Overextension Z-scorePROBABILITY MOVEMENT
What it is. Measures how unusual a recent probability move is compared with the market's own historical movement, computed in log-odds so a 5¢ move at 90¢ counts for more than 5¢ at 50¢. Baselines use a volatility floor and minimum history; displayed values are capped at ±12σ, and each score carries a confidence label.
How to read it. It is a research flag, not a recommendation. Very late event moves and effectively resolved markets are filtered from the leaderboard entirely.
When to ignore it. Ignore it when the confidence chip says LOW or BUILDING BASELINE, when volume is thin, or when an event is seconds from resolving — a score there is settlement mechanics.
Example. Example: +3.2σ HIGH on a live MLB moneyline with heavy volume means the probability moved about three times harder than this market's norm — worth investigating why.
Crowd Cost Basis / DistanceVOLUME STRUCTURE
What it is. Estimates where the most volume has recently changed hands by attributing snapshot volume deltas to 5¢ price buckets. The heaviest bucket is the crowd's approximate anchor; distance is how far current price sits from it; conviction is how concentrated volume is around it.
How to read it. A market far from its crowd basis is priced away from where most participants entered. The metric is approximate: hourly snapshots assign volume to the mid price at collection time.
When to ignore it. Ignore it when the THIN chip shows (too little profiled volume) or in brand-new markets where the profile is a handful of prints.
Example. Example: price 68¢, crowd basis 25¢, conviction 74% — most money entered far below; current holders are sitting on large unrealized moves.
Reversion Follow-ThroughHISTORICAL BASE RATE
What it is. After similar historical unusual moves (|Z| ≥ 2), how often was part of the move given back within the next window, and how much on average. Terminal settlement moves are excluded so resolution dynamics don't pollute the base rate; results split by move direction.
How to read it. This is the empirical scoreboard for the mean-reversion question, shown only when enough comparable events exist.
When to ignore it. Ignore it while the BUILDING SAMPLE chip shows, and never read it as a prediction for any single market — it is a population base rate.
Example. Example: 58% of upward spikes in Economics partially reverted (n=41), average 22% of the move given back — upward overreactions in this family historically cooled off.
Vol vs PeersMARKET BEHAVIOR
What it is. Current hourly logit volatility compared with similar markets in the same category, lifecycle phase, and expiry bucket (falling back to the expiry bucket alone when the cohort is small). Volatility always rises near resolution; phase-matching isolates markets that are unusually active for where they are in their life.
How to read it. High multiples flag contested, news-driven, or thin markets — a discovery feed for where probability is moving most.
When to ignore it. Ignore it for markets minutes from close and for cohorts of one or two peers.
Example. Example: 3.4× on an upcoming NBA game four hours before tip means pre-game probability is churning far more than comparable pre-game markets.
CalibrationSETTLEMENT ACCURACY
What it is. Compares historical implied probability buckets with actual YES settlement rates, priced at each market's last non-pinned snapshot so terminal 99¢ prints don't flatter the curve. Buckets under the minimum sample are grayed out.
How to read it. Persistent gaps describe systematic pricing behavior — the favorite-longshot pattern appears here when present.
When to ignore it. Ignore grayed LOW-N buckets and category curves with small settled samples.
Example. Example: the 70–79¢ bucket settling YES 62% of the time (n=180) suggests markets in this range historically overestimated the outcome.