Don't claim you know it.
Prove it.
Semideus Cognition turns any source into understanding you can prove — you teach it back, you test it, and you survive the spaced retrieval. One Mastery Score per topic, and it moves only on proof.
“You don't claim you understand. You teach it back, and the score decides. You don't claim you remember. You survive the spaced retrieval.”
Three channels. None self-graded.
Consumption feels like progress while nothing crosses into you. The loop closes only on typed proof — three signals, each weighted into one number.
Explain it. Get scored.
Teach a concept in your own words. Claude grades it against your source's ground truth — clarity, accuracy, depth, analogy — and names the exact gaps and contradictions. No 'great job.'
Questions from your own sources.
Generated from what you chose to learn and scored question-by-question against a rubric — not an answer key you can game.
Survive the forgetting curve.
A card isn't 'known' because you answered it once. It has to survive 30-, 60-, and 180-day retrieval floors — or it gets pulled back into rotation.
A Socratic tutor that won't let it slide.
State a claim drawn from one of your own sources. The tutor escalates through three postures — Probe → Refute → Falsify — and at the end you get a reasoning score with epistemic humility as the headline signal. Your claim is what gets pressure-tested; the rubric is what gets reported.
Probe
State the claim. Define your terms. Here the tutor is curious, not adversarial.
Refute
Defend the weakest assumption. The tutor presses on the one you'd rather not look at.
Falsify
Withstand the counter-case. What would change your mind — and is it true?
Open /socratic inside any project — bring something you don't get or a claim worth defending, and the tutor walks you there question by question. The transcript and the proof persist; the gaps get cards back into your review queue.
One score. Moves only on proof.
Teach-back, tests, and retention fuse into a single Mastery Score per topic — 50 / 30 / 20. You can't nudge it by hand. There is no “mark as learned” button.
- Computed from proof events, never self-report
- Gaps surfaced in plain language — no academic chips
- Graceful fallbacks when a signal is missing
Any source becomes an interrogable second brain.
Drop a paper, PDF, article, link, or GitHub repo. It's extracted, chunked, and embedded — then distilled into a citation-backed wiki, decks, and tests. Nothing generic: every artifact is grounded in what you chose to learn.
- Projects → paths → modules → decks → wiki
- Every claim traces back to a source passage
A mind that gets better at getting better.
A flashcard app helps you memorize. A meta-learning OS instruments your own learning process — and tunes it.
Cognitive profile
The system models how you learn — analogy domains, prior knowledge, preferred depth — and shapes every deck and test to it.
Spaced retrieval
SM-2 scheduling with 30-, 60-, and 180-day retention floors. A card you saw once isn't 'known' until it survives the long re-probe.
Pay for proof.
Not a bigger library. A higher Mastery Score you can stand behind.
Turn one source into a living wiki and start building durable recall.
- Capture sources & build your wiki
- Spaced-repetition review
- Weekly mastery check-ins
Every plan includes monthly AI credits for teach-back, tests, and Socratic sessions — top up anytime with credit packs, and credits never expire.
Stop consuming. Start proving.
Join the founding cohort proving what they know — instead of collecting another reading list.