How to track Polymarket whales and smart money

What the exchange publishes, what it does not, and where it gets read wrong.

Polymarket is unusually legible: positions settle on a public blockchain and the exchange serves read-only APIs over that record, so any wallet's open book, entry prices, settled results and trade history are readable by anyone. Legible is not easy, though: the number most often quoted about a wallet — its lifetime profit — comes out wrong in almost every naive implementation, by millions of dollars, in the wrong direction.

What a wallet actually publishes

The positions endpoint returns the current book: every holding with size, average entry, current price, mark-to-market and realised components, and a redeemable flag. Closed positions returns the settled record, trades and activity return the timestamped tape, the leaderboardranks wallets by profit over four windows and eleven categories, and public profile data carries a wallet's creation time. Prices and order books come from the gamma and CLOB APIs.

What is not public matters more. A wallet is an account, not a person. Intent is invisible, since the chain shows the position and never the reason, and so is anything held off the venue — a directional-looking position may be one leg of a hedge. Resting orders carry no owner; the book is aggregate levels. And none of this exists on Kalshi, which publishes no per-user data at all: Kalshi signal is anonymous aggregate flow off the public tape, never an identified trader.

Why lifetime profit cannot come from the positions endpoint

That endpoint returns open positions plus resolved ones nobody claimed. Winners are redeemed out of the book automatically; losers have nothing to claim, so they stay. The settled residue is therefore almost entirely unredeemed losses, and summing it is confidently wrong: checked live, a wallet with more than eight million dollars of lifetime profit read as down $1.37M, because two unredeemed losers were the only settled rows left in its book.

Total profit has to come from Polymarket's own profit endpoint, the figure the exchange shows on the profile. Unrealised profit then comes from genuinely open positions only — not redeemable, priced strictly between zero and one — and realised profit is the difference. The settled feed has a matching trap: it sorts by realised profit descending, so a naive first page is a wallet's biggest winners. Sort by time. Redemption breaks both.

The CLOB, and why one order arrives as many rows

Polymarket runs a central limit order book— not a bookmaker, not an automated market maker. You trade against other participants' resting limit orders, matched by price then time. Each outcome is its own token with its own book, best price listed last. A market therefore has no price but a ladder, and the number that matters is the executable ask on the contract you want.

That structure is also why raw tape misleads. One human order routinely lands as many fills seconds apart: a single wallet's sell campaign once printed as sixty-one near-identical clips, which an unclustered tracker renders as sixty-one whale trades. We merge fills sharing a wallet, market, outcome and side whose quiet gap is under three minutes — dollars sum, price becomes the volume-weighted average, and the row shows how many clips it merged.

What the Sharp Score actually computes

Profit alone ranks bankroll and luck as readily as skill, so wallets get re-scored. The Sharp Score starts at 50, adds three bounded terms and clamps to 1–99: profit up to ±24 through a saturating curve, where $250K of window profit collects three quarters of the term; return on volume — profit divided by dollars traded — up to ±18 on the same shape, saturating faster, since 20¢ per dollar traded already takes three quarters of it; and hit rate up to ±8, only once at least five positions have settled. The bands are fixed at 90 elite, 75 sharp, 60 solid, 45 mixed, below that cold.

Its limits are structural. It looks backward, so a skilled newcomer stays invisible until trades settle; full enrichment runs for the top of each fifty-row board, not every row; a single wallet's page has no leaderboard volume figure to divide by, so its efficiency term uses a proxy denominator; and maker-shaped wallets are flagged, not removed, because over $3M of volume at under 0.8% return on volume is a rebate business, not a read. Past performance does not guarantee future results; the methodology page has the rest.

negRisk, and multi-outcome events

An event flagged negRiskhas an exhaustive, exclusive outcome list: exactly one leg resolves YES, enforced by the adapter the flag is named for. Without it, leg prices summing under a dollar may just mean the winning outcome was never listed. It also changes the unit of analysis — a wallet holding NO across three legs is shaping one position over one outcome space, not expressing three opinions — and every leg still needs its own executable ask, since one archived leg carrying a phantom 99¢ quote with no book behind it poisons the whole event. Kalshi's comparable flag is weaker by design: mutually exclusive there means at most one leg resolves YES, not exactly one.

Reading a whale trade without over-reading it

Size is not conviction. A large print can be one leg of a hedge, maker inventory offset within minutes, a position being unwound rather than opened, or — before clustering — one order counted many times. Read the entry price, not just the stake: winning positions entered near 45¢ is skill, winning them at 90¢ is paying for favourites. Read the whole book behind the trade, because a leg of a spread looks directional alone. And ask why this position would exist in this market now.

Where each of these lives

Each job above is a tool, and each has an open page documenting what it computes and cannot tell you:

• Sharp Leaderboard — the public leaderboard re-ranked by Sharp Score, maker-shaped wallets flagged.
• Live Sharp Feed — tracked wallets' trades as they print, fills already clustered into one decision.
• Portfolio Tracker — one wallet's book: official lifetime profit, open marks and settled record, kept apart.
• Sharp Consensus — where tracked capital agrees, weighted by score, dust and near-resolved rows dropped.
• Insider Radar — fresh wallets taking large, concentrated positions — a question, never an accusation.
• Overround Scanner — negRisk events priced leg by leg off executable asks, untradable legs excluded.
• Watchlist — the few addresses worth following, read as one combined book.

The tools are members-only at $49.99/mo — one plan, no free tier, no trial. Their documentation, the glossary, the guides and the methodology stay open.

The other venue works nothing like this one: Kalshi publishes no per-user data at all, so a Kalshi "whale wallet" is not a thing that exists — only anonymous flow off the public tape. If you are weighing one against the other, the venue comparison puts the fee models, resolution rules and multi-outcome semantics side by side.

WhaleTracks is an independent analytics product, not affiliated with or endorsed by Polymarket. It reads Polymarket's public, read-only APIs and public on-chain data, places no trades and holds no funds. Informational analytics, not financial advice.