Key takeaways
- Pick the level where you get most answers right but not all of them. As a working rule of ours — not a research constant — aim for roughly four correct in five, and move when you are reliably above that.
- Practice that feels comfortable is usually practice you have outgrown. The Bjork lab's work on desirable difficulties describes conditions that depress performance during training while improving what you retain afterwards.
- A difficulty label is not a score multiplier; it changes the task. Knowing which property it changes tells you whether moving up trains the thing you actually want.
- In arithmetic modes the ladder usually adds digits. In memory modes it changes several things at once — item count, grid size and how long the reveal lasts — so a single step up can be a much bigger jump than it looks.
- Adaptive settings adjust within a level rather than across levels, so the floor you choose still decides the worst case.
Pick the level where you get most answers right but not all of them. As a working rule — ours, not a number from research — aim for about four correct in five, and move up when you are reliably above that for two or three sessions in the same mode. That single rule resolves most difficulty questions, and the rest of this article is about why it works and what a difficulty label is actually changing when you move it.
The instinct people bring to practice apps is the opposite. Comfortable settings produce good-looking sessions, and good-looking sessions feel like progress. They mostly are not.
Why comfortable practice underdelivers
The Bjork Learning and Forgetting Lab at UCLA has spent decades on a distinction that matters here: performance during training is not the same as learning. Conditions the lab calls desirable difficulties — spacing, testing, mixing problem types — tend to depress how well you do while practising and improve what you retain afterwards. Easy conditions do the reverse. They feel efficient and leave less behind.
That is uncomfortable advice, because the visible signal in a practice app is exactly the one the research says to distrust. A session at a level you have outgrown produces a high score, a high accuracy figure and fast response times, and none of those tell you whether anything changed.
The four-in-five rule is a practical way of staying on the useful side of that line without needing to measure anything. Enough success to keep going; enough failure that the session has something to teach.
What a difficulty label physically changes
"Hard" is not a marking scheme. In a well-built practice app it is a different task, and knowing which property changed tells you whether moving up trains what you came for.
In the arithmetic modes of IQ Test & Brain Training Games — our own app, built by Reign Creative — the ladders are explicit about it. Multiplication Rush runs tables of 1–9, then 1–12, then two-digit by one-digit, then two-digit by two-digit, then a mixed setting. Each step adds digits, which changes which stage of the calculation breaks first: recall at the bottom of the ladder, holding partial products at the top. Mixed Operations runs Easy, Medium, Hard, Expert and Adaptive, where the change is less about digit count and more about which operations turn up and in what combination.
The memory side moves several properties at once. Its seven levels change how many items appear, how large the grid is, how long the reveal lasts and how long you hold the answer before typing it — Beginner sits at three or four items on a 2×2 grid with a slow reveal, Champion at up to twenty items on grids as large as 7×7 with the fastest reveal. One step up the memory ladder is therefore a bigger jump than one step up an arithmetic ladder, and it is worth expecting that rather than concluding you have gone backwards.
The practical consequence: when you want a harder version of the same skill, move within a mode. When you want a different skill, change mode.
The two directions people get wrong
Too low. The tell is that you finish sessions without noticing them. Accuracy sits at or near the top, response times are flat across weeks, and the score only moves when the timer is generous. This is pleasant and it is maintenance, not training. If you want to stay here deliberately — some weeks that is the right call — at least call it maintenance.
Too high. The tell is guessing. Accuracy falls below roughly half, response times get erratic rather than slow, and you stop being able to say afterwards what you got wrong. A drill you cannot reconstruct teaches nothing, because there is no error to learn from. NRICH, the mathematics education project at the University of Cambridge, frames fluency as efficiency, accuracy and flexibility together — and a level where you have lost the thread is not building any of the three.
The recovery from too high is not always a level down. It is often the same level without a clock, which is a different intervention: the untimed accuracy route keeps the content and removes the pressure. Our mental math practice guide covers which methods are worth having in place before the harder digit counts become reasonable.
What adaptive settings do and do not decide
Adaptive difficulty is often read as "the app will pick for me". It does less than that, and the limits are the useful part.
In this app, adaptive adjustment reacts to the run in progress: three correct answers in a row lengthens the task by one step, two misses in a row shortens it by one, and the whole thing is clamped so a bad patch cannot drop below the floor of the level you chose. It tunes inside a band. It does not choose the band, and it will not rescue a level that was wrong to begin with.
That clamp is also why the initial choice still matters. If you pick Champion on a bad day, adaptive adjustment will make the session survivable but it will not hand you Beginner. Choose the floor deliberately, then let the adjustment handle the within-session variation.
When to move, and what to do instead of moving
Move up when two or three consecutive sessions in the same mode, at the same difficulty, land comfortably above your target with steady response times. One good session is noise.
When the numbers say you should move but the next level is clearly too far — a common gap on the memory ladder — there are two alternatives worth trying before you force it. The first is mixing problem types rather than raising difficulty: The Learning Scientists describe interleaving as switching between ideas while you study, and note that it feels harder than working through one type at a time while being more helpful in the long run. The What Works Clearinghouse practice guide on organising instruction makes a related recommendation about interleaving worked examples with problems to solve. The second is holding the difficulty and tightening the constraint — the same questions with a shorter clock, or with a single mistake ending the run.
How a quiz app decides what counts as hard in the first place is a related and genuinely separate question, covered in how quiz difficulty actually works. For what this whole category of practice does and does not support, our brain training explainer is the evidence-led version. More sits in the brain training games hub and the education and brain category, and the app itself is described on the IQ Test & Brain Training Games app page.
Good to know
Frequently asked questions
What accuracy should I be aiming for in practice?
Why does practice that feels easy stop helping?
Should I use an adaptive setting instead of choosing?
How long should I stay at a level before moving up?
Sources
- Research — Bjork Learning and Forgetting Lab (opens in a new tab)University of California, Los Angeles — accessed
- Organizing Instruction and Study to Improve Student Learning (opens in a new tab)What Works Clearinghouse, Institute of Education Sciences — accessed
- Developing Number Fluency — What, Why and How (opens in a new tab)NRICH, University of Cambridge — accessed
- Learn To Study Using… Interleaving (opens in a new tab)The Learning Scientists — accessed
About the author
Reign Creative LLC — The independent app studio that builds the apps covered on this site and publishes every article here. On Google Play it publishes as Reign Collective Apps.
We use AI tools to help research, draft and edit our articles. Each article lists the sources its factual claims rely on, and details about our apps come from their live Google Play listings. No article on this site has been reviewed by an independent expert unless the reviewer is named on it.
Spotted an error? Tell us — how corrections are handled is set out in our editorial policy.