AI scores. People decide. Always.

Ready never auto-rejects and never auto-hires. Every score is advisory to a human reviewer. The hiring manager opens the candidate's session, reads the evidence behind each rubric line, and makes the call.

That posture matches the requirements across the EU, the US, and the UK. The legal framing differs by country; the product behavior is the same everywhere.

• No automated rejection. The model does not block a candidate from advancing. A reviewer reads the session first.
• No automated hiring. The model does not extend offers or move candidates into final stages.
• Final decision belongs to the employer. The score is a recommendation. The hiring manager makes the call, with the full evidence panel in view.

Every simulation produces a transcript, a scenario state record, a voice pattern record, and a rubric-anchored score. The hiring manager opens the candidate's session, reads the evidence behind each rubric line, and decides.

Every rubric line carries an override control. The reviewer adjusts the score, adds a written reason, and saves. The override is logged with the reviewer's identity, the original AI score, the new score, and the reasoning. Override patterns feed back into prompt and rubric calibration so the system sharpens with human pushback rather than drifting without it.

• Evidence first. Every rubric line points to a transcript segment, scenario event, or voice pattern observation.
• Line-level override. Every scored claim can be adjusted individually. Reviewers do not have to accept or reject the whole score.
• Reason required. Every override carries a short written justification. Numbers on their own are not enough.
• Full audit trail. Original score, override value, reviewer identity, and timestamp recorded.

Ready is built to monitor pass rates per protected class on every customer deployment. When the lowest-passing class falls below the regulatory fairness threshold relative to the highest-passing class, the customer receives an alert and the assessment is paused for review until calibration is verified. No candidate is screened during a paused period.

Aggregate vendor-level audits are not enough. Where the law requires it, an independent third-party bias audit will be commissioned per customer before the first regulated candidate is screened, and refreshed annually.

• Fairness threshold per customer. Applied per customer and per protected class. Drift triggers an alert.
• Auto-pause on drift. Assessment suspended; customer notified; review and calibration required to resume.
• Audit commitment. Where a jurisdiction requires it, an independent third-party bias audit will be commissioned per customer before the first regulated candidate is screened, and refreshed annually.

Enterprise accounts can enable a second-opinion tier. A trained reviewer from Ready inspects borderline candidate sessions before they reach the hiring manager. The reviewer confirms the AI score, adjusts it, or flags the session for additional review. The hiring manager sees the AI score, the reviewer's notes, and any disagreement between them. Nothing is hidden.

Borderline is defined per contract. Common rules include sessions inside a configurable score band, sessions with low evidence confidence, and sessions a hiring manager has flagged. The tier is always opt-in.

• Configurable trigger. Score band, evidence confidence, or hiring manager flag. Each enterprise defines its own border.
• Trained reviewers. Ready staff trained on the rubric, the bias controls, and the override protocol.
• Disagreement is visible. The hiring manager sees both the AI score and the reviewer's notes side by side.
• Always opt-in. Off until the employer enables the tier and the candidate has consented to it.

Ready treats candidate experience as a primary product metric. Every candidate sees an explicit notice that they are speaking with an AI assistant, an explicit time estimate, an explicit human-review commitment from the employer, the rubric they are being scored against, and a brief structured summary after the simulation.

Sessions are short by design. A technical failure grants an automatic retake.

• Explicit AI notice. The candidate is told before the session begins that they are speaking with an AI assistant.
• Rubric shown to the candidate. The dimensions being scored are visible before the session starts. No black-box scoring.
• Structured feedback after the session. Every candidate receives a short structured summary, regardless of outcome.
• Retake on technical failure. Wifi drops, browser crashes, microphone issues. The candidate gets the session back automatically.