Prediction market statistics, from scored data

Do top prediction-market traders beat the prices they pay?

Across the 8,033 wallet-category records the board shows calibration for, trades resolve about 4.0 points better than the price paid. These are the top-ranked wallets, not the average trader, and not Polymarket prices at large. The full distribution, stated price versus resolved frequency:

Pooled: 8,963,896 scored trades from top-ranked wallets state 48.1% on average and resolve 50.3% (+2.2 points); in the 20–80 cent mid-range, 3,259,871 trades state 49.6% and resolve 53.5% (+4.0 points). Source: OVERROUND scored fills, verification gate passing, 2026-09-02.

Does profit measure skill? The all-time P&L top five, sized by evidence

Official all-time profit (Polymarket’s public leaderboard, fetched 2026-09-02) beside the number of independent resolved events behind it. Correlated legs in one event family count once.

Of the 10 largest all-time P&L wallets on Polymarket, 4 clear OVERROUND’s public evidence floor. Among those, the median skill-adjusted rank is outside the top 1,000. Ranking by profit and ranking by demonstrated accuracy answer different questions. Past accuracy does not predict future outcomes.

How accurate are OVERROUND’s own alerts?

Of the 799 featured alerts OVERROUND has sent subscribers, 680 have resolved. 427 resolved in the alerted direction and 253 resolved against it, a 63% rate on resolved alerts. The ledger counts one row per alerted position as of the 2026-08-05 correction (corrections log); earlier figures on this page counted re-alert snapshots as separate rows. Every alert and its grade is public at overround.pro/receipts. Voided markets are excluded from the record; unresolved alerts are counted as sealed and publish in full at resolution. The ledger publishes its own arithmetic identity, so edits would be detectable. As of 2026-09-02.

Do the rankings hold up out of sample?

OVERROUND published pre-registered, rolling-origin backtests of its own rankings on 2026-07-15, committed before results existed to publishing either way. In six of nine quarterly windows, alert-eligible wallets went on to beat the prices they paid by about 0.7 to 1.4 cents per share with confidence intervals clear of zero (largest window n = 2,236,857 wallet-events). In the weakest window they did not demonstrate forward edge, and the report leads with that. Full tables, intervals, and limitations: overround.pro/research.

Coverage

The public board currently ranks 489,978 wallet-category records across 61 categories, computed nightly from public on-chain fills. Live counts: overround.pro/leaderboard. As of 2026-09-02.

Sources

• OVERROUND scored fills (primary dataset): public on-chain Polymarket activity graded per the methodology; the graded alert slice is public at /receipts with a JSON mirror. Continuously updated; page snapshot as of 2026-09-02.
• Polymarket public leaderboard: polymarket.com/leaderboard, the source for official all-time P&L figures. Accessed 2026-09-02.
• Polymarket public APIs: docs.polymarket.com, the documented public market and data endpoints used for external cross-checks. Accessed 2026-09-02.

Citing this page

• Cite as OVERROUND (overround.pro) with the page date. Numbers update as markets resolve; the date pins the snapshot you quoted.
• License: CC BY 4.0. Reuse and republish these statistics freely with attribution.
• Methodology behind every number: the methodology page (price-relative scoring, independence rules, exclusions, limitations).
• Questions or data requests: hello@overround.pro.