Prediction Markets methodology

Market probability reflects the current price on the underlying prediction market (e.g. Polymarket) and changes continuously as people trade.

CoinAugur AI consensus is an independent estimate produced by multiple leading AI models. It is generated far less frequently than the market moves, and is shown as discrete snapshots — never interpolated as if it were continuous.

By default, models are given the exact question, the outcome list, the resolution rules and source, and the relevant dates — but not the current market price. This reduces anchoring, so the consensus is a genuine independent view to compare against the market rather than an echo of it.

Each model returns a probability for every outcome, which is validated (outcomes complete, probabilities in range and summing to one, credible sources present). Invalid responses are excluded. We then take the median probability per outcome across the valid models and renormalise the distribution to 100%. Median aggregation is robust to a single extreme model.

We measure how tightly the models cluster on the favoured outcome (agreement) and their overall spread (dispersion). Confidence is derived from agreement, the margin between outcomes, evidence quality, the number of valid models and time to resolution — it is not the models' self-reported confidence.

Provider prices refresh on a fast cadence. AI consensus refreshes on a configurable schedule (e.g. daily) or when a market moves materially, subject to a cooldown to control cost. Consensus never regenerates on a normal page load.

For sports, the AI consensus is a pre-event estimate. It is frozen at event start and clearly labelled — we do not publish in-play AI probabilities from a pre-match model. Live market prices and scores may continue to update.

Markets resolve strictly by the provider's published rules. Every AI consensus snapshot is preserved and never overwritten. When a market resolves we record the official outcome and stop refreshing its consensus.

On resolution we score the consensus, the market and each model against the actual outcome using multi-class Brier score and log loss. We will not publish accuracy claims until enough markets have resolved to be meaningful.

• Probabilities are estimates and outcomes are uncertain.
• AI models may be wrong, and market probabilities may change.
• The difference between AI and market probability is not a guaranteed edge, mispricing or trading signal.
• CoinAugur does not execute trades or bets and does not custody funds.
• Provider links are external, and market data may be delayed.
• Resolution is controlled by the underlying provider's published rules, not by CoinAugur.