See evaluate → approve → export in one thread
Institutional buyers use this loop before production Copilot scope opens. Inspired by agent event traces (Google ADK-style inspection) — orientation only on www; live evaluate runs on your tenant after pilot intake.
Click each step — event trace updates
Same semantics as production: policy-bound workflow with human gate before export.
Session: Evaluate started
Complete all 5 steps to unlock export achievement — then start sandbox or apply for pilot.
Human-in-the-loop (AG-UI pattern)
High-risk Copilot scope routes to a named approver before any export is valid. Low-risk sandbox paths auto-record only.
tle_id: tle_demo_001 decision: conditional confidence: 0.72 approver: pending — CISO delegate rid: (your Request ID)
Orientation sample — not live tenant data. Verify export integrity.
Approved commercial surfaces only
Public www uses a locked component catalog — cards, badges, exports — not arbitrary agent HTML.
Run this on your governance meeting agenda
Apply for Copilot Readiness or Trust Brief — include your Request ID in intake.
Policy change demo
Five-minute buyer story — policy change → Trust Ledger
Publish policy v3.2, invalidate stale v3.1 evaluations, re-brief, then evaluate Copilot rollout intent. Sample decision allow · 0.80 after re-brief.
- 1 Policy baseline
- 2 Invalidate + re-brief
- 3 Evaluate + TLE
Policy updated — Copilot Acceptable Use v3.2 (guest sharing blocked in production scope).