Methodology
ZRG Autopilot — autonomous prep, human approval gate
On critical predictive alerts Autopilot prepares supplier shortlists (coverage map), RFQ texts and contract skeletons. No send and no contractual binding without explicit approval by an authorized person — enforced in code, not UI convention alone.
Step 1
Trigger
Predictive alerts above configured thresholds (impact, confidence) or manual trigger via POST /api/v2/autopilot/trigger. Each run is stored in autopilot_runs and logged in the Black Box.
Step 2
Drafts
Top-5 alternatives from the coverage map, LLM-generated RFQ text per candidate and a watermarked contract draft (DRAFT — NOT LEGALLY BINDING). Rate-limited LLM calls per tenant.
Step 3
Approval gate
Only users with can_approve_rfq (owner/admin or explicitly enabled team members) set approved_by/approved_at. Send (POST …/rfq/:id/send) fails server-side without approval — including DB constraints.
Step 4
Scope
Autopilot is decision support and drafting — not legal advice, no guarantee of supplier quality or contract terms. Contracts bind only after signature by authorized representatives.
Intentionally frictionful
There is no one-click accept-all. Approval requires deliberate action per RFQ — by design, so authorized persons review content before anything goes outbound.
ZRG Autopilot creates automated drafts from market and portfolio data. No send without approval. ZRG accepts no liability for content or terms of sent enquiries.