How the AI scores you — in full

Where AI is used

• The Experience Translator maps the story you write about your work onto our public skills taxonomy. Every claim it produces quotes your own words back to you as the source, and everything it produces is labelled a claim — unverified by definition.
• The rubric evaluator scores challenge submissions criterion by criterion (0–4) against the published rubric. It must reference something specific in your submission for every score, and it is explicitly instructed not to score writing quality, grammar, fluency, length, or formatting.
• The employer intake turns an employer’s plain-language job description into a suggested skill spec. It recommends; the employer edits and decides.

Each of these runs on Anthropic’s Claude when an API key is configured, and on a deterministic offline engine otherwise. Every score is labelled with which engine produced it.

Notice before you are evaluated

Before any challenge you can see: the full rubric, the time cap, the tools policy, and the fact that AI performs the evaluation. Nothing about the scoring is revealed only after you submit. If an employer directs you to a challenge for a specific opening, that page is your advance notice of the assessment and its criteria.

Human oversight — the contest path

• Any candidate can request human re-evaluation of any score: review@provenly.dev with the submission id. A human re-reads the submission against the same rubric. The human result stands.
• No candidate is rejected from an opening on this platform by a solely automated decision, and employers are contractually bound (Terms §4) to remain the human decision-maker.
• If you prefer not to be evaluated by AI at all, ask for the alternative process at the same address — human evaluation against the same rubric.

Bias posture

• The system never receives your name, age, gender, photo, nationality, or address as scoring input. It scores text against criteria.
• Rubrics are behavioural (“names the cost of the decision”), not stylistic — because style is where language background leaks into scores.
• We publish score distributions per challenge and monitor them for drift. Before any score is used for regulated hiring decisions in jurisdictions that require it (for example New York City), an independent bias audit will be commissioned and its summary published on this page.

What we refuse to build

No emotion recognition. No video or voice analysis. No personality inference. No social-media scraping of candidates. No prediction of “culture fit”. These are either prohibited for workplaces under the EU AI Act or incompatible with the premise that people should be judged on inspectable work — usually both.

Regulatory posture

Candidate-evaluation AI is classified high-risk under Annex III of the EU AI Act. We are building toward the provider obligations that regime requires — risk management, data governance, technical documentation, logging, human oversight — on the published compliance timeline, and this page will track that status. US framework laws we track include NYC Local Law 144, the Illinois Human Rights Act (as amended effective 2026), the California ADMT regulations, and Colorado’s SB 189 disclosures (effective 2027). Details of data handling are in the Privacy Policy.