What it solves
Most students confuse feeling prepared with being prepared. They re-read their notes, watch a video at 2×, ask an LLM to summarise the chapter — and walk into the exam with a fluency that has nothing to do with retrieval. The result is the universal post-exam shock: “I knew this. Why couldn’t I write it?”
Exam Session collapses that gap by making the exam happen — privately, on your own canvas, days before the real one.
How it works
You finish a study session. You open the Atlas menu and tap 🎓 Interrogami — or type “interrogami” into the canvas chat. Fluera reads the concepts you’ve actually drawn on the canvas (cluster cache plus stroke OCR) and generates a closed-book exam from them. Five, seven or ten questions. Mixed types: open-ended, multiple-choice, true/false, formula recall.
Before each answer, the confidence picker asks one thing: how sure are you, one to five? Move the dial. Commit.
Now answer — by hand, on a mini-canvas scratchpad, exactly the way you’d write it on the exam itself. No autocomplete. No copy-paste. No “regenerate response.” Just you and the page.
When you submit, Fluera evaluates the answer token by token, streaming feedback as it reads. If you were wrong and confident, the hypercorrection shock UI fires — a deliberate, hard-to-ignore visual that primes the most durable form of correction the brain can produce. Then Fluera asks you to rewrite the correct answer in your own words: the generation effect locks it in.
After three correct in a row, difficulty boosts automatically. After every chunk of four questions, a short pause for feedback before the next set — chunking and interleaving by design.
When the session ends, the FSRS scheduler updates per-concept review intervals from your actual performance, not your self-reported ease. Blind spots flow back to Fog of War for surgical retrieval practice on Days 1, 3, 7, 14.
The science behind it
Three decades of evidence converge on one finding: the most durable corrections come from the most surprising failures.
Butterfield & Metcalfe’s hypercorrection effect (2001) is the central mechanism. High-confidence errors, once corrected, are retained more permanently than low-confidence ones — the surprise of being wrong when you were sure modulates hippocampal encoding via amygdala arousal. The confidence picker is not aesthetic; it primes this mechanism.
Robert Bjork’s desirable difficulties framework (1994) explains why an open-ended hand-written answer beats multiple choice: the friction of generation is exactly what builds memory. Roediger & Karpicke (2006) round it out — a single retrieval attempt produces stronger long-term retention than four re-reading sessions.
Manu Kapur’s productive failure research (2008) supports the hardest design choice: questions that begin slightly above your current level produce more lasting learning than questions tuned to your comfort. Adaptive difficulty escalates deliberately, not defensively.
What’s coming
- Post-exam analytics — per-cluster confidence calibration and hypercorrection magnitude graphs.
- Voice-mode Exam — spoken questions with handwritten answers, for clinical or oral-exam preparation.
- Persistent surgical path overlay — after Exam, Fog of War highlights only the missed concepts on the live canvas, scoped for next-day review.
- Multi-subject Exam Sessions — one session that interleaves clusters from multiple notebooks, weaponising transfer.