Trained ability
Matrix reasoning
Spotting the pattern in a grid, then working out which piece is missing.
Train three kinds of thinking. Then check whether it worked.
Most brain trainers give you one exercise and a score that only goes up. This one gives you fourteen, across memory, reasoning and speed — and five tests you never practise, so the progress you see is your own.
No account. No ads. Works offline.
記憶kioku
Memory
Holding several things in your head and keeping them current while more keeps arriving.
論理ronri
Reasoning
Working out what follows from what you know, and stopping there.
速度sokudo
Speed
Taking in more of what's in front of you, in less time.
Every exercise finds your level in the first few minutes, then moves with you as you improve. You are never bored, and never drowning.
違いThe differencechigai
You get told the truth about your own progress.
Most apps test you on the game you just played, so the number always rises — whether or not anything underneath it did. Here, training and testing are kept completely apart. Five tests you never practise, at a difficulty that never changes, taken before you start and again after each real stretch of training. So the trend you are looking at is about you, not about how good you have got at one particular game.
独自What's differentdokuji
There isn't another one like this
- Three kinds of thinking, fourteen exercises, one engine. Memory, reasoning and speed. Most trainers pick one and stop there, or hand you thirty small games that are fun for a week. The breadth is the whole idea.
- It fits you within minutes. Nothing to configure and no level to guess at. It watches how you are doing, settles on a difficulty you can just about hold, and keeps moving it as you get better.
- Every puzzle has exactly one right answer, and we prove it. A second program solves each one on its own before it ever reaches you. You will never lose a point to a question that was not fair.
- You cannot win by reading the pictures. An automatic check hunts down puzzles that accidentally give the answer away, so what improves is your thinking rather than your eye for a tell.
- Five tests you never practise — and one of them we deliberately do not train, so you can tell real change from simply getting good at tests. It is the opposite of what a trainer is normally built to do.
- Encouragement you actually earned. Applause when the difficulty really rises, a streak that forgives one day a week, and a picture of how far you have come. Every number on the screen is something you did.
- Your training lives on your phone. No account, nothing to sign up for, and it works with the aeroplane mode on.
We built the most complete trainer we knew how to build, and then built the instruments to check whether it actually works. You get both — and the second one is the rare part.
正直Honestyshōjiki
You won't have to take our word for any of it
Plenty of people who train seriously say it changed something real for them. The controlled trials mostly haven't managed to show it — though most of them ran a few weeks, on students with no reason to push. The honest answer is that it's unresolved, and anyone selling you certainty either way is selling you something.
So we did the useful thing instead of the loud one: we built the measuring instruments into the app and handed them to you. Take the tests before you start, train, take them again, and read your own before-and-after. You'll get better at these exercises — that part is certain. Whether it reaches further is something you'll be able to see for yourself.
測定Measurementsokutei
The part nothing else in this category has.
The five tests
Four of them measure the kind of thinking the training goes after: spotting a pattern in a grid, working out what comes next in a sequence, holding several facts in your head at once, and remembering under pressure. The difficulty never changes, and you never practise them.
Trained ability
Figural series
Five shapes in a row, each following a rule. Work out what the sixth one has to be.
Trained ability
Relational integration
Holding several facts in your head at once and working out what they add up to.
Trained ability
Symmetry span
Remembering a sequence while something else keeps interrupting you.
対照The control · taishō
3D mental rotation
Imagining an object turned around. We deliberately don't train this one.
If it climbs as much as the other four, then you didn't get better at reasoning. You got better at taking tests. Most apps would rather you never found that out.
Why we picked an ability we're least likely to improve →鍛錬Trainingtanren
Fourteen exercises, and there's no trick to learn
Every problem is built fresh each time, so there's nothing to memorise, and each family closes a different shortcut.
記憶Memorykioku
Hold, update, compare — under time pressure.
Relational n-backJudge whether the current relation matches the relation n steps back.
Premise n-backDoes this premise share its structure with the one n back?
Interrupted memoryHold a sequence while something keeps interrupting.
Tonal spanHear a run of tones, then tap them back in order.
論理Reasoningronri
One skill, many relations — each family is its own ladder.
Hidden shapesWhich of these four patches is hidden in the field?
Order chainsOrder by size, by when, or by how long.
Same or oppositeTrack same-as and opposite-of through a chain.
Directions in 3DCompass plus above and below.
SyllogismsAll, some, no, and what actually follows.
Logical connectivesIf-then, or, and their traps.
Mixed relationsTwo relation kinds, one chain of reasoning.
Alternating reasoningThe kind of puzzle changes from one to the next. Work out which rules apply, then apply them.
速度Speedsokudo
Less time, more of the field.
Sweep pairsTwo quick slides in a row. Did each one go up or down?
Split-second glanceIdentify the centre and place the edge — from one masked flash.
工夫The designkufū
Why it keeps working.
01
It fits you, not the other way round
There is no level to pick and nothing to set up. The app finds a difficulty you can just about hold — right at the edge of comfortable — and keeps nudging it as you improve. Beginners and old hands get the same treatment.
02
Every puzzle is fair, and that is checked
A second program solves each puzzle before you see it. If more than one answer holds up, the puzzle is thrown away and rebuilt. Every wrong answer is provably wrong, not just different from the one we had in mind.
03
It never turns into a routine
Difficulty here is not one slider. Twenty-three separate things can get harder — how much you hold, how fast it arrives, how convincing the wrong answers are — and only one or two move in any session. A trick that carried you last week quietly stops working.
秘匿Your datahitoku
A score for the round you just played, and the sum that produced it
Every session ends with a number, and with the arithmetic written out underneath: how many you got right, what pure guessing would have scored on that exercise, how far up you climbed from one to the other, and the multiplier for the difficulty you played at. You can check it on a calculator, and we think you should be able to.
It’s a game score and it says so. It isn’t a measurement of you — that’s what the five tests are for, and we keep the two apart on purpose.
There’s also a self-check in Settings, which reads your own stored sessions and tells you whether the machinery is doing its job: whether the difficulty really climbs when you clear 85%, whether it’s mixing up the exercises and the dials, whether interruptions are being spotted, and when your next test is due. It reads and never writes, so running it can’t change anything. Where it hasn’t got enough data yet, it says so rather than showing you a reassuring tick.
What keeps you coming back
Training is hard, and hard things need more than willpower. So there is real encouragement here — and every piece of it is something you actually did.
Applause when you level up. Not for showing up, and not for a streak: for the difficulty genuinely rising because you cleared the band. There is a rarer version when an exercise goes further than you have ever taken it, and it says so differently, because that one means more.
A streak that forgives you. One missed day inside any seven does not end it. A rule that deletes six weeks over one bad Tuesday is not motivating, it is just unkind. The skipped day is not added to the count either — it bridges the run rather than padding it, so the number stays honest.
One reminder a day, if you want one. At a time you choose. Never a second one, silent on a day you have already trained, and it switches itself off after a fortnight of being ignored. Nothing in it about losing anything — it is off until you ask for it.
A picture of how far you have come. This is the one people are most surprised by. Because the app holds you near 80% correct forever, every session feels exactly as hard as your first, and your own improvement is invisible from the inside. So each exercise shows the distance travelled in plain units: you began holding 2 back. You now hold 5.
And the folding. 鍛錬 is a metalworking word — heat the steel, hammer it, fold it back on itself, repeat. Every fold doubles the layers, and ten folds is the thousand-layer figure smiths quote. Here a fold costs two hours of real training, so the metal deepens over months. It cannot be hurried, opening the app earns nothing, and it ends at ten. Then there is just the training.
Where your data lives
On your phone. There is no account, nothing to sign up for, and the app works with the aeroplane mode on — which is also the honest way to try it, since nothing about your history depends on us.
Export it as JSON whenever you like. Back it up encrypted through your phone’s own file picker, no cloud account needed. On Android it backs up by itself as you train, keeps the last five, and tells you loudly if it ever stops.
Online features — a leaderboard, and pairing up with someone else training — are being built. When they arrive they will be optional, and what they send is written down in the privacy policy before either one goes live.
記録Noteskiroku
8 min
Five ways the puzzles leaked their own answers
An automated audit measures how much each surface feature of a problem reveals about its answer, and fails the build past 0.05 bits. It has caught five real leaks — including one that printed the answer on the item.
6 min
One dial and one number
Dual n-back gives you a single difficulty knob and a single metric. Matching a relation rather than an item changes what the task demands — a structure to update instead of a list to hold.
Find out for yourself
The tests are in the app. Take them before you start, then take them again in a month.