One Hundred Questions, Six Sections, One Explanation

A hundred questions is a real ask, and it deserves a reason better than thoroughness. The reason is explanation. Matching is only useful if it can tell you why, and a why built on four answers per section is a guess dressed as a rationale.

The full profile works through all one hundred questions across six sections, plus optional practical constraints, in roughly eighteen to twenty-five minutes. More signal per section makes the explanation behind each family placement specific rather than gestural.

• Interests: the subject matter you are drawn to.
• Work values: what a job has to provide for you to stay in it.
• Preferred activities: the shape of the work rather than its topic.
• Work environment: where and with whom.
• Self-reported skill confidence: what you believe you are good at.
• Learning and career goals: what you want the next few years to build toward.

The skill section records self-reported comfort only. It is deliberately kept separate from tested ability, and no aptitude or reasoning score can enter the calculation at all: the engine strips those inputs before matching runs.

That separation protects both directions. A confident person is not handed careers they cannot do, and a person who tested badly on a practice suite one evening is not quietly steered away from a field they would be good at.

Results show which sections pulled a family toward you and which pushed it away, rather than an unexplained score. That makes disagreement productive: if a family ranks high because of a work-values section you answered carelessly, you can see that, and you can rerun.

It also makes the ranking less authoritative in a useful way. A number invites belief; a breakdown invites reading. The most valuable outcome of a hundred-question profile is often not the top family at all, but noticing that one section is pushing hard against every option you were considering.

After the six sections the profile offers practical constraints. They are optional because plenty of people do not have firm ones yet, and forcing an answer would manufacture precision that is not there.

If you do have them, answer them. A match that ignores a constraint you actually have is worse than no match, because it costs you the time it takes to discover the problem.

It is also worth answering them as they are rather than as you hope they will be. A constraint that is real for the next two years is a real constraint, even if you expect it to lift; a shortlist built for the situation after it lifts is a shortlist you cannot act on now, and the version you can act on now is the one that produces movement.

The practical risk in a long questionnaire is not boredom, it is drift: the answers you give in section five stop being calibrated the same way as the answers you gave in section one. That produces a profile that is internally inconsistent in ways nothing can detect, because every individual answer looks reasonable.

Two things help. Answer quickly, because a first reaction is usually a better reading of what you actually prefer than a considered one, and considering invites you to answer as the person you would like to be. And take the break in the middle rather than at the point you notice you need one: the section boundary is a cleaner place to stop than the middle of a run of similar prompts, and the profile saves automatically wherever you leave it.

The same answers always produce the same result, and the engine version is recorded with it. Nothing is sent to a server: sessions and results live in this browser's IndexedDB, you can resume where you left off, and you can delete any of it at any time.

Determinism is worth naming because it is unusual. There is no model in this loop, nothing is sampled, and two people giving identical answers get identical results. A matcher that varied between runs would be impossible to reason about and impossible to explain.

It also makes the version label on a saved run mean something. If the engine changes, results produced by the old version stay exactly as they were and stay labelled, so a comparison across months is either valid or visibly not. A stochastic system could offer neither guarantee, which is the practical argument against reaching for one here.