The brain-training industry sells a clean story: practise the games, get smarter. The uncomfortable finding that keeps recurring is far narrower — people mostly get better at the trained task, and that improvement travels poorly to anything else. If you want to be honest with yourself about improvement, start by separating three different things.
Three things that look identical on a score sheet
- Test familiarity: you have seen this kind of question before and lost less time working out what it wants.
- State: you slept, you are not stressed, you were not interrupted, you cared about doing well.
- Underlying capability: the thing everyone actually means, and the slowest of the three to shift.
The distinction shows up clearly in ordinary skills. Someone who spends a month on crossword puzzles gets noticeably faster at crosswords: they learn the setters' habits, the recurring abbreviations, the shape of a hidden-word clue. They have not become better at reading contracts or debugging code. That is task learning, and it is the honest description of what most brain-training results show.
Retake almost any reasoning test and your score usually rises. That rise is mostly the first two. It is real in the sense that the number changed; it is not evidence of the third.
What is worth your attention
- Sleep. Short sleep degrades attention and working memory quickly and reliably, and no training programme compensates for it.
- Physical activity and general health — unglamorous, and better supported than most products sold on the promise.
- Deep learning in something real: a language, an instrument, mathematics, a craft. You become genuinely more capable in that domain, which is worth more than an abstract score.
- Test conditions, if a score matters to you: one sitting, no interruptions, not at 2am.
Most people already have first-hand evidence for the sleep point. Compare how you handle a complicated week at work after two solid nights and after a night of four hours with a sick child. The difference in how much you can hold in your head is obvious, immediate, and larger than anything a puzzle app will give you. It is also the cheapest intervention available and the one nobody sells, because there is nothing to sell.
Telling a real change from noise
This is where a confidence interval earns its keep. If your first result was 104-118 and your second is 108-122, nothing has been demonstrated — the ranges overlap almost completely, and the difference sits comfortably inside measurement error. A change only starts to be interesting when ranges barely overlap, it holds across several attempts, and it shows up in a specific domain rather than drifting everywhere at once.
Runners hit the same trap with a single fast parkrun on a cool morning with a tailwind. One result is an event; a trend across months, on comparable courses, is a fitness change. Treat your score history the same way, and notice that a genuine domain-specific shift — say, verbal reasoning climbing after a year of serious reading while the other axes stay put — is far more believable than every axis rising at once a week later.
Which is also the case against chasing your own score. Retake often enough and you will nudge the number upward through familiarity while learning nothing about yourself. Your first attempt, taken properly, is usually the most informative one you will ever have.