OpenHedge
Real-time reliability scoring for every prediction market. When prices reflect one whale instead of the crowd, we catch it.
OpenHedge applies information theory — Shannon entropy, concentration indices, and convergence analysis — to every prediction market on Polymarket. The result is a single number: a Crowd Reliability Score from 0 to 1. When that number drops, you know the price has stopped being wisdom and started being one person's bet.
Live Market Scanner
Every 30 seconds, OpenHedge pulls live order book data from Polymarket's CLOB API, runs it through four analytical tiers, and outputs a Crowd Reliability Score for each market.
The scanner doesn't predict outcomes — it predicts whether the crowd's prediction can be trusted. A market priced at 70¢ with a CRS of 0.85 means the crowd is probably right. The same market with a CRS of 0.25 means someone has likely overwhelmed the signal.
Markets flagged as OPPORTUNITY have low CRS scores but high Kelly fractions — the math suggests a contrarian position may be justified.
| Market | Price | CRS | Signal |
|---|---|---|---|
| Fed pauses rate hikes | 58.7¢ | 0.84 | TRUST |
| BTC $200K by Dec | 28.1¢ | 0.51 | UNCERTAIN |
| OpenAI IPO 2026 | 67.8¢ | 0.78 | TRUST |
| SpaceX Starship orbit | 34.2¢ | 0.31 | OPPORTUNITY |
| GTA VI released 2026 | 71.5¢ | 0.72 | TRUST |
| ETH flips BTC 2027 | 4.1¢ | — | INSUFFICIENT |
| SOL surpass ETH TVL | 12.4¢ | — | DEAD MARKET |
| Apple foldable 2027 | 22.0¢ | 0.89 | TRUST |
Prediction markets are growing fast.
So is manipulation.
Prediction markets were supposed to be humanity's best forecasting tool — aggregating thousands of independent judgments into a single probability. But capital concentration is breaking that promise. When one trader can move a market by 15 points, the price stops being a forecast and starts being a position. And the downstream consequences are real.
Existing tools track price movements and volume. OpenHedge tracks something no one else does: whether the crowd generating those prices can be trusted at all. We don't compete with trading bots or analytics dashboards — we provide the reliability layer that tells you whether those tools' inputs are trustworthy.
How Prediction Markets Should Work
A prediction market works when thousands of independent traders each bring their own information, analysis, and judgment. The price of a contract — say, 65¢ — reflects the crowd's collective belief that an event has a 65% chance of happening. This is the Efficient Market Hypothesis applied to forecasting: the price aggregates all available information because each trader is incentivized to be right.
In a healthy market, no single trader dominates the order flow. The entropy of the distribution is high — meaning the "votes" are spread across many participants. The Herfindahl-Hirschman Index (HHI) stays low. The price converges gradually as new information arrives.
How They Break
Markets break when one participant — a whale — has enough capital to overwhelm everyone else's signal. They don't need to be right. They just need to be big. A single trader placing $45 million in directional bets can shift odds by 15+ percentage points, drowning out thousands of smaller, more informed traders.
Mathematically, the entropy of the market collapses. The HHI spikes because one entity controls a disproportionate share of volume. The price is no longer an aggregation of crowd wisdom — it's a reflection of one person's position and capital.
Why It Matters
When a manipulated prediction market price gets reported as "the crowd thinks there's a 78% chance," every downstream decision built on that number is compromised. Media organizations cite prediction markets as leading indicators. Traders use them to size positions. Policymakers reference them in public discourse.
The fundamental assumption — that the price reflects collective intelligence — is broken, but no one can see it from the outside. The price looks the same whether it emerged from 10,000 independent judgments or one whale's $50M bet. That's the problem OpenHedge solves.
What We Do About It
OpenHedge computes a Crowd Reliability Score from 0 to 1 for every active prediction market. It uses Shannon entropy to measure the diversity of the order flow, the Herfindahl-Hirschman Index to detect concentration, convergence analysis to track how prices are settling, and KL divergence for diagnostic context.
When the CRS drops below 0.4, we flag the market as potentially compromised. The score doesn't tell you what the right price is — it tells you whether the current price can be trusted as a genuine reflection of crowd wisdom. That distinction changes everything.
Four-Tier Architecture
Each market passes through four analytical tiers in sequence. Only Tier 2 determines the CRS — the other tiers filter, contextualize, and translate the score into action. The architecture is deliberately modular: each tier can be tested, validated, and improved independently.
Gate
Is this market alive enough to analyze? Markets with wide spreads or negligible volume produce unreliable entropy readings. The gate filters dead and illiquid markets before wasting compute on noise.
CRS Core
The heart of OpenHedge. Measures three signals — entropy anomaly (how unusual is the current distribution?), entropy rate (is the market converging or diverging?), and HHI (is volume concentrated?). These combine into a single CRS from 0 to 1.
Diagnostic
KL divergence quantifies how far the current distribution has drifted from a reference. It doesn't change the CRS, but it helps explain why a market scores the way it does — useful for investigation and backtesting.
Decision
Translates the CRS into actionable position sizing using the Kelly criterion. A high-CRS market gets standard Kelly. A low-CRS market gets scaled down or flagged as a contrarian opportunity — depending on the direction of the anomaly.
The mathematics behind OpenHedge draws from Shannon's information theory (1948), the Herfindahl-Hirschman Index used in antitrust economics, Kullback-Leibler divergence from statistical inference, and the Kelly criterion from optimal betting theory. Each component is well-established in its own field — OpenHedge's contribution is combining them into a real-time detection system for prediction market manipulation.
Interactive CRS Calculator
Drag the sliders to see how the three core inputs of the T2 CRS Core interact. Entropy anomaly measures how unusual the current order flow distribution is. Convergence rate tracks whether the market is settling toward a consensus or oscillating. HHI measures trader concentration — how much of the volume comes from a few large players.
The weights aren't equal. HHI and entropy anomaly matter most — a concentrated market with anomalous order flow is the strongest signal of manipulation. Convergence provides context but carries less weight.
This is a simplified version of the T2 computation. The production system uses additional smoothing, lookback windows, and Bayesian priors calibrated against historical manipulation events.
Data Pipeline
From raw order book data to actionable intelligence in under 30 seconds. The pipeline ingests live data from Polymarket's CLOB API via Polygon, runs each market through the four-tier analysis, and outputs reliability scores and Kelly-optimal position sizes. Currently tracking 847 active markets with approximately 2,000 data points per market per scan cycle.
Roadmap
Tech Stack
Built by Siew's Capital
OpenHedge is a research project by Siew's Capital, founded by Brayden Siew. It started with a simple question: if prediction markets are supposed to be the best forecasting tool we have, how do we know when they stop working?
The answer turned out to be information theory. Shannon entropy, originally developed for communication channels, provides a natural framework for measuring the "health" of a market's order flow. When entropy is high, many independent voices are contributing. When it collapses, someone is drowning out the signal.
This project is being developed for submission to the Yau Science Award 2026 — an international mathematics research competition for high school students. The research combines mathematical statistics, information theory, and financial economics into a practical tool that works on live data today.
The core insight is that manipulation leaves a mathematical fingerprint. When one entity dominates a market, the probability distribution of trades becomes less random — entropy drops. When multiple accounts controlled by one person are trading in coordination, concentration metrics spike. These signals are invisible to anyone watching the price chart, but they're clearly visible in the information-theoretic structure of the order flow.
OpenHedge is not a trading tool, and it's not a prediction model. It doesn't try to forecast whether Bitcoin will reach $200K or whether SpaceX will reach orbit. It answers a more fundamental question: can you trust what the market is telling you right now? In a world where prediction markets are increasingly cited by media, referenced by policymakers, and used by traders to size positions worth millions, that question matters.
Stages 1 through 5 of the terminal tool are built and computing CRS on live Polymarket data right now. The backtester is in development. Live trading integration is planned.