How we'll prove
we're honest.

Our approach, in depth

MispriceHQ exists to answer one question for every event contract we cover: is the market price right? To answer it credibly, a model has to form its own opinion before it ever looks at the market. That is the first and most important design rule of the engine we're building.

1. Price from scratch, never from the market

For each contract, the planned model assembles features from primary data — rate curves and macro releases, options-implied paths, polling aggregates, on-chain crypto flows, and platform-level liquidity — and produces an independent probability. The current market price is deliberately excluded from the model's inputs. If we fed the market price in, the model would learn to echo the crowd, and the gap between the two would be meaningless. Keeping the prior independent is what makes disagreement informative.

2. Disagreement is the signal

We define edge as the difference between our model probability and the market price, in percentage points. The sign tells you direction — whether we think the event is more or less likely than the market does — and the magnitude is a rough proxy for conviction. A contract trading at 34% that our model prices at 51% carries a +17pp edge. Edge is not a promise; it is a hypothesis that gets graded the moment the market resolves.

3. Honesty by construction

Most forecasting products show you only their hits. We intend to do the opposite: publish the score for every category, including the ones where we have no edge. When a category's model cannot beat the market baseline out-of-sample, the interface will say so — the markets there will sit visibly grey rather than glowing with false confidence. The boring, honest grey is the point. A research tool that only ever looks confident is not a research tool; it is marketing.

Why we lead with calibration, not win rate

A forecaster who says “70%” and is right 70% of the time is doing excellent work, even though they are “wrong” 30% of the time. Win rate — how often the higher-priced side happens — rewards confident guessing and punishes honest uncertainty. It is the wrong yardstick for probabilistic forecasts. The right yardstick is calibration: when we say 70%, does the thing happen about 70% of the time?

The standard way to measure this is the Brier score — the mean squared error between the predicted probability and the realized outcome (1 for yes, 0 for no). Lower is better; 0 is perfect, 0.25 is what you'd get by guessing 50% on everything. We grade against the market-consensus baseline in the same category, so the only number that matters is whether our Brier score beats the price you could have just taken off the screen. We explain this in depth in our guide to why Brier score beats win rate.

Calibration is also why the reliability diagram above is the first chart on this page. When the engine is live, every published probability becomes a dot on that chart the day its market resolves. Points on the dashed 45-degree line are perfectly calibrated; points above mean we were under-confident, points below mean we were over-confident. There is nowhere to hide on a reliability diagram, which is exactly why we use it.

What you should expect at launch

We would rather ship late and honest than early and wrong, so a few commitments are worth stating now:

• Sample data until it's real. Everything you see on the site today is placeholder data for design purposes, marked as such on every surface. We will not quietly swap in “live” numbers — the launch of real model output will be explicit.
• Publish only where we have edge. A category goes live only after its model beats the market baseline on held-out, already-resolved markets. Categories that don't clear that bar stay in research and are labeled accordingly.
• Every number ships with its calibration. Probabilities will always be accompanied by the category's track record, sample size, and confidence interval — never a bare figure.
• We disclose our limits. Base rates in some categories are genuinely noisy; regimes shift; tails happen. Where our confidence interval is wide, we'll show it wide.

Until then, the most useful thing on this site is the published methodology you're reading and the research library. Both are written to the same standard the models will be held to: sourced, dated, and honest about uncertainty.

Nothing on this page is financial advice. Event contracts are derivatives that carry risk of total loss. Model accuracy, once published, will describe past performance and is not a guarantee of future results.