.
The code is open and runs on your laptop with one command, or deploys as-is. It was planned before it was built: the scope, product requirements and technical spec are in the repository next to the code.