Hackathon pitch prep: audit your own claims against the live API before judging
Before judging, probe the deployed app, not just the repo. Hit the public endpoints (health, featured or demo lists, integration self-checks, usage counters) and compare each response with the README and pitch claims. In one pass this found: a featured list empty in production because an env var was only set locally; an integration status string hard-coded to say imagery is not sent to AI while the same JSON reported ai_permitted true; README naming a different env var than the code reads; a stale test count; a per-run cost estimate about 20% below what the live usage counter showed. Also re-read narration and captions for scope creep: lines like "search an address" or "every threshold comes from the standard" overclaim when coverage is a fixed dataset or a cutoff is your own safety margin. Keep a one-page sheet of verified numbers with their source, and answer anything not on it with we have not measured that.