Every hire goes through a human. The agent does the work; the recruiter makes the call. Side-by-side scorecard + transcript + audit trail. One click to override.
Trusted by recruiting teams shipping audit-grade hiring
01 · Escalation
Three triggers send a candidate to the recruiter queue: interview score in the 50–69 borderline band, coding-challenge score in the 50–69 band, or a delivery-verification flag for material claim mismatch. The Lark bot pings the recruiter; the dashboard surfaces a callback queue with the reason for escalation prefilled.
02 · Decision
The recruiter sees the agent's recommendation, the supporting citations, and a single primary CTA matching the recommendation. Two ghost CTAs cover override paths — request second round, send to coding challenge anyway, or reject. Every override reason is logged with the same provenance shape as the original decision; auditable why a human went a different way.
03 · Parity
All three surfaces call the same record_hiring_decision() handler. The Lark bot is for managers who live in chat; the dashboard is for recruiters working sets of candidates; the API is for HRIS integrations. The 21-tool Lark bot list (list_candidates, get_interview_report, bulk_invite, …) is documented in the trust report.
Override-ready
GDPR Art. 22 honoured at the surface, not in a footnote. Every recommendation lands in a human-review queue with the rubric line, the evidence, and a single button to override — and that override is itself logged with the same provenance shape.