Automated scoring, human review and candidate rights

A report is a decision aid, not a decision

We do not make hiring decisions and we do not recommend rejections. The hiring organisation decides, using the report as one input among several, and the report says so on its face.

This matters legally as well as ethically: Article 22 of the GDPR restricts decisions based solely on automated processing that produce legal or similarly significant effects. Our design keeps a human decision-maker in the loop by construction.

How scoring works

There is no artificial intelligence anywhere in the scoring path. Not a language model, not a classifier, not a model of any kind. The same answers always produce exactly the same report, which is why it appears the second a candidate finishes.

The assessment is a fixed, versioned bank of situations. No option is the correct one — every option is something a competent professional might do — and each option carries a written weight across six capabilities and four style axes.

• Twelve situations are scored. For each one the candidate marks what they would do first and what they would never do; both count.
• The weights of the chosen and the ruled-out option are added up per capability, then expressed on a nought-to-a-hundred scale against the maximum that bank could produce.
• Fit is the same numbers seen through the hiring company's brief: they say which capabilities the role cannot do without, and that changes the recommendation without changing the measurement.
• The style axes are never scored. They have no better end, and the report says so.
• No personality inventory is used at any point, and nothing the candidate writes is scored.

When we refuse to publish a result

Some answer patterns cannot be honestly read: the same position chosen over and over, the whole thing completed faster than it can be read, the two attention checks failed, or the first and second halves of the assessment disagreeing with each other.

When that happens no report is produced, the company is told only that the assessment could not be read, and their credit is returned. We would rather return the money than sell a number we do not believe.

Human review

Any candidate can ask for a person to review their result, for any reason, without justifying it, and with no effect on their application. We will look at their answers against the bank and either confirm the profile or correct it, and tell you which.

What we never ask anyone

We collect no demographic data. Not age, not sex, not nationality, not education, not photographs, not anything that describes the person rather than their professional judgment. This is a deliberate decision and not an oversight: data we do not hold cannot leak, cannot be subpoenaed, and cannot quietly enter a score.

The one thing a hiring company tells us about a role is its level — junior, mid, senior or lead — and that is a statement about the scope of the job, not about whoever applies for it.

What the hiring company tells us afterwards

Three weeks after the first report on a process, we send the company one email — one per process, never one per candidate — asking whether the role was filled, how useful the reports were, whether they would use it again, and what would make it better.

We do not ask who they hired, we do not ask about any individual, and no candidate is named. What we learn is whether the product is any use, and we hold it against the process rather than against a person. Nothing from it goes back to a candidate or appears on anyone's report.

What candidates tell us about the situations

After a candidate finishes — never before, and never while it could affect anything — they can flag any situation as confusing or unfair, and add anything else they want to say. It is optional, the hiring company never sees it, and their result is already sent by then.

It is how we find out that a situation assumes a working context not everyone has had. A question like that scores beautifully and measures the wrong thing, and no statistic we run would catch it on its own.

What candidates can ask for

Every candidate has the following, and exercising any of it has no effect on the outcome of their application.

• A copy of their own report, automatically and permanently.
• A copy of all data held about them.
• Human review of their result.
• Correction of anything inaccurate.
• Deletion of everything, confirmed when done.
• An explanation of the criteria and their weights — which are published on this site in any case.
• All of it through the form at /legal/data-request.

Who sees a result

The organisation that invited the candidate. Results are not sent to candidates automatically: a result is scored against one company's specific brief, and outside that context the numbers mislead more than they inform. A candidate can request their own copy at any time and we send it with the explanation attached.

Results are never published, never ranked publicly, never sold, and never shared with another organisation without the candidate's explicit, specific permission.

Bias and limitations, stated plainly

Any assessment encodes a point of view about what good work looks like, and ours is no exception. We publish the capabilities and their weights so that the point of view is inspectable rather than hidden.

The weights are a reasoned hypothesis. They have not yet been calibrated against professionals whose ability is independently known, and until they have, nobody — including us — can say how well this predicts anything. We will not pretend otherwise while that work is outstanding, and we will publish what it finds, including if it is unflattering.

We do check, continuously and with the real data, what we can check: that no situation is answered the same way by almost everyone, that each one agrees with the capability it claims to measure, that where an option sits on screen does not sway the answer, and that the strongest option is not simply the longest one. Those tests say the instrument behaves consistently. They do not say it is right.

So: we do not claim the instrument is neutral, and we do not claim it is validated. We claim it is written down, identical for every candidate, deterministic, and open to challenge — and that we would rather publish this paragraph than a number nobody can check.