Built on how memory works.
anti/agent is built on a few findings from the science of learning that have held up for decades. The research shows that these principles work; it does not prove that any app works, this one included. Here is each one, in plain words, and what anti/agent does with it.
01Retrieval practice
Answering beats rereading.
What the research found
Pulling an idea out of memory strengthens it far more than reading it again. Rereading feels productive because the text looks familiar; recalling it is what makes it stay.
What anti/agent does
Every session is answering, never rereading. Even the first meeting with a new idea starts with a guess.
02Spacing
Spacing beats cramming.
What the research found
The same practice, spread over days, is remembered much longer than practice packed into one sitting. The best gap grows as the memory gets stronger.
What anti/agent does
Each idea has its own schedule, worked out by FSRS, the open-source scheduler Anki adopted in 2023. An idea comes back when you are about to lose it, never on a fixed calendar.
03Interleaving
Mixing related ideas beats one at a time.
What the research found
Practising look-alike ideas side by side teaches you to tell them apart, which practising each one alone does not.
What anti/agent does
A session mixes the ideas of one chapter, and some cards ask which of two close ideas a description belongs to.
04Self-explanation
Explaining beats recognising.
What the research found
Producing an answer, and explaining why it holds, builds knowledge you can use in a new situation. Picking the right option from a list builds much less of it.
What anti/agent does
Questions move from recognising to recalling to explaining as you know an idea better. What a form can prove limits what it can count: no number of right choices alone marks an idea as known.
05Desirable difficulties
A little difficulty helps.
What the research found
Conditions that make practice feel harder, like guessing before being taught or answering after a gap, often make learning last, as long as you can still succeed.
What anti/agent does
You guess before you are taught. A miss comes back once, as an easier question of another kind, so a hard moment never becomes a wall.
06Learning progress
Practise where you are making progress.
What the research found
Learners progress fastest on material just past what they know, and a tutor that follows where progress is happening does better than a fixed sequence.
What anti/agent does
Ideas open in the order they build on each other, and each session picks the chapter where you are learning the most.
07Knowledge tracing
Knowing is not a feeling.
What the research found
Answering right several times in one sitting shows that something is recent, not that it is learned. A model of what you know has to be updated from evidence, and cautiously.
What anti/agent does
An idea counts as known only after right answers on two different days. Answers given the same day count for less, and when a grader cannot tell whether you were right, nothing is recorded.
What the AI does, and what it does not.
The AI reads your sources, finds the ideas, puts them in order and writes the questions. It does not decide what you practise or when: the engine does, with the rules above, and it is the same code on every device. And it does not teach from its own knowledge: every fact comes from your sources and opens the passage it came from.