AI-generated generic voice; exact spoken transcript. Recorded synthetic local results, not live execution.

00:00:00.000 - 00:00:11.000 | Slide 1
Eight synthetic cases. Recorded local results, not live execution. Requires Python three point ten or newer. No API calls.

00:00:11.000 - 00:00:24.000 | Slide 2
Case C zero four requires document D zero seven. The regressed prediction instead retrieves and cites D ninety-nine, which does not exist in the fictional corpus.

00:00:24.000 - 00:00:40.000 | Slide 3
These commands were executed locally using Python's standard library. The baseline exits zero, with no gate failures. The deliberate regression exits one; both runs retain data files and readable Markdown reports.

00:00:40.000 - 00:00:53.000 | Slide 4
Regression recall at three and required citation coverage are both seventy-one point four three percent. Decision-label match is seventy-five percent. Fourteen automated gates fail.

00:00:53.000 - 00:01:07.000 | Slide 5
Valid source IDs cannot prove correctness. Case C zero five cites D zero eight but makes a false compliance claim. Human review should block it. These checks cannot judge answer meaning.

00:01:07.000 - 00:01:18.000 | Slide 6
AI Design: six dollars, ninety-nine cents. Complete collection: nineteen dollars, ninety-nine cents. Prices are in U.S. dollars.