Judging Rubric
Build an autonomous agent that solves any real-world task on the open web.
Scoring Overview
Technical Complexity
Can it handle the "real" web? Resilience against JS-heavy pages, CAPTCHAs, and multi-step workflows. We're looking for robust, production-grade architecture.
Tool Integration
Mastery of TinyFish + OpenAI. Deep, clever integration where the tools are central to the solution's "magic," not just an afterthought.
Utility & Impact
Does this solve a real pain point? High leverage and clear PMF. Could this be a standalone company or a major industry shift?
Innovation
The "Wow" Factor. Is the approach novel or surprising? We want to see something that feels like it shouldn't be possible yet.
Demo & Pitch
Clarity. Did the team tell a compelling story and execute a polished live demo?
Awards & Prizes
Each team can only win one category. Prizes are awarded in the following order:
- Top 3 — selected first from the overall leaderboard.
- Spot prizes — awarded from the remaining teams (teams that already won Top 3 are not eligible for spot prizes).
Top 3 Teams
Overall winners based on total score across all five criteria. Selected via pairwise comparison by our judging panel.
Spot Prizes
Deep Sea Architect
Best Use of TinyFish
Technical elegance + "Aha!" moment. For the team that solves a massive, positive problem using the Open Web in a way that feels like magic.
Most Likely to Be the Next Unicorn
Best Product Vision
Awarded for the best product vision and product-market fit. This feels like it could be a real company.
Most Likely to Go Viral
For the Wrong Reasons
For the most unhinged, "Black Mirror" style use of autonomous agents. Impressive, terrifying, and unforgettable.
The Rube Goldberg
Most Over-Engineered Solution
For the team that builds an unnecessarily complex (but beautiful) way to solve a problem. Extra points for elegance in absurdity.
WTF (What the Fish)
Most Bizarre Use Case
Reserved for the most bizarre, hilarious, or "why does this exist?" use case of the day.
Important: Declare pre-existing work
If your project was not fully built during the hackathon, you must clearly declare what existed before and what was built on the day. Only the work produced during the hackathon will be evaluated by judges. Failure to disclose pre-existing code may result in disqualification.
You will be asked to confirm this during submission. If you used an existing project as a starting point, provide a clear description of your hackathon contributions in the submission form.
How Judging Works
For Participants
- Opt in for judging on the submission form in the Team tab of the participant dashboard. Make sure all fields are filled in, including your current table number.
- Code freeze at 4:00 PM. All submissions must be in by then. There is a ~5 minute grace period, but the hard cutoff is 4:05 PM.
- After submissions close, we may ask some teams to relocate tables so all judged teams are packed together. The dashboard will tell you if you need to move. Please cooperate quickly so judging can start on time.
- Judging starts at 4:30 PM. Judges will visit your table. You'll have about 4 minutes to demo your project and answer questions. Be ready with your demo running.
- Winners are announced at 5:30 PM during the closing ceremony.
For Judges
- We use Gavel, a pairwise comparison judging system. You'll be assigned tables to visit via the Gavel webapp.
- At each table, you'll have approximately 4 minutes with a countdown timer. Watch the demo, ask questions, and evaluate based on the rubric above.
- After each visit, you'll vote whether the current project is better, same, or worse than the previous project you saw. The algorithm handles the rest.
- At the end, Gavel produces a score leaderboard. We select the Top 3 from the leaderboard, then award spot prizes from the remaining teams.