demo-script.md

Genuine demo script

Use a fresh local session and record the visible browser state. Do not substitute screenshots, fabricated traces, or model prose for tool activity. If the recording uses a time cut while a local run completes, disclose the cut on screen or in narration.

1. Open the local app and show the local-model readiness state. State that the example is synthetic and that absent availability means unavailable.
2. Load the baseline pantry example. Point out that each coverage row is an exact staffing target. Review and explicitly confirm the inputs.
3. Request: “Inspect this case, find a schedule, and prepare a review using evidence from your tools.” Show the actual tool trace, the independently verified schedule, workload metrics, and the separate approval control. Do not approve until the result is visible.
4. Approve the verified baseline schedule and download the approved bundle. Open the ZIP locally and show that it contains schedule, verification, and calendar snapshot files.
5. Start again with the cancellation example. Confirm inputs, request a schedule, then show the repair result and its replacements metric. Explain that it compares against the most recent approved real schedule.
6. Load the infeasible example, confirm it, and request an infeasibility diagnosis. Show the status and evidence without calling it a formal certificate.
7. Request hypothetical options to restore V02 availability for S3 or raise both V01 limits. Show that simulations are labeled hypothetical and that a proposal needs a separate confirmation before it changes the case.
8. End by showing that no change, schedule approval, or export happens automatically from model output.

If a local run is trimmed for pacing, state the cut and retain the unedited trace or a timestamped recording as supporting evidence. Do not claim measured latency, user outcomes, contest status, or a completed full local-model evaluation unless recorded separately.