Loading
Learning science, product design, method notes, safety essays. Every Brainback decision is downstream of one of these — published in the open.
Brainback is an opinionated reading of a century of work on how memory behaves. The opinions are ours; the findings are not, and each one is named so you can go and argue with the source.
Students who read a passage once and were tested remembered more a week later than students who read it twice. Retrieval is not measurement of learning; it is the learning.
in the product: Nothing happens until you attempt. The gate is in the API, not the prompt.
Conditions that slow acquisition down — spacing, interleaving, generating rather than reading — improve long-term retention. Easy practice feels better and works worse.
in the product: Hints cost points, and the ladder makes you climb instead of jump.
Memory for new material decays roughly exponentially, and each successful review flattens the curve. The interval matters more than the hour.
in the product: Every wrong attempt becomes a card, scheduled by SM-2 rather than by mood.
Material you produce yourself is remembered better than identical material you read — even when your production was wrong and then corrected.
in the product: A wrong attempt is worth more here than a right answer handed over.
Learners are poor judges of what they know, and the gap between confidence and accuracy predicts exam surprises better than either number alone.
in the product: You commit to a confidence before you see the diagnosis; 30% of the Brain score is that gap.
The 100+ years of research that make Brainback's design non-optional.
Handing a student the answer skips exactly the moment learning happens.
ReadBjork's principle — and why productive struggle is the whole point.
ReadOne of the most robust findings in learning science, quietly built into every recall review.
ReadOne creates capability. The other creates the feeling of capability.
ReadWhy refusing to give the answer is the product's whole design brief.
A single hint is a random walk. A ladder is a curriculum.
ReadOne number, three components, weekly cadence.
ReadEfficiency mode is normal AI. Except when it isn't.
ReadThe privacy stance behind the parent dashboard.
ReadThe economics behind a product that could technically be free.
ReadA brief history of student-facing AI and what it optimized for.
ReadThe right kind of frustration is a feature. Here's how we tell them apart.
ReadWhy the tutor ends every reply with exactly one guiding question.
ReadWhy we generate recall cards from your errors, not from pre-made decks.
ReadHow Brainback actually implements the theory, line by line.
What Socrates got right — and what needed a system prompt.
ReadSM-2 is 40 years old and still the right default. Here's why.
ReadKnowing what you don't know is 30% of your Brain score. Here's why.
ReadYou can recognize a concept without knowing it. Explaining it back is where the gap surfaces.
ReadCard generation, review UX, and the edge cases every implementer meets.
ReadEvery attempt asks the student their confidence. Here's what happens next.
ReadSystem prompts stay warm. Per-problem context stays warm. Cost stays down.
ReadWhy the minimum is twenty characters, and what we would have to run to know.
ReadOne paragraph, delivered warmly, thousands of times.
ReadGuardrails, output guards, adversarial resistance.
What the numbers say — from the published literature, cited.
→the essays are downstream
Every one of these turns into a line of code inside Brainback.