Polymarket whale tracking: telling a wallet that matters from a wallet that is merely large

Polymarket whale tracking is usually sold as a plumbing problem: get the trades faster, get them in a channel, get an alert when someone spends six figures. The plumbing is the easy half. The hard half is deciding which wallets are worth a notification at all — and on this venue, the honest answer is that almost none of them are, and that the criterion which separates the few from the many is not size.

This article is about that criterion. What a whale is on a venue where anyone can open an address in a minute, how a wallet earns the right to be watched, what a public leaderboard gets wrong and why, and what following a large wallet structurally cannot tell you. Every figure below is one of three things: measured in our own database on 17 August 2026, read from a live call to our public API with the time of the call, or attributed to a named source with its link. Figures that move — counters that grow with every hour of ingestion — are either rounded to an order of magnitude or given with the exact minute they were read, because an exact count published without its instant is false an hour later.

What Polymarket whale tracking actually measures

A "whale" is not a category the venue defines. It is a label an observer applies, and the label is only as good as the rule behind it. Three rules are in common use, and they are not equivalent.

Notional size. A wallet is a whale if its tickets are large. This is what almost every alert bot uses, because it needs nothing but the trade stream. It is also the weakest, for a reason developed in the next section.

Cumulative volume. A wallet is a whale if it has traded a lot in total. Better, in that it filters out the account that made one large trade and vanished — except when the volume figure is broken, which on this venue it frequently is.

Realised profit. A wallet is a whale if it has made money. Intuitively the right rule, and the one people ask for. Its problem is not coverage: on 17 August 2026, 79,751 of our 145,141 tracked wallets carry a non-zero profit and loss figure. Its problem is that the figure arrives late, and that it sits in the same row as market counts that do not survive ten seconds of inspection — the fourth section of this article is about exactly that.

None of the three measures what an observer actually wants, which is: does this wallet's arrival tell me something about where the price is going? That is a fourth quantity, it has a name — alpha — and it is measurable directly, without reference to size, volume or claimed profit. Getting there is the rest of this article.

One boundary first, because it shapes everything that follows. We track wallets, not people. There is no address clustering in our data: measured on 17 August 2026, wallets.tags and wallets.metadata are empty on 145,141 rows out of 145,141, and no table or column in the database carries a notion of entity or cluster. A single operator running eight addresses appears as eight wallets, and nothing we publish can tell you otherwise. Any product that claims to have merged addresses into people should be asked how, and on what evidence.

Size is not conviction: a $20,195 ticket at 0.999

Here is a real observation, taken from our trade table on 17 August 2026 at 01:18:18 UTC. Two wallets classified as whales traded the market Will Bitcoin dip to $60,000 August 10-16?, in the same second:

• —one bought NO for $20,195.45 at a price of 0.999;
• —the other sold NO for $15,763.60 at the same price.

A size-triggered alert fires twice here, and both alerts are worthless. The market had, at that moment, essentially resolved: the window in the title had closed, and its stated end was 04:00 UTC the same morning — a little under two hours and forty-two minutes after the trades. Buying NO at 0.999 is not a forecast that Bitcoin will not dip to $60,000. It is a wallet putting twenty thousand dollars to work for a tenth of a cent per contract — a capital operation, not an opinion. Selling at the same price is the mirror image: someone taking capital out a few hours early rather than waiting for resolution.

Twenty thousand dollars is a large ticket by the standards of the flow we record — our base holds trades of $1,000 and up, nothing below, and its smallest recorded amount is exactly $1,000. It held 2,123,153 trades with a usable amount when we counted, at 10:43 UTC on 17 August 2026, and it holds more now: ingestion is continuous. Of those, 2.0% — more than 42,500 tickets — are $20,000 or more, against a median ticket of $1,784. And that large ticket carries no directional information whatsoever.

Meanwhile, two and a half minutes earlier, at 01:15:43 UTC, a wallet classified as a fish put $11,188.62 into NO at 0.9685 on Will Bitcoin reach $140,000 by December 31, 2026? — a market that runs to the end of the year. Near-certain pricing again, a completely different horizon, and a label a size-based feed would have filtered out. Neither of these tickets survives the price-band filter described below, and that is the point: the label tells you nothing about whether a ticket is a forecast, and the price tells you most of it.

This is the central failure of size-based Polymarket whale tracking, and it is not an edge case. Nearly half of the flow we record happens at near-certain prices: 48.4% of those trades are struck at 0.95 or above. Filtering by size does not steer you away from that region, because size does not vary with it: the median ticket at 0.95 and above is $1,765, against $1,800 inside the 0.05–0.95 band. Among the tickets of $20,000 or more, more than 19,000 sit at 0.95 or above. If you alert on notional, close to half your attention goes to trades that had already stopped being predictions.

The correction is mechanical and costs nothing: filter by price band before you filter by size. Our own alpha measurement discards every trade outside 0.05–0.95 for exactly this reason. Prices at the extremes are mechanically likely to be right, and including them would let a wallet accumulate a spotless record by buying certainties.

How to identify a wallet worth tracking, step by step

The measurement we publish is alpha: the price movement that follows a wallet's trades, in the direction of its position. It does not depend on the wallet declaring anything, it does not depend on resolution, and it does not depend on the venue's own statistics being correct. Here is the procedure, in the order the steps have to happen. Every count below was re-measured on 17 August 2026.

1. Fix a window and state it. Ours is trades from 2026-04-01 to 2026-06-24. A wallet's alpha is a property of a period, not a permanent attribute, and a number published without its window is not checkable. State also where the data actually sits inside the window: of the 323,647 trades that fall in ours after the price filter, 322,056 are from April. Our ingestion has holes, and a window that hides them is a window that lies.

2. Restrict the price band. Trades priced between 0.05 and 0.95 only. Without this, the extremes dominate — they are half of what we record — and the measurement becomes a count of how much cheap certainty a wallet bought.

3. Measure the movement after each trade. For every trade, take the price move up to the next trade in the same market, signed by the direction of the position. This is the raw observation. There were 307,540 of them in the window.

4. Aggregate by market before you aggregate by wallet. This is the step everyone skips, and skipping it invalidates everything downstream. Trades in the same market are not independent observations: a wallet that placed 98 orders in one trending market has 98 correlated data points, not 98 pieces of evidence. Skip this correction and the ranking is topped by exactly that wallet — 98 observations, all in a single market, an average move of 80.0 points and a t-statistic of 3,921. That is a number to be read as a bug report, not as a result.

5. Require breadth and significance. At least 25 distinct markets and a t-statistic above 2.5, plus a plausible volume figure. Of 16,551 wallets with at least one scored observation in the window, 604 clear the 25-market threshold, 64 of those also clear t > 2.5, and 62 survive once wallets carrying an impossible volume are removed. That is 0.37% of the wallets evaluated, and 0.04% of the 145,141 wallets we track. The filter is not a shortlist. It is a rejection of essentially everybody.

One thing has to be said about what comes out of that. The per-wallet figures we publish are in-sample: they are measured inside the window stated in step 1, and we label them as such. A wallet at the top of that ranking moved price during that window; whether it does so in the next one is a separate question, and not one the ranking answers.

The wallets that come out of this process do not look like the ones a volume leaderboard surfaces. The eight at the top of our published table carry between 26 and 53 distinct markets and t-statistics between 4.2 and 10.1. Their cumulative traded volumes, read on 17 August 2026, span $363,509 to $2,212,976 — real money, but nowhere near the top of any volume ranking, where the leaders are three orders of magnitude larger. Breadth and consistency are what the filter selects for; size is not in the criterion at all.

Why a public whale leaderboard is mostly noise

Our own API exposes a leaderboard, and we do not put it on the marketing pages. The reason is worth spelling out, because the same defects are present in every public ranking built on this data.

Measured on 17 August 2026, across the 145,141 wallets in the table:

• —every single one carries a win rate, which is the first defect rather than a feature: the column is never empty, so its presence tells you nothing. 54,913 carry a non-zero value;
• —36,354 wallets show a win rate of exactly 100% — and 18,914 of them have traded exactly one market, 28,710 three or fewer. Only 557 of the 36,354 have twenty-five markets or more behind the number;
• —21,257 wallets carry a smart score above zero, while the seven scoring 80 or above have between 5 and 58 distinct markets — three of the seven under ten markets;
• —14 wallets carry a total volume above $100 million, and 11 of those have at most one market. The wei that were never divided down are still in the trade table too: 2,209 trades carry a raw amount, the largest of them 8.5 × 10⁷⁰ USDC. Those are precisely the rows that rise to the top when you sort by volume.

That last one is not theoretical. Calling GET /api/leaderboard?limit=5 on 17 August 2026 at 10:24 UTC returned, in third place by volume, a wallet classified as a whale with a total volume of $945,696,387.85, a markets count of 1, a trades count of 1, a profit and loss of 0 and a win rate of 0. Fourth place: $694,563,782.35, one market, zero trades. Fifth: $590,909,090.91, one market, one trade. A ranking whose top five contains three single-trade wallets with nine-figure volumes is not a ranking of anything.

The two wallets that did look substantive in that same call showed win rates of 73.8% and 88.6% — over 662,694 and 331,093 markets respectively. Before those percentages mean anything, someone has to explain how a single address traded more than half of the database's 1.2 million-plus markets. We do not have that explanation, so we do not publish those figures as a claim.

The general rule, and it applies well beyond this venue: a rate without its denominator is not a measurement. A wallet at 100% over one market and a wallet at 62% over four hundred are printed in the same column, in the same font, and one of them is information. This is the same discipline we apply to our own signal accuracy, worked through in prediction market signals: what a signal is worth without its denominator.

What Polymarket whale tracking cannot tell you

Even a wallet that passes every filter above tells you less than the alert implies. Four structural limits, none of which better data would fix.

You cannot get their price. By the time a trade is public, it has already moved the book. The alpha we measure is the movement after the trade — which is to say, it is precisely the part the follower is buying into rather than capturing. Against the cost of taking liquidity on this venue, which we publish as roughly 3.5% of notional alongside our calibration table in the public register, a follower needs the subsequent move to exceed both that cost and the slippage the original trade caused.

You never see the exit. Positions on Polymarket are closed by selling or by resolution, and a wallet that quietly sells back into a rally leaves a much smaller footprint than one that bought it. Alerts are systematically biased toward entries.

You cannot see the hedge. A wallet buying YES on one market may be flattening exposure held elsewhere, on another market, on another venue, or in a form that never touches a public chain. What looks like conviction may be risk reduction.

You cannot see the person. As established above, there is no address clustering in the data. One operator across six addresses reads as six independent wallets agreeing with each other, which is exactly the pattern a naive consensus indicator would score highest.

Add to that the general result about who makes money here. Bloomberg reported on 28 April 2026 that across roughly two million Polymarket addresses, traders were down about $131 million net while automated participants took gains out of the same markets (the article is here). The mechanisms behind that number — cost on notional, long-shot pricing, resolution wording, sizing — are worked through in why most prediction-market traders lose money, and every one of them applies to a follower just as much as to the wallet being followed.

A whale-tracking workflow that survives contact with the data

What is left after all that is narrower than the marketing, and more usable.

• —Rank by alpha, not by size. Size tells you a wallet is capitalised. Alpha tells you its arrival preceded a move. Only the second is a reason to look.
• —Read every rate with its denominator. Distinct markets, not trades. A wallet with 400 trades in 6 markets has 6 observations.
• —Discard the extremes. Trades outside 0.05–0.95 are capital operations far more often than they are forecasts, and trades at 0.95 and above alone are 48.4% of the flow we record. The $20,195 ticket at 0.999 above is the archetype.
• —Watch clusters, not individuals, and know what a cluster is worth. Several independently qualified wallets converging on the same side within a short window is a stronger observation than any single ticket — with the caveat that they may be one operator, and nothing in the data can rule that out.
• —Price the follow before you take it. 3.5% on the way in, plus the move you missed. If the position needs a 20-point move to break even after costs, that is the trade you are actually considering. The position size calculator makes the arithmetic explicit, including the part where a positive edge still gets you liquidated at the wrong size.
• —Re-run the ranking. Alpha is measured over a window and decays. A leaderboard computed in April and displayed in August is a historical document.
• —Read the resolution rules. A contract resolves on wording applied to a named source at a stated time. A whale who read the rules and a follower who read the title are not in the same trade. The mechanics are set out in how prediction markets work.

Where EdgeMarket fits, and what each tier actually opens

We publish measurements, not forecasts, and the tier boundaries follow the cost of producing them rather than an artificial scarcity.

Observer is free. It opens the public register with its full calibration table — including the bands where our own signal is worse than the market price — the ten-row live tape, and the wallet audit sheet for any address, at five trades of depth.

Operator, at $39 a month or $390 a year, opens what requires the ranking to be computed and maintained: the alpha leaderboard, the 50-row feed desk with wallet aliases and classifications, audit depth at 200 trades, the operator consensus view, and alert delivery on Telegram at the trade size and market categories you set.

Principal, at $79 a month or $790 a year, adds the REST API and outbound webhooks — the tier for running this measurement yourself rather than reading ours. The full grid is on the pricing page; annual billing is twelve months charged as ten.

Frequently asked questions about Polymarket whale tracking

What counts as a whale on Polymarket?

There is no official definition. Our classification is a size-derived label attached to a wallet, and we use it for display rather than for ranking — the ranking is built on measured alpha over at least 25 distinct markets. A wallet can be classified as a whale and have no measurable alpha, and that combination is far more common than the opposite: on 17 August 2026, 3,262 wallets carry the whale label, while 62 wallets of any label clear the alpha filter.

Can I copy whale trades on Polymarket?

Mechanically you can place the same trade. Whether it is worth doing is a separate question, and the arithmetic is unforgiving: you enter after the move the original trade caused, you pay roughly 3.5% of notional to take liquidity, you cannot see the hedge, and you will not see the exit. Copying is a strategy whose costs are known in advance and whose benefits are not. We do not offer trade execution, and nothing we publish is a recommendation to enter a position.

Is a high win rate a good way to pick wallets to follow?

Not on this data. Measured on 17 August 2026, 36,354 of the 145,141 wallets in our table show a win rate of exactly 100% — and 18,914 of those have traded a single market. That is the arithmetic of a handful of trades, not a track record. Any rate you are shown should arrive with the number of distinct markets behind it, and if the denominator is under 25 the rate is not yet a measurement.

How many wallets are actually worth tracking?

By our filter — 25 distinct markets minimum, t above 2.5, plausible volume, measured over 2026-04-01 to 2026-06-24 — 62 wallets out of 16,551 evaluated, as re-measured on 17 August 2026. That is roughly one in 267 of the wallets we looked at closely, and 0.04% of the 145,141 tracked overall. The number will change when the window moves, which is the point of stating the window.

Does whale tracking work on markets that have already resolved?

No, and this is where most alert feeds waste their subscribers' attention. A market whose outcome is settled in practice still trades, at prices near 0 or 1, in large size, because capital is being parked and unparked. Those trades trigger every size-based alert and contain no forecast. Filter the price band first.

Do you merge wallet addresses into entities?

No. There is no address clustering anywhere in our data, and we say so rather than implying otherwise. Every figure we publish about "wallets" is a count of addresses. If a single operator runs several addresses, our data shows several wallets and cannot tell you they are related.

The honest summary of Polymarket whale tracking is that it is a filtering problem where almost everything gets filtered out — 62 wallets from a base of 145,141 — and where the surviving signal is measurable but small, decays with time, and costs 3.5% to act on. That is a less exciting proposition than a whale alert channel, and it has the advantage of being true.

Everything we assert about our own measurements is published with its coverage and its denominator in the public register, including the parts that do not flatter us. The Observer tier is free and requires no card: read the register, audit a wallet, watch the tape, and decide whether the measurement is worth paying for before you pay for it.