About & methodology

Where the data comes from

Everything is built on public sources. Market metadata, questions, categories, and resolution come from the Polymarket Gamma API; wallet positions and trade activity from the Polymarket Data API; settlement and balances are anchored to the Polygon blockchain. Data is refreshed continuously by background collectors. Figures can lag the chain by minutes and are only as accurate as the upstream sources.

How we find wallets

Candidate wallets are surfaced from four independent angles, so coverage doesn't depend on any single heuristic: trading volume (active wallets in our trade history), market winners (top traders mined from high-volume resolved markets — ground truth), co-occurrence (wallets that repeatedly position alongside known winners, i.e. shared information or framework), and curated Dune queries across volume, calibration, and recency. Surfacing a wallet is not an endorsement — it only becomes "tracked" once it clears the scoring bar below.

How we score a wallet

Scoring only counts BUY trades in markets that have resolved — the only trades with a known outcome. From those, we use the informative subset: entries priced below 90¢ are kept, near-certain entries (≥90¢) are dropped because buying something already priced at near-certainty carries no predictive signal (it's carry, not edge). A wallet needs at least 30 resolved informative trades to be scored at all.

On that subset we compute:

• Brier — calibration of the implied probabilities the wallet paid against actual outcomes. Lower is better; 0 is perfect.
• p-value — statistical significance that the wallet's edge is real, not luck, via a score test against the prices it paid.
• ROI — realised return per dollar across resolved bets.
• Churn ratio — trade frequency relative to position turnover; wallets above 10× are flagged as likely market-makers and excluded from the tiers, since their edge isn't directional.

Tier definitions

Sharp is the strict tier: significant, calibrated, and profitable. Profitable is significant and profitable but not necessarily well-calibrated. Both demand statistical significance and a minimum sample — a hot streak alone never qualifies. Flags also decay: a wallet with no on-chain trade in the last 30 days drops off the tracked smart-money surfaces until it trades again.

Selectivity in numbers: of the 3667 wallets we have fully scored so far, only 6% clear the Sharp bar and 9% the Profitable bar. The tiers are earned, not handed out.

Conviction, consensus & spheres

Conviction normalises a position against the wallet's own typical size — a $34 bet from a wallet that usually risks $9 is high conviction (≈3.8×), which is more telling than raw dollars. Consensus is where several tracked wallets independently land on the same outcome; we report it against the implied price (the edge), because raw hit-rate flatters crowded longshots. Markets are grouped into spheres (Politics, Crypto, Sports, Economy, Geopolitics…) so a wallet's edge can be read per domain — sharp in Politics and break-even in Sports are very different things.

Freshness — data accrues forward

Track records and consensus validation are captured going forward, not backfilled: positions and their resolutions are snapshotted as they happen, because once a market resolves and is cleaned up the position-level history is gone. Some panels will read "accruing" or show a small sample early on — that is honest, not a bug. Treat young samples as preliminary.

What this isn't

WhaleGraph does not give advice, recommendations, or signals to copy. A wallet's past performance does not predict its future; statistical edge is measured over a finite sample and can be noise. It is independent and not affiliated with Polymarket. See the Terms & Disclaimer and Privacy Policy for the full statements.

Get in touch

Questions, corrections, or feature ideas are welcome — see Contacts.