video1.0<iframe src="https://www.loom.com/embed/5ef9f941c6644d158340f6eb480238a3" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/5ef9f941c6644d158340f6eb480238a3-df47a962fbf782a0-full.jpg253.166A stalking threat plays out across 8 turns. Same model, same conversation — run two ways. Without a safety layer, the per-turn classifier sees three separate empathetic messages. With Sango Guard, the targeting pattern (named target + location + daily routine) triggers high-urgency state by Turn 2. By Turn 4 — "Tomorrow morning. When he's alone." — Guard is shaping the response before it ships. Demo includes the audit trail export at the end: one-click PDF with plain-English explanations for every flagged turn. The kind of record California's 24-hour incident reporting rule expects you to have. ~4 minutes. If you want a walkthrough on a sample of your own conversations, contact at the end.