Eleven exercises built so they cannot be shortcut, and a test battery you never train on — including one measuring an ability the app deliberately leaves alone. That last part is what lets you tell a real change from simply getting better at tests.
No account. No ads. No trackers. Works with the network off.
Why this one is different
01
Every problem is judged by an independent solver that was written and hand-tested before the generator existed and imports nothing from it. An item ships only if that solver confirms exactly one answer is correct. A generator that grades its own homework will eventually pass a broken item.
02
An audit measures how much each surface feature of a problem — premise count, negations, term positions — leaks about its answer, and the build fails if any single feature carries more than 0.05 bits. It has caught five real leaks so far, including one where the wording of the answer was effectively printed on the item.
03
The staircase adapts across nine dimensions and rotates which one it pushes. Practice a strategy that beats the axis currently under load, and the axis changes. Item count is not one of the axes, because more premises scales reading, not reasoning.
04
Four tests at fixed difficulty, taken every four weeks. They never feed the adaptive engine and the training never touches them. Your training level measures the staircase as much as it measures you; these do not.
05
One of those four tests measures 3D mental rotation — an ability this app deliberately does not train. If it rises as much as everything else, what improved was your test-taking. If the trained measures move and this one does not, that is the pattern worth believing.
06
No streak to protect, no daily goal, no notifications begging you back, no single “brain score” that only ever goes up. Those features exist to make you open an app. They are not training, and some of them actively corrupt the measurement.
The exercises
Two working-memory tasks that ask about the relation between items rather than the items themselves, and nine reasoning families that each break a different shortcut. Every problem is generated fresh from a seed, so there is no item bank to memorise.
No brain-training program has been shown to reliably raise general intelligence. Not this one, and not the ones with the advertising budgets. Improvements on a trained task are easy to produce and easy to sell; showing that they carry over to anything you actually care about is the part that decades of research keeps failing to establish.
So Tanren Sensei does not ask you to believe a transfer claim. It gives you the instruments to check one, on yourself, and reports what they say — trend lines per test, no composite, no conversion into a score with a familiar name. If the honest answer for you turns out to be “nothing moved”, the app is built to show you that clearly rather than bury it.
What the evidence does and does not support →
Most cognitive training gets easier the longer you do it, because what you learn is the task rather than the skill. You find the chunking trick, the answer-shaped distractor, the position the answer tends to sit in — and your score climbs while the underlying ability sits still.
Every design decision here is aimed at that failure. Item identity is never the answer in the n-back tasks, so chunking has nothing to grip. Premises arrive scrambled out of chain order, so the reasoning is never handed to you pre-assembled. Distractors are wrong on a different axis than the one under test, so spotting the negation buys nothing. And a statistical audit runs on every build to catch the shortcuts nobody thought of.
Training, results and history are computed and stored on your device. Today the app makes no network requests at all — the two optional online features are built but switched off in every shipping build, so there is nothing yet to opt into. That is not a policy promise sitting in a document; it is held up by tests that fail the build if any file outside one specific network client can reach the network.
The app ships with the test battery built in. Take it before you start, take it again in a month, and decide for yourself whether anything moved.