Prediction markets are quietly becoming one of the best signals we have. When you make people back an opinion with money, the noise falls away and what's left is close to the truth. But the data underneath is scattered, primitive, and impossible to trust at the moment it matters.
We started Mostly Right because we kept hitting the same wall. We wanted to study how these markets form prices, where they're wrong, where they're early, and every time we started, we were rebuilding the same plumbing from scratch. Pulling weather observations that three public feeds disagree on. Reconciling settlement rules by hand. Discovering, days later, that a backtest didn't reproduce live because a number had been quietly rewritten upstream.
So we built the layer that should already exist: one client SDK for prediction-market data and the real-world data that settles it: fetched on demand, kept raw-as-reported, validated at fetch time, and reproducible to the row. Not a trading app. Not another exchange. The infrastructure underneath. Increasingly used not by a person at a terminal, but by an AI agent doing the research. We built for that.
We chose to focus on outcomes you can verify: weather first, then prediction-market settlement. Things that resolve against real data, not opinion. The builders who show up now are deciding what prediction markets become, and we'd rather they become trusted infrastructure than a punchline. Read our full mission →
