Governance
The method is the product.
Anything that decides something about a person has to be explainable to that person. Ours is arithmetic, disclosed in full, and reproducible by anyone holding the same inputs.
Assurance grid
Four pillars we hold ourselves to.
Deterministic scoring
Fixed weights of 0.45 skills, 0.25 qualifications, 0.30 experience — published, not a general description.
Auditability
Every element of a result can be taken apart and shown to the person it describes.
Data ownership
The record belongs to the person. It can be exported or deleted at any point.
Consent
Nothing is shared with an organisation beyond what the person has explicitly authorised.
How the measurement is made
Four properties, not four promises.
01
Disclosed weights
The measurement is a fixed arithmetic blend of skills, qualifications and experience, weighted 0.45, 0.25 and 0.30. Those proportions are the same for every candidate and every role, and they are published here rather than described in general terms.
02
Structured requirements
Roles are not read from advert copy. They are matched to occupational requirements held in a published taxonomy, so two employers writing the same job get the same requirements.
03
Reproducible
The same inputs always produce the same result. There is no sampling, no drift, and no model version that quietly changes an outcome.
04
Auditable
Every element of a result can be taken apart and shown to the person it describes. A result that cannot be explained to a candidate is a defect.
The platform measures and explains, and a person can be part of the conversation — see Coaching.
What the AI is not permitted to do
A language layer, held on a short leash.
Language layer only
The model rewrites and clarifies. It has no access to the scoring engine and cannot alter a measurement.
Temperature zero
The same input produces the same output. Nothing is left to sampling.
No-new-facts guard
Every generated line is checked back against the candidate's confirmed record before it is shown. A claim that is not in the record is blocked, not softened.
Where it does not yet reach
A named limitation, not a hidden one.
Weighting experience will always favour people who have had the chance to accumulate it. That structurally disadvantages people returning after a break, people leaving prison, people who have been out of work a long time, and people changing direction late. We regard that as a defect in the blend rather than a property of the candidates, and it is under formal equality impact assessment before the platform is used in any public employment programme.
Questions we are asked
Governance FAQ.
Does CareerPathIQ decide whether someone gets a job?
No. It produces a measurement and an explanation of that measurement for the member. CareerPathIQ does not rank or select candidates, and any recruiter, employer or adviser judgement remains entirely theirs.
Who can see a member's analysis and reasoning?
The member does, in full. An employer or recruiter sees only what the member has explicitly permitted, for the purpose agreed at the time. Career Profiling stays private self-insight and is never shared for screening or selection.
Can a candidate see how their own result was produced?
Yes. The method is disclosed, the requirements come from a taxonomy anyone can inspect, and every element of a result can be traced to a specific piece of the candidate's own evidence.
Does the AI write things the candidate did not say?
No. The model works on language only. Every generated line is checked back against the candidate's confirmed record, and any claim that is not in the record is blocked rather than softened.
What happens if a candidate disagrees with a result?
The inputs are the candidate's own and they can correct them. Because the method is fixed and disclosed, a corrected input produces a predictable change rather than an unexplained one.
Bring the questions you ask your suppliers.
Method, audit trail, data handling, equality duty. We would rather answer them early.