DeBet

Prediction markets reward being right. DeBet rewards being smart. The first argument-quality exchange where your payout is determined by the strength of your reasoning, not the flip of an outcome.

Comment

Inspiration

We've all been in an argument where someone says "I'd bet money on it" and then nothing happens. That throwaway phrase stuck with us. What if you actually could? Not just on outcomes, but on the quality of your reasoning? We wanted to build something that rewards people for thinking well, not just thinking loudly. That's where DeBet came from.

What it does

DeBet is a debate-driven prediction market where users take positions on real-world cultural and financial questions, write a short argument defending their stance, and get scored by AI on the quality of their reasoning. The better your argument, the more points you earn. Points are redeemable for real money, meaning the sharpest thinkers actually get paid.

How we built it

We built DeBet using Next.js and TypeScript on the frontend, with Supabase handling our database, auth, and row-level security. We used Supabase migrations to manage our schema incrementally, including market economics, blind yes/no voting, argument limits, ledger tracking, and duplicate entry prevention. To ensure objective evaluation, our scoring system ensembles Claude and GPT-4o to neutralize single-model biases. Unlike traditional LLM grading which operates as a ‘black box’ subjective scorer, DeBet utilizes AI to identify and quantify raw parameters – Evidence (E), Logic (L), and Fallacy (C) – which are then fed into a deterministic Bayesian-inspired formula. By treating these factors as interdependent multiplicative terms in a post-processing step, the final score remains mathematically grounded: if either the factual basis or the logical connection is missing, the overall strength collapses toward zero. Everything was built in VS Code with Claude Code handling a lot of the heavy lifting on the agentic coding side.

Challenges we ran into

The grading system was the hardest part by far. Getting the AI to score arguments consistently and fairly across wildly different topics took a lot of iteration. We also had to think carefully about the market economics how to structure payouts so that good reasoning is actually incentivized and not just rewarded for length or confidence. Getting the RLS policies right in Supabase to make sure users could only see and edit their own entries added another layer of complexity.

Accomplishments that we're proud of

We shipped a fully functional end-to-end product in one hackathon session. The database schema alone went through 9 migrations as we kept refining the logic, and we're proud that the core loop — pick a side, write your argument, get scored, earn points — actually works and feels like a real product.

What we learned

Prompt engineering is genuinely hard. Getting Claude to be a fair and consistent judge across completely different debate topics required a lot more nuance than we expected. We also got much more comfortable with Supabase's RLS system and learned how to structure a real-time app without it feeling slow waiting on API responses.

What's next for DeBet

We want to introduce real money payouts through Solana that’s why we added the wallet connect to basically hint at it, pull in live news topics automatically so the platform stays current, and build a reputation system so the best arguers build a track record over time. Long term, DeBet could become a new kind of signal layer where cultural conviction is actually quantifiable and tradable.

Advantage

Our topics are more flexible and faster to be verified than existing prediction markets such as https://polymarket.com/ For instance, everyone can express personal opinion on the debate like “will AI replace humans in 2030”, but we don’t need to wait until 2030 to see which side wins. We are just evaluated by how strong the arguments are.

Potential Consulting Space with Responsibility

Nowadays, if you get any advice on the internet for free, it’s hard to tell how much confidence the commenter holds. What if someone just randomly comments but does not take the responsibility? Get serious advice from high stake arguments! When you see a commenter bet $100 on its own comment, you know it’s serious.

Built With

Share this project:

Updates