Heuristic
Make difficulty earn its keep
Harder is useful only when the extra effort serves the learning process you actually need.
When it fits
- A practice method feels harder and you are tempted to call that difficulty evidence that the training is more effective.
When to avoid it
- The cited reviews synthesize learning theories and domain evidence; they do not supply a universal optimal failure rate or difficulty percentage.
Why it matters
Treat difficulty as a designed intervention, not a virtue. Ask what process the challenge forces: retrieval, discrimination, generation, transfer or something else. Then check delayed retention or transfer rather than celebrating slow, frustrating practice. Reviews of desirable difficulties warn explicitly that perceived effort or disfluency should not be confused with benefit, and that complexity and learner expertise change when added difficulty helps versus overloads.
An example
Removing the answer and requiring retrieval can create useful effort. Making the font tiny, instructions ambiguous or the interface annoying creates difficulty without necessarily training the target skill.
Check your result
You can name the learning mechanism the difficulty is meant to activate and a delayed or transfer measure that would justify keeping it.
Keep this limit in mind
- The cited reviews synthesize learning theories and domain evidence; they do not supply a universal optimal failure rate or difficulty percentage.
Connected ideas
Useful withSeparate better monitoring from better performance
Evidence and sources
A 2026 review concludes that testing, spacing, interleaving and productive-failure approaches can function as desirable difficulties, while warning that greater perceived difficulty or disfluency should not itself be treated as evidence of better learning.
The review synthesizes multiple literatures and does not provide one scalar difficulty target for individual learners or tasks.
A 2024 review argues that the effect of added difficulty depends on material complexity and learner expertise: difficulty can aid retrieval-based learning but excess extraneous load can hinder schema formation.
The proposed integrated model is theoretical and should be used as a design boundary, not a validated personalization algorithm.