From one team, with data, offered as fans who are coming back
First, what worked: a published rubric (five criteria, one to ten each), a serious venue, real sponsors, and a full room. Most hackathons never get that far. This is written by the team that scored all 29 projects on your rubric during the event, so every claim below has a number behind it.
Participants were told AI was optional. The rubric, meanwhile, never mentioned AI at all: it measured impact, idea, execution, feasibility, and usability. Both statements are defensible, and together they left the single most important strategic question of the sprint undefined: does the room reward AI that assists, or AI that acts?
The scoreboard answered it. All four ribbons went to projects where AI visibly acts: answering calls, building schedules, dispatching help, deciding reimbursements. Projects where AI assisted quietly, or where the AI lived in the stack without a moment on stage, clustered lower. Several were excellent engineering that effectively competed with one hand tied, because their teams optimized a target nobody had defined. They were not disqualified by any rule. They were out-read by teams that guessed the meta-game right.
The fix costs one slide: define what "AI optional" means for scoring, and show one example of a winning pattern before teams commit their 48 hours. Ambiguity does not average out. It compounds.
A snapshot of the event page two days before the deadline still listed a single judge. The panel filled in gradually during the final days. Teams that study judges (backgrounds, priorities, what a physician-judge wants to see versus an engineer-judge) build differently from teams that prepare blind. Some of our competitors never saw the full panel before submitting.
Twenty-five of twenty-nine teams will never hear a word about their work. No scores, no notes, nothing. Those twenty-five built real software for your event, and the silence teaches them that submitting was not worth it. We closed part of that gap with this site, but it should be structural, not volunteer work.
1. Define the AI expectation precisely on day one, with one slide showing what a winning pattern looks like in this event's judging culture.
2. Publish the full judge panel, with backgrounds, at kickoff. It changes how teams spend every hour.
3. Guarantee feedback: every submitting team gets a scorecard within one week, on the published rubric. Our scoring method and templates are yours to copy, free.
4. Judge the demo: require the core loop to run live on stage. It rewards teams who finish their product and punishes teams who finish their slides, which is the incentive you want.
We lost this event ourselves, and we will be back for the next one. We would rather lose to a well-run event than win a confusing one, and we suspect the 28 other teams feel the same.