Upload a policy, then a CSV of agent conversations. Each row is evaluated against the policy and cited inline — never a fabricated rule. Approved, disputed, and corrected rows persist so the calibration loop is demoable end-to-end.
policyId and batchIdare mirrored in this browser’s localStorage so a reload restores the demo. Manager actions are persisted server-side but anonymous — this is a single-tenant demo (no workspaces / users / auth in this milestone). Quotas, billing, and aggregated reports are not wired yet.—Batch: —customer_message and agent_response are required. Optional: conversation_id, agent_name, customer_name_or_id, channel, started_at, conversation_text, resolution_status, csat, tags — captured if present (external_id is still accepted as an alias of conversation_id).No evaluations yet. Press Run evaluation above.