Methodology

Logit-space movement. Prices are clamped to 1–99¢ and converted to log-odds. All movement statistics are computed on logit changes.

Event lifecycle. Markets classify as upcoming, live, late-live, post-event, pinned, continuous, or unknown. Event starts are parsed from date-coded tickers where possible. Only rankable phases reach leaderboards; exclusions carry stated reasons.

Overextension. The recent window's logit move is standardized against the market's own history, with a volatility floor, minimum-history gates, a ±12σ display cap (raw values preserved internally), and a confidence label.

Vol vs peers. Hourly logit volatility divided by the median of markets in a similar lifecycle/expiry bucket.

Crowd cost basis. Snapshot volume deltas assigned to 5¢ price buckets; the peak bucket is the anchor. Approximate by construction and labeled low-confidence on thin profiles.

Reversion. After |Z| ≥ 2 moves, the following window is measured for partial retracement. Pinned and post-event markets are excluded; small samples display as building.

Calibration. Settled markets' price buckets vs actual YES rates, sample sizes always shown, low-sample buckets greyed.

Data quality. Snapshots are collected hourly from public market data; dormant markets are not tracked. Metrics carry confidence labels, and no metric on this site is investment, trading, financial, legal, or tax advice.