winning-ideationlisted

# Generating ideas that can actually place first The evidence base for everything here is `references/winner-corpus.md` — twelve winning submissions across three hackathons, six of them from a single hackathon so the placement differences are readable. Read it when calibrating; don't rely on recall. ## Four angles, run in parallel One agent producing ten ideas converges on a house style. Four agents with different angles diverge, which is the point. - **technical-wow** — the demo that makes a judge lean forward, with a hard engineering spine underneath. - **social-impact** — a real, nameable beneficiary and stakes that matter. - **sponsor-native** — impossible without the required technology, not merely using it. - **underserved-niche** — a specific audience nobody builds for. ## The three tests **The inversion test.** Nearly every winner in the corpus can be stated as "X, not Y" in one sentence: the model goes to the data, not the data to the model; authorization lives in the database, not the UI; vision is the last resort, not the first tool. Aim to write that sentence for your idea — if you cannot, it is not finished. This is the Originality criterion made mechanical. **The thesis test.** One line on why *this* required technology, phrased so a competitor using a different technology could not claim it. See `sponsor-tech-thesis`. **The demoability test.** Name the single visceral thing a judge sees inside three minutes — one moment, not a feature list. If the dem