No "best practices" and no "enterprise-grade security". What feeds into a score, what does not, who reads what, and what each revoke button undoes.
The interview assessment comes exclusively from the audio transcript: content, structure, clarity, fluency. Face, expression, clothes, lighting and whatever is behind you feed into no score at all. The video exists for a human recruiter to watch and for anti-fraud verification — and that separation is enforced in the code, not only here.
Every score cites the excerpt that produced it. There is no number without the CV line or the transcript quote behind it — it is a database constraint, and not an intention of the code.
The employer of the job you applied to sees your profile, your answers, your assessment with the evidence, and the video. No other employer sees anything — not even the name.
An independent recruiter who wants to present you to a job needs your authorization first: until you accept, the company sees not even your name, and the text they wrote about you is the same one you read.
Above the requirements cut, nothing is decided by machine. Every decision carries a named person, and the database refuses to record one without them. If you open the result of an application and the screen says there is no named owner, it is because there genuinely is not — and in that case it can be investigated.
The requirements cut is the only automatic rejection, and it is binary and factual: work authorization, location, language level, years of experience, certification. All published on the job page before you apply, and the rejection says which one failed.
All of that lives in privacy, in your account, without writing to anybody.
CV, answers and transcript are personal data, and where they are processed is a compliance decision — not a pricing one. The model that does the assessment is configuration, and choosing it takes into account where the processing happens, under GDPR and LGPD.
Every assessment records which model produced it and on which version of the engine. It is the question a challenge asks six months later, and it has an answer.