Where is the human step?
Map whether review happens before rejection, before advancement, after ranking, during appeals, or only during periodic quality checks.
Human review
"Human in the loop" is a claim, not an evidence file.
Vendor diligence should identify what the human reviewer actually sees, what they can change, and how the workflow records review before an AI-assisted hiring output shapes a people decision.
Review proof
Evidence to ask for before accepting the claim.
What does the reviewer see?
Capture whether the reviewer sees scores, rankings, summaries, explanations, source fields, limitation notices, or confidence indicators.
Can the reviewer override?
Ask whether reviewers can override AI outputs, what approvals are required, and whether the system nudges or constrains override behavior.
What gets recorded?
Document logs for review completion, overrides, reasons, escalation, candidate status changes, and vendor-side troubleshooting.
What guidance exists?
Collect reviewer instructions, misuse warnings, role-based training materials, and support paths for uncertain or edge-case decisions.
Where should the tool not be used?
Ask for limitation statements covering unsupported job families, languages, geographies, data quality issues, and assessment contexts.