Polymarket Scanner

Convexly's published board is a closed set of 50 Polymarket addresses, selected once on 2026-04-19 and never added to. A cron rescores those same 50 rows every day at 06:00 UTC; it does not scan Polymarket for new wallets. Convexly surfaces that board, ordered by scored resolved-position count with a publication floor of 20 scored resolved positions, at convexly.app/research/wallet-rankings, the top 20 by PnL vs the top 20 by Edge Score side by side at convexly.app/truth-leaderboard, and a free live single-wallet lookup for any 0x address at the Polymarket wallet analyzer. Your first wallet needs no signup; more wallets free with an account. No private key, no wallet signature.

What the scanner ranks and why

Polymarket's own leaderboard ranks by realized PnL, which is the natural thing to rank by and which our V1 methodology paper reports is a noisy proxy for cross-sectional behavioral skill. The Spearman rank correlation between calibration (Brier) and realized PnL on the full 8,656- wallet V1 cohort is +0.148. In the original V3b fit, the Spearman rank correlation between the composite and signed log PnL, measured out-of-fold with 10,000-shuffle Fama-French permutation null, was +0.514 (p less than 0.0001); the constants were refit on 2026-07-13, so that fit statistic does not describe current scores (dated method-change note at convexly.app/methodology). Every row on the board carries the composite, though the board itself is not arranged by that composite rather than by the composite. In its one out-of-sample forward test, a pre-set per-wallet temporal holdout, the frozen composite held a Spearman of +0.11 (95% CI [0.05, 0.18]) with forward PnL, below the +0.30 bar set in advance; treat Edge Score as a descriptive behavioral profile, not a validated skill ranking.

The three pillars that drive the scan

• Posture (z-scored negation of skill_brier): captures wallets whose profit is earned despite imprecise calibration. Polymarket V1 fitted coefficient +0.79.
• Conviction (z-scored log concentration): captures portfolio concentration in a single dominant event. Polymarket V1 fitted coefficient +2.72, the dominant pillar.
• Discipline (z-scored negative log position count): captures fewer, larger resolved positions. Polymarket V1 fitted coefficient minus 1.15.

Coefficients are re-estimated per venue. Manifold coefficients diverge meaningfully, including a sign flip on discipline, per the V1-M extension paper.

How often the scanner refreshes

A GitHub Actions cron runs daily at 06:00 UTC. It rescores the 50 rows already on the board, rebuilds the JSON payload, and redeploys the static board pages. The cron passes the same file as both its candidate list and its output, so it rescores the existing members and never screens Polymarket for new ones: the 50 addresses were selected once on 2026-04-19, from a 2026-04-15 snapshot, and none has been added since. That first selection required at least 5 resolved positions and at least $5,000 realized PnL, so membership was conditioned partly on having made money and no count taken over these rows describes Polymarket wallets generally. Individual wallet lookups through the single-wallet analyzer are a different mechanism: those are live against the Polymarket Data API with a 24-hour cache on each (wallet, date) key. Off-board 0x addresses are still analyzable on demand, but they do not enter the board and their scores do not shift the reference distribution. Corrected 2026-08-10: freezing the cohort keeps the results recomputable by Convexly, not by a reader. That cohort has never been published, so nobody outside Convexly can rerun the V1 fit today.

Scanner vs single-wallet analyzer

The scanner is the published 50-wallet board. The Polymarket wallet analyzer is the single-wallet lookup. Same underlying methodology. Same coefficients. Different entry point. Most traders who search “Polymarket scanner” actually want one of three things: the 50-wallet board, rescored daily (public), the PnL-vs-Edge comparison view (public), or an ad-hoc lookup for a specific 0x address (public; first wallet free with no signup, more wallets free with an account).

Institutional access to the full reference cohort

API access to the full reference cohort with per-pillar scores, wallet labels, and historical snapshots back to the 2026-04-15 selection snapshot is available through the institutional pilot path. See convexly.app/enterprise for details, or read the open methodology at convexly.app/research/edge-score-methodology-v1.