How to practise so that it actually sticks
Two hundred puzzles done quickly teach less than thirty done properly.
It is possible to do a great deal of practice on reasoning puzzles and get almost no better. It happens constantly, and it is discouraging in a particular way, because effort is clearly going in.
The problem is almost never the amount. It is what happens in the seconds after each puzzle.
The thing that makes practice work
The skill being built is rule recognition — seeing a grid and knowing immediately what family of rule is running. You build that by naming rules, and by nothing else.
So the critical moment is not solving the puzzle. It is the five seconds afterwards, when you either say what the rule was or move straight to the next one. Practice without naming is exposure, and exposure is a weak teacher.
A session that works
- Ten puzzles, no timer. Fewer than you think, and slower than feels productive.
- Write the rule before checking. One sentence. 'Count goes up by one, shape constant down columns, shading alternates.'
- Check the answer.
- For wrong answers, find the failure point. Not the correct answer — the exact step where your rule broke. Did you miss an attribute? Misread a cell? Assume a progression that was a cycle?
- For right answers, check your rule was the real rule. A right answer from a wrong rule is a warning, not a success.
Ten puzzles done like that takes about twenty-five minutes and teaches more than a hundred done at speed. It is also considerably less pleasant, which is the main reason people do not do it.
Keep an error log
Write down every mistake with its cause. After twenty entries a pattern appears, and it is always narrower than expected — most people have two or three recurring failure modes, not twenty.
- 'Missed the shading rule' appearing six times means run shading as its own pass, always
- 'Assumed progression, was a cycle' appearing four times means check the options before committing on that specific question type
- 'Picked the mirror image' appearing five times means check handedness before finalising any rotation answer
A list of three personal failure modes is worth more than any general advice, including everything on this page, because it is about you rather than about the average solver.
Log the near misses too. A puzzle you got right but were unsure about is the same weakness as one you got wrong — it just happened to land well this time. Those are the ones that quietly cost marks under real time pressure.
Spacing beats cramming, by a lot
Twenty minutes a day for a fortnight produces substantially more improvement than four and a half hours in one Sunday, even though the total time is similar.
The effect is well established across almost every kind of learning. Recognition needs consolidation between sessions; a long single session mostly produces fatigue, and fatigued practice rehearses sloppy habits.
When to introduce the timer
Only after naming rules has become automatic. Adding time pressure early consumes working memory on clock-watching, pushes you toward guessing, and rehearses exactly the habits you are trying to replace.
The sequence that works: accuracy without time, then accuracy with a generous time limit, then realistic timing. Most people invert this, start with realistic timing, and plateau immediately.
Variety matters more than volume
Doing a hundred matrix puzzles builds matrix-puzzle recognition and little else. Mixing matrices, sequences, odd-one-out, analogies and spatial questions in the same session is harder and transfers much better, because you also practise identifying which kind of problem you are looking at — which on a real mixed paper is the first decision you have to make and the one nobody practises.
What good progress actually looks like
Not a smooth climb. Typically: a fast early gain as you learn the rule families, then a long flat stretch that feels like failure, then another step up when recognition becomes automatic.
The flat stretch is where most people conclude they have hit their ceiling and stop. It is consolidation, and it is the part that makes the next step possible. If you are naming rules and logging errors, you are improving during it even when the scores do not move.