Glossary

The machine-readable version of everything here is GET /v1/metrics/catalog. Direction matters: for some metrics lower is better — the catalog declares it per metric.

Win rate vs prediction accuracy

Two figures that sound alike and answer different questions about the same positions. Every page shows them under these names.

Win rate — Made money: the share of settled positions that ended in profit. The money question. On a trader's page it counts the positions they closed (sold out, or held until the market settled) when that rate is available.

Prediction accuracy — Picked the winner: the share of settled positions where the side they bought won, whether or not they still held it. The pick question. It is not adjusted for the price paid, so a wallet that buys favourites scores high by construction; read it next to the edge vs the odds.

The two can disagree: selling a winning side early at a loss lowers win rate but still counts as a correct pick, and selling a losing side at a profit does the reverse.

Calibration (the skill core)

Brier score — mean squared error between entry probability and the realized outcome. Buy YES at 70¢ and it happens: error² = (1 − 0.70)². A coin-flipper scores ~0.25; elite forecasters sit near 0.15–0.18. Lower is better, and it's hard to fake — luck averages out over enough resolved markets.

Calibration / reliability diagram — bucket a wallet's entries into 10 probability bins and compare each bin's implied probability to the realized outcome rate. "Says 70% → happens 71% of the time" is a calibrated forecaster.

Calibration error (ECE) — the 10-bin summary: sum of |accuracy − confidence| per bin. Lower = better calibrated.

Brier skill score — 1 − Brier/Brier_base_rate; positive means better than always guessing the base rate.

Brier reliability / resolution — the Murphy decomposition: reliability (lower better) isolates miscalibration; resolution (higher better) rewards forecasts that actually discriminate outcomes.

Bonder-adjusted Brier — Brier excluding entries above p = 0.85, neutralizing the "buy 97¢ favorites" strategy that game-ifies raw Brier.

Peer score / relative log score — Metaculus-style log-probability versus the crowd: did you know more than the market did?

Alpha

Alpha vs mid — size-weighted gap between the pre-fill mid and your fill price, signed by side. Positive = you systematically buy below / sell above where the market was. The purest "edge at entry" measure.

Average entry edge — mean (resolution probability − entry probability), signed. Did the things you bought at 30% actually happen 45% of the time?

Information coefficient (IC) — correlation between entry edge and realized outcomes; consistency of being right, not just magnitude.

Counter-consensus alpha — alpha restricted to fills against the prevailing direction; the contrarian's report card.

Early/late-entry alpha — alpha sliced by when in the market's life you entered (first 25% vs last 25%).

Risk

Sharpe / Sortino / Calmar — risk-adjusted return: mean daily return over volatility (Sharpe), downside-only volatility (Sortino), or max drawdown (Calmar), annualized.

Max drawdown — worst peak-to-trough equity decline. CVaR (5%) — the mean of your worst 5% days; the tail you should actually fear.

P&L (the commodity numbers)

Realized / unrealized / total P&L — closed-position profits; mark-to-market on open positions; their sum. Computed PM-methodology- faithful (avg-cost, transfers ignored) — see Trader profiles.

Profit factor — gross profit / gross loss. ROI — capital-weighted return. TWRR — chain-linked daily returns, removing deposit/withdrawal bias.

Labels

Composite score — the weighted blend of the above that orders the default leaderboard; the Builder lets you make your own.

Smart-money tags — behavioral labels (sharp, calibrated, whale, early-mover, …) with scores and rationale, from GET /v1/trader/{proxy}/tags.

pnl_under_review — this wallet's P&L is in the reconciliation tail and being re-verified against on-chain truth; rankings exclude it until it clears.