Probity checks what a candidate wrote against what can be verified. Every finding has a source. A human decides.
We take only the text and the metadata. The file itself is not stored.
Registries, GitHub, metadata, timeline and 5 more. Every finding comes with a link.
“Is this the person?” and “Is what they write true?” — separately.
The system suggests the next step. The decision is recorded under your name.
One overall score hides what exactly is wrong. And an automated conclusion carries legal weight (CJEU case C‑634/21). Hence two separate scales, and sources.
Not “a neural network decided”, but a document, a registry or a link.
Who authored the file, when and in which time zone it was created
Whether the dates of jobs, studies and technologies add up
Whether the company existed in those years
Whose account it is, whether the activity is real and where the links lead
Whether the links open and whose work is there
Whether the profile matches the résumé
Mentions in open sources, the age of the email domain
One file author across “different” candidates
Fills in facts where there is no ready-made source. Never produces red on its own
The candidate card goes through 3 questions from the top down and lands in one of 4 buckets.
10 résumés from a LinkedIn search. They looked normal. One was real.
Honestly: the recruiter pre-selected this batch on suspicion. An ordinary pipeline has far fewer fakes.Only the text, metadata and a hash
That is biometrics under the law
Criminal records, health, religion, views — no
Being absent online is not evidence
No automatic rejections
No “hire / don’t hire” conclusions
Including what we missed
No sign-up. The queue, the cards, the two scales — just as in real work.