Robert and Elizabeth Bjork's central insight is that the conditions producing the best performance during study are often not the conditions producing the best learning. Some difficulties help. Knowing which ones is the difference between a study tool that works and one that merely feels nice.
At a glance
| Desirable difficulty | Why it helps | Where you meet it |
|---|---|---|
| Spacing practice out | Forces retrieval from a colder memory | Widening review gaps |
| Retrieving rather than rereading | The attempt strengthens the trace | Answering before revealing |
| Mixing topics | You must work out which method applies | An interleaved review queue |
| Varying the conditions | Memory generalises rather than binding to context | Reviewing in different places |
| Generating before being told | An attempt, even a wrong one, prepares the answer | The tutor asking first |
Learning is not performance
The distinction underneath everything else: performance is how well someone does during or immediately after study; learning is what remains available later.
They come apart constantly, and in both directions. A student who re-reads a chapter performs well on an immediate quiz and poorly a week later. A student who struggled through recall attempts performs worse immediately and better a week later.
This matters because performance is the only thing anyone can see. A child who has just re-read their notes feels ready. A parent watching a smooth study session sees productivity. Both are reading a signal that does not predict what it appears to predict.
Which difficulties are desirable
The Bjorks' framing — set out in their 2011 chapter Making things hard on yourself, but in a good way — is that difficulty helps when it engages the processes that produce durable memory, and hurts when it simply blocks them.
Desirable: having to retrieve rather than recognise. Waiting long enough that recall is uncertain. Mixing topics so you must identify the right approach. Attempting an answer before being told it.
Not desirable: material pitched so far beyond a child that they have no basis for an attempt. A question that is unclear rather than hard. A card written so badly it cannot be answered honestly. Anxiety, distraction, exhaustion.
The dividing line is roughly: does the child have something to bring? A card they can almost recall is productive struggle. A card about a topic never taught is just failure, and failure without traction teaches nothing except that the subject is not for them.
That distinction is doing real work in how a tutoring product should behave. It is the reason a tutor should hint rather than answer — a hint restores traction while preserving the effort — and the reason cards are only made about material a child has actually covered in a lesson.
How this shapes the flashcard loop
Several design choices are direct applications.
The answer is hidden until your child commits. If the answer is visible, there is no retrieval, and the whole exercise collapses into reading. The reveal is a deliberate action.
The gaps widen. A card reviewed while still fresh is easy and teaches little. The schedule pushes each review toward the edge of forgetting, which is uncomfortable and correct.
Tricky is not punished. Since productive struggle looks like partial failure, a system that penalises failure penalises the mechanism. Nothing is deducted for a Tricky, and it is not reported to you as a score — because a child who feels judged for struggling will start avoiding the struggle.
The queue mixes topics. Consecutive cards about the same idea let the first prime the second, removing the difficulty that was doing the work. So the queue interleaves across decks and shuffles within them.
Where we deliberately reduce difficulty
Not all friction is virtuous, and a product that treats every obstacle as a desirable difficulty is just badly made. Three places we remove it on purpose:
Card creation. Writing good cards is genuinely hard, and it is difficulty of the undesirable kind — it blocks the learning rather than constituting it. So the tutor writes them.
The daily queue. Twenty cards, always finishable. An unbounded backlog is not a desirable difficulty; it is a reason to stop.
Settings. No ease factors or interval modifiers for a child to tune. Choosing scheduler parameters is not learning, and getting it wrong has consequences a child cannot diagnose.
The test we try to apply: is this difficulty part of the thinking, or is it in the way of the thinking? Retrieval is the thinking. Formatting a card is in the way.
The uncomfortable consequence
If desirable difficulties are real, then a study product optimised purely for how pleasant it feels is optimised against its own purpose. Every review made easier, every hint given sooner, every gap shortened to avoid discomfort trades durable learning for a better session.
That does not license making things gratuitously unpleasant — it means the difficulty has to be the right kind, and everything else should be as smooth as possible. Hard where it counts, frictionless where it doesn't.
FAQ
What is a desirable difficulty?
A condition that makes learning feel harder in the moment but produces more durable memory — such as spacing practice out, mixing topics, or recalling rather than rereading.
Isn't struggling just discouraging for a child?
Undesirable difficulty is. The distinction is whether the effort is engaging the learning process or blocking it — a card you can almost recall is productive; a card you have no basis for answering is not.
What is the difference between learning and performance?
Performance is how well someone does right now; learning is what remains later. They routinely come apart, which is why a study session that goes smoothly can produce less durable knowledge than one that felt shaky.