A score is only as good as the evidence behind it.

We don't score “a person.” We score specific, job-relevant traits — the way someone reasons through ambiguity, how they make a hard call, how clearly they bring others along, how fast they learn from a mistake. Each role defines which traits matter before anyone is interviewed, so the same rubric applies to every candidate for that role.

We deliberately do not infer protected characteristics, personality “types,” or anything a resume keyword could stand in for. If a trait can't be tied to observed behaviour in the conversation, it doesn't get a score.

The pipeline is intentionally boring and inspectable. In order:

• Structured interview. A consistent set of questions per role, adapted in follow-up depth — not in difficulty — so everyone is measured on the same ground.
• Language understanding. Responses are turned into semantic embeddings and matched against the behaviours each trait is defined by. We compare meaning, not keywords, so a strong answer phrased plainly scores as well as a polished one.
• Evidence linking. Every point of a trait's score is anchored to the specific passage that earned it. No passage, no points.
• Calibrated confidence. Thin or ambiguous evidence widens the uncertainty band and lowers confidence — the model says “I'm not sure” out loud instead of guessing precisely.
• Bias guard. A separate check screens each generated question and score for adverse-impact patterns before results are shown.

Most tools give you a number that looks certain whether or not it should be. Placedon reports a score andthe range it's confident in. A trait shown as 82 with a tight band means the evidence was clear and consistent. The same 82 with a wide band means “promising, but ask more.” That distinction is the point — it tells you where a human should look closer rather than pretending the machine already knows.