How it decides

92 questions in the bank, of which between 40 and 80 get asked. Which ones and how many depends on what you answer. This is what happens underneath.

Four layers, always in this order

First it works out where you lean, and only then decides what else is worth asking.

Interests 30 questions
Holland's six types. These all get asked: they are the skeleton.
Aptitudes 30 questions
What you believe you are good at. Only the ones that fit your three highest interests.
Values 18 questions
What you ask of a job. Each turn goes to whichever layer is worst covered.
Context 14 questions
How long you can study, whether you can move, whether you can take the risk. These break the ties.

Before every question, the engine keeps the one that best separates your five closest fits. That is why no two people take the same test.

Before asking

Five candidates nearly tied: nothing is known.

After

A good question is one that splits the pack.

It matches the shape of your profile against each vocation. The four layers do not weigh the same.

Interests50%Aptitudes18%Values18%Context14%

Interests are worth as much as the other three layers put together. Arguable, and on show on purpose.

Being alike is not enough: a vocation can run head-on into something you said clearly.

You, 65 or moreThe vocation, 25 or less
Something you care about counts for nothing there.
You, 35 or lessThe vocation, 65 or more
Something you ruled out is essential there.

There is no fixed number. It stops once your ranking settles, and at 80 answers at the latest.

At least 40 answers: A hard floor, whatever happens.
Something from all four layers: So the result cannot lean on a single one.
70% confidence: A flat profile never gets there, however many questions you answer.
A stable ranking: Your top 5 stops moving for several answers in a row.

It does not work out what you are good at: There are no exercises to grade. It measures what you say about yourself, which is not your performance — even if that is what really moves a decision.
It does not invent a result when there is none: If your answers come out too flat, the test says so instead of picking a winner at random.
It does not keep who you are: No name, no email, no account. Results stay on your device and nothing travels at all — not even an anonymous identifier.
That 92% is not a probability: The app rescales the best result to 92 when presenting it, keeping the distances. It is a screen convention. If you want the raw affinities, the finder does not rescale them.

The only question left is which of the 101 is yours.

Free on Google Play.