# Why KUL Talent Refuses to Give You a Fake 93% Match Score

> KUL Talent uses evidence states and explicit unknowns instead of unexplained candidate-match percentages.

A precise-looking number can make a complicated hiring decision feel scientific. The useful question is whether the evidence underneath it deserves that certainty.

## 93% of what?

A job seeker opens a role, a system scans the CV, and a few seconds later the interface says “93% match.” It looks simple. But what moved the number from 89 to 93? Was it title similarity, keywords, years of experience, location, a licence, work authorisation, or an inference the user cannot inspect?

A battery can be 93% charged because the number has a stable definition. Candidate fit does not have one universal denominator. A job description is not a perfect specification, a CV is not a complete record of a person, and two recruiters can reasonably care about different evidence.

## Evidence before certainty

KUL uses a small vocabulary that says what kind of evidence exists instead of compressing every uncertainty into one score.

- **VERIFIED** — A fact confirmed through an authoritative or validated source.
- **SUPPORTED** — The available evidence strongly supports it, without pretending it is independently verified.
- **CLAIMED** — A candidate or employer stated it, but KUL has not independently verified it.
- **INFERRED** — The evidence makes it reasonable to infer, but it should not be presented as a confirmed fact.
- **UNKNOWN** — There is not enough information yet. The right next action may be verification, not a lower score.
- **CONFLICTING** — Available evidence disagrees. The conflict should be visible instead of silently resolved by a model.

This is less theatrical than a giant percentage. It is also easier to audit, challenge and turn into an action.

## Unknown is a valid answer

Imagine a role requires a professional licence and the candidate's profile does not mention it. Missing evidence is not automatically negative evidence. Inferring the licence from a job title can be wrong, while silently lowering an opaque match score hides the real issue.

The responsible state can simply be: **licence unknown — verify.** The uncertainty becomes a useful next step instead of an invisible penalty.

## Fit can still be useful

Rejecting fake precision does not mean refusing to make judgements. KUL can distinguish direct fit, transferable fit, needs-verification situations and confirmed mandatory constraints. The difference is that the conclusion stays connected to evidence.

A recruiter should be able to ask why a candidate is in a state. A candidate should be able to see what could be clarified. “Because the model says 84%” is not enough.

## AI should reduce work, not manufacture authority

AI can structure information, compare evidence, draft, search and prepare. But fluent output is not the same thing as verified truth. The more consequential the decision, the more important it becomes to preserve the boundary between a claim, an inference and a fact.

**Show the evidence. Show the unknowns. Make the next useful action obvious.**

That is the product principle. If it produces fewer dramatic percentages and better decisions, KUL is comfortable with the trade.

Canonical article: https://www.kultalent.com/insights/why-no-match-score
Published: 2026-10-14