SEMIDEUS
Log inStart free

How we grade

Nothing moves because you said so.

Here is what grades you, what it compares you against, and what a grade does not mean. If something on this page stops being true, it is a bug.

  1. Explaining an idea

    Graded by Claude (currently Sonnet 4.6) against a fixed rubric.

    Compared against: The lesson built from your material when one is attached. Otherwise, the concept itself.

    Accuracy
    Are the claims correct and consistent?
    30%
    Coverage
    Did you address the ideas that matter, in a sensible order?
    25%
    Depth
    Do you explain why and how, not only what?
    25%
    Transfer
    Can you apply it, give your own example, or build an analogy?
    20%

    Each part is scored from 1 to 5. You get the gaps by name, any contradictions, and follow-up questions.

  2. The Socratic tutor

    It only asks questions and never gives you the answer. When the session ends, your reasoning is graded against a rubric built on the Paul-Elder and Facione critical-thinking standards.

    Compared against: Your own reasoning during the session, and your own source material.

    Logic
    Do your conclusions follow from your reasons?
    25%
    Evidence
    Are your claims grounded in your sources?
    20%
    Humility
    Do you separate what you know from what you assume, and update when you should?
    20%
    Fairness
    Do you take on the strongest version of the other view?
    15%
    Open-mindedness
    Do you consider other explanations?
    10%
    Clarity
    Can someone follow what you mean?
    10%

    The numbers are not shown to you during or after the session. Your dashboard shows whether your reasoning held or is still settling.

  3. Teaching Daimon

    You teach Daimon, our teachable AI. It then takes a short quiz, and you predict its score first.

    Compared against: How well your teaching let it answer.

    Pasting the source text back scores zero, with no model call at all.

  4. Rebuilding from memory

    You name the ideas your own map links to a concept.

    Compared against: Your own map of ideas, by direct comparison. No model is involved.

    It counts at half the weight of an explanation, and it can never finish or seal an idea on its own.

  5. Tests

    Questions generated from your sources, in seven formats, from fill-in-the-blank to explaining why.

    Compared against: A weighted rubric for each question, not an answer key you could memorize.

    You see a score and feedback for every question.

  6. Spaced review

    Cards come back on a schedule set by FSRS, the open algorithm also used in Anki. You rate how well you recalled each one.

    Compared against: Your own recall, over days and weeks.

    Review keeps memory alive and moves an idea's estimate. It can never finish or seal an idea.

Finishing and sealing an idea

  • An idea is finished when you show you understand it: an explanation, a Socratic session, teaching Daimon or a test, on top of solid recall.
  • It is sealed when you prove it again at least 72 hours later, with an explanation, a Socratic session or by teaching Daimon. The database enforces the gap.
  • The seal mints a Proof of Understanding signed with Ed25519. It is private until you share it, and anyone can check the signature against our published keys.
Published keys

What a grade does not mean

  • It is not a proctored exam. Nobody watches you take it.
  • AI grading can be wrong. Our graders have not yet been checked against human raters, so treat a grade as a strong signal, not a final word.
  • A record attests two graded performances, days apart. It does not attest identity beyond the name given, closed-book conditions, or that the memory will last.
  • We never claim a better brain. We measure what you understand.
Start free

Free plan. No card needed.