Model Confidence vs Market Probability

Any yes/no question traded on a prediction market can be viewed two ways. The market-implied probability comes from the contract price: what participants, in aggregate, are currently willing to pay. A model confidence comes from structured analysis of the available evidence, independent of what the market is doing.

Neither is the "right answer." They are different instruments. Comparing them is useful precisely because they are built differently.

A market price reflects everything its participants know and believe, weighted by how much they are willing to commit. In a liquid market this can be a strong summary of public information.

It also carries the market’s habits: reactions to headlines, positioning, fees, and the biases covered in how prediction market probabilities work. In a thin market, a price may reflect only a handful of orders.

A model confidence is built by examining evidence for and against a proposition and allocating confidence accordingly. In QSE, that output is a confidence allocation across five states rather than a single number, so it shows how settled or unsettled the evidence is, not just which side it leans toward.

A model does not see order flow or trader sentiment. It sees evidence. That makes it slower to react to noise, and also means it can miss information the market has already absorbed.

When the model’s confidence and the market’s implied probability are close, the analysis is broadly consistent with the price. When they differ materially, that difference is called divergence. Some tools describe a meaningful, filtered divergence as edge.

For example, if a market prices a Yes contract at 40 cents and an evidence-weighted analysis lands meaningfully higher, the two readings disagree. That disagreement is a reason to look closer, not a conclusion.

Divergence measures disagreement, not correctness. The market may know something the evidence does not yet show, such as late news or informed participants. The model may be weighting evidence the market has not priced. Only settlement reveals which was closer, and even then a single outcome says little. Reliability shows up only across many resolved cases.

That is why QSE publishes a public track record and applies conservative limits before anything is published. Unusually large claimed divergence is treated with more suspicion, not less.

Public explanations describe the approach: evidence is gathered, factors are weighted, and confidence is allocated. The specific weights, calibration settings and internal scoring are not published. What you can inspect is the output, the top contributing factors, and the resolved outcomes over time.

Edge Spotter signals are published when QSE’s confidence allocation and a Kalshi market price diverge within defined limits, and every signal is later settled against the real outcome. Market Console lets you pick a market yourself and see the model and market readings side by side. Terms such as divergence and implied probability are defined in the glossary.

This is probabilistic analysis, not financial advice. It is designed to help you reason about uncertainty, not to tell you what to do.