Skip to main content

​
Purpose

Every scored AI output in Exterview is checked, recorded, and made reviewable before a person acts on it. This page covers what runs today and what is Upcoming.

​
Automated quality checks

Every scored output passes through automated checks before it reaches a reviewer. An output that fails is not passed through silently. It stays visible for a person to review.

Schema check

The output has the structure a report needs, with nothing missing or malformed.

Score range

Every score sits inside its allowed range.

Citation presence

Claims point to evidence from the session, not unsupported assertion.
Upcoming: an additional model-based review of each output, and ongoing drift and regression monitoring over time.

​
Scoring Logs

Admins see the results in Settings → Operations → Scoring Logs. It shows the status of the automated checks, a log of recent scored rounds with their check results and the model that produced each score, and a per-role view of scoring. A check that did not run is shown as not run, never counted as a pass.

​
What is recorded with every score

  • Model and prompt provenance. Scoring Logs show which model produced each score. The prompt version behind each score is also recorded for audit, but it is not shown in Scoring Logs.
  • AI score overrides. A reviewer can override an AI score. The override is kept with a mandatory written justification.
  • Decision reasons. Every change to a candidate’s decision requires a stated reason, recorded with who made it.

​
What the AI employees did

  • Show what the agent did. When you ask Smaya, a transparency panel above the answer shows the steps the agent took to reach it.
  • Named hand-offs. When one AI employee hands work to another, the reply names the employee that did the work. Hand-offs run as you, with your permissions.
  • Approval before writes. Creating a role, publishing a job, or contacting a candidate always asks a person to approve first. Upcoming: an Agent Catalog of the pre-built agents behind your AI Employees, Agent Lineage (a map of how work hands off between agents, built from live hand-offs), exportable run traces, and a control plane showing each role run and what it is waiting on.

​
Auditability

Because checks, overrides, and decisions are recorded, the chain is reviewable after the fact: what was produced, what was checked, what was changed, and which person made the decision. See Auditability.

​
Responsible AI, operationally

Content-based scoring, explainable outputs, mandatory human review, and no autonomous adverse decisions are enforced and recorded at this layer. No machine path sets a candidate to hired, rejected, or withdrawn. See Responsible AI for the principles.

​
FAQs

Yes. This layer improves the quality and traceability of AI outputs; it never replaces the human who owns the outcome. No adverse decision is made autonomously.
It stays visible for human review rather than being used silently. A person sees the check result and the evidence before acting.
Yes. Check results, overrides, and decision reasons are recorded and reviewable, and feed the audit trail. See Auditability.