scripts/demo_script.md

Model Court Demo Video Script

0:00-0:15 Problem

High-stakes AI decisions need adversarial review. A single model gives one confident answer; Model Court produces evidence, disagreement, Dissent, and a Human Gate.

0:15-0:45 Tribunal

Open the app, choose the escalation demo, and convene court. Point to Evidence Items, Role Agent arguments, Cross-Examination, Agreement Map, Decision Ranking, and the Judge Verdict.

0:45-1:30 Outcome Coverage

Run the approval demo and denial demo. Show that the same Tribunal structure produces different ranked Verdicts from different insurance facts.

1:30-2:10 AMD/Qwen Parallel Story

Switch to AMD/Qwen Benchmark when the endpoint is available. Show first-wave Role Agent timing: Claimant Advocate, Carrier Counsel, and Risk Officer run concurrently, while Evidence Clerk, Domain Expert, Agreement Clerk, and Judge remain ordered.

2:10-2:30 Audit Trail

Open Markdown Report and Full Audit Trail JSON. Emphasize that final action requires human approval and that benchmark evidence is preserved when timing data exists.