How the demo suite is made
The method varies by demo, but the same principle applies throughout: show what the model does, mark uncertainty, and separate source-based results, curated corpora, and synthetic examples.
Static snapshots
Public featured examples are stored in data/prewarm/runs and served as completed run results. This makes the demonstration fast and reproducible, and reduces the risk that a public visitor encounters infrastructure errors or unpredictable live costs.
AI-generated interpretations
When a demo uses AI for summarisation, classification, or comparison, output is validated against a schema and shown with a methods panel, source references, or a disclaimer. Results should be read as analytical hypotheses, not authoritative answers.
Privacy and analytics
Analytics is opt-in for deployment and intended for simple visit statistics, not individual tracking. The Plausible script is rendered only when both PLAUSIBLE_DOMAIN and PLAUSIBLE_SCRIPT_URL are set.