MarketGrade: Trust Scores for Prediction Markets

Inspiration

Prediction markets are increasingly cited as sources of truth for elections, policy decisions, and major events. While markets often “know” earlier than polls or news, we noticed a critical blind spot: market probabilities are treated as equally trustworthy, even when the market structure behind them is fundamentally different.

A 70% probability from a deep, diverse market is not the same as 70% from a thin, whale-dominated one — yet existing tools make no distinction. We built MarketGrade to answer a more important question than what the market thinks: when should the market be trusted?

What it does

MarketGrade is a Hex-native decision intelligence app that assigns credit-style trust ratings to prediction markets.

It:

• Scores each market with a Market Trust Index based on liquidity, trader diversity, volatility, time-to-resolution, and historical accuracy
• Translates scores into intuitive AAA–BB style ratings
• Detects risky crowd behaviors like overconfidence, herding, and whale dominance
• Performs market autopsies to explain why high-confidence markets sometimes fail
• Adds an AI second-opinion layer that actively challenges market consensus
• Converts raw probabilities into risk-adjusted probabilities for decision-makers

The result: users can explore markets interactively and immediately see not just what the odds say — but whether they deserve belief.

How we built it

We built MarketGrade entirely in Hex using:

• Semantic modeling to standardize market metrics across categories
• A novel Market Trust Index combining five structural factors
• Crowd behavior analytics to detect manipulation and narrative lock-in
• Dynamic reactivity using Hex inputs and Jinja templating
• An AI explanation layer that adapts responses based on market structure and user questions
• App-mode design with collapsible methodology for a clean decision-maker experience

Due to limited public historical Kalshi trade data, we demonstrated the full methodology using synthetic data — designed to be directly transferable to authenticated Kalshi API or historical datasets.

Challenges we ran into

• Limited public Kalshi data required designing a framework that proves methodology without full trade-by-trade histories
• Balancing technical rigor with usability — ensuring advanced analytics remained intuitive
• Designing AI outputs that were skeptical and analytical, not just summarizations
• Creating an app-like experience while preserving transparency into the underlying methodology

Accomplishments that we're proud of

• Creating a credit-rating system for prediction markets — something that doesn’t exist today
• Demonstrating that markets with identical probabilities can have dramatically different reliability
• Building an AI layer that challenges consensus instead of reinforcing it
• Delivering a polished, fully interactive Hex app rather than a static analysis
• Showing how prediction markets can be used responsibly as decision-grade intelligence

What we learned

• Markets are often early — but their structure matters more than their confidence
• Liquidity and trader diversity are stronger predictors of reliability than category alone
• AI is most valuable when it questions assumptions, not when it repeats them
• Hex enables an entirely new class of analytics: interactive, explainable, and decision-focused

Built With

• hex