Every block puts three statements against each other that cost exactly the same to admit — three strengths, or three things that wear you down, never a mix. There's no better version of yourself to reach for, which leaves only the question of which is truer. It opens with twenty-four plainer questions placing you on the four dichotomies that carry most of the signal.
Answer quickly and don't work through the details. Long deliberation gives you time to reason about what your believed type would say, which is the main way results go wrong. When a block offers nothing you'd claim, rank it anyway — the near-ties carry most of the information.
Best taken twice, a week or two apart, and pooled at the end. That was worth more in testing than any change to the questions — you vary more than the test does. Where two sittings disagree, the disagreement is itself the finding.
This is a socionics test. It estimates which Model A type fits you, tells you what that estimate is worth, and says so when the honest answer is that it can't tell.
Four dichotomies carry most of the signal, and the test measures them in socionics terms rather than the MBTI ones they are usually confused with. The distinction matters more than it sounds.
Extraversion here is not about social energy. Following Jung, it is whether your attention rests on objects themselves or on the relations between them. A socionics introvert can be highly sociable, and asking whether parties tire you out measures a different trait entirely.
Thinking and Feeling is not agreeableness. It is which information you process more readily — how something works, or where people stand with one another. Logical types can be warm and ethical ones harsh; the question is what you notice first.
Intuition against Sensing is potential against the concrete. Rational against irrational is whether a formed judgment drives your action or your perception stays open — not whether you keep a tidy schedule.
An earlier version asked the MBTI-style questions. Testing showed that cost eight to twenty points of accuracy, and made the test overconfident — the answers were evidence about a neighbouring trait, presented as evidence about this one.
Twenty-four agree-or-disagree items place you on those axes. Then twenty-four blocks of three statements, where you mark the one most like you and the one least. Every block is matched so no option is more flattering — three strengths, or three things that wear you down, never a mix. Eight further items check your consistency and how flatteringly you are answering, and both are reported with the result. Fifty-six in all, about eighteen minutes.
You get your type and its full Model A stack, with each of the eight positions described specifically — not what a blind spot is in general, but what it means that this particular element is yours. Also your quadra and club, your position on each axis shown continuously rather than as a letter, all sixteen types ranked, and a comparison of the top three showing where they differ.
Sometimes it declines to name a type. If no stack passes seventy per cent confidence, the page leads with your quadra and lists the closest candidates without picking one. That is not a failure: at that confidence a named type is right about sixty per cent of the time, while the quadra reading holds near ninety-seven. Roughly one respondent in eight lands there.
Every result carries two numbers: what the scoring believes, and what that belief is worth once the limits of a questionnaire are applied. The second is always lower, and it is shown rather than applied quietly.
Those limits are real. A questionnaire reaches your self-image, not your behaviour, and socionics has no external standard to check it against — no diagnosis, no outcome, nothing a result can be scored against later. So a ceiling is built in: the highest anyone can score for an exact type is seventy-two per cent, and eighty-six for quadra.
The page also says where its own percentages come from: simulated respondents, generated using the same assumptions the scoring uses to read them. They show that the arithmetic works, and nothing more. No real respondent has taken this test and been checked against anything.
Retest instability is the largest real source of error here, and pooling attacks it directly. Every sitting produces a code. Save it, take the test again a fortnight later, paste the first code in at the end, and the two combine — worth about ten points in testing. Further sittings add well under a point each.
You can also pool someone else's reading of you. An informant sees your visible behaviour better than you do, and Model A is largely a theory about behaviour, so the two readings are wrong in different directions. Where a self report and an informant agree, that is the strongest evidence this test can produce.
It is also the only check you can run yourself. Every percentage on the results page comes from simulation. Two sittings that disagree is a real finding about you, and worth more than either answer alone.
Free, no ads, no account, nothing stored — see for yourself. If you want to support it, buy me a coffee.