# Five-minute technical walkthrough

1. **Context (45 sec):** Explain the hypothetical housing desk and the four queues. State that data is synthetic and customer requirements remain assumptions.
2. **Design (45 sec):** Show the architecture. Explain why inference is local and why human review is a first-class outcome.
3. **Demonstrate (90 sec):** Route “My laptop cannot connect to wifi.” Compare approaches. Try “Can you help me tomorrow” to show abstention. Try “The heater is broken and my housing bill is incorrect” to expose a multi-issue failure.
4. **Evidence (60 sec):** Show the 24-case comparison, coverage, accepted accuracy, and review-case recall. Explain why the rule baseline currently wins. Do not call cosine scores confidence.
5. **Next steps (60 sec):** Ask the audience which queues differ from their workflow. Record corrections in a consented, anonymized pilot; test independently before integration.

## Readiness questions
Can you explain TF-IDF without jargon? Identify two failure cases? Distinguish implemented privacy properties from proposed controls? Describe the rollback path? Explain why a higher-coverage model can still be the worse choice?