How to track a Polymarket wallet's positions

The short way

Paste the address. That is genuinely the whole of it for one look, and it works because of how Polymarket is built rather than because a vendor was generous: the exchange's own repository describes a model in which an operator matches orders off-chain while settlement happens on-chain, non-custodially, swapping conditional-token assets against a collateral token. The executed swaps are public transactions on Polygon. Every product below is reading the same events.

Predicts.guru is the cheapest place to start, because there is nothing to start: no account, no signup, no wallet connection, no paid tier. Paste a Polygon address or search a Polymarket username and you get realised and open profit, ROI, win rate, volume, open and closed positions, the markets traded and the full trade history, with a comparison view for up to ten addresses side by side. Its own home page counters put the corpus at roughly 3.2 million wallets, 1.88 billion trades and 604 million positions.

If you want the same address again next week rather than once, that is a different purchase, and the two things being sold are how often you may look and how often the product re-reads. Those are not the same product at the same price, and picking the wrong one is the usual mistake here. The section below is ordered by that distinction rather than by price.

One thing to be clear about before any of it: what you are reading is a record of what an account did. This page is about getting that record right, and nothing in it is a reason to do the same thing.

What the options are

PolyWallet is the small end and does one job. You paste a public address — its FAQ states plainly that the service is read-only and that no private keys, signatures or wallet connections are involved — and it syncs that address's trades, positions and balances, marks open positions against live Polymarket prices, and lays the result out as daily, weekly and monthly profit and loss with a calendar heatmap. Five tracked wallets are free; Premium at 9.99 USD a month raises that to 500, turns on live unrealised profit and loss, and unlocks CSV export. The mark is the honest part: the vendor says it uses mark-to-market arithmetic that matches what Polymarket itself shows, which makes the unrealised figure a quote of the venue rather than a model of it.

OrcaLayer indexes Polygon itself rather than consuming somebody else's leaderboard, and it is the one that publishes what it throws away. Its methodology page classifies a farmer numerically — an address whose average buy price sits above 95 cents — and a bot at more than 500,000 trades averaging under 20 dollars, states that four Polymarket protocol router contracts are stripped platform-wide, and says win rate is measured over resolved markets rather than fills, NegRisk-corrected, with profit and loss computed first-in, first-out. Twenty wallet lookups a day are free with an account and five without; Pro at 9.99 USD a month lifts the cap and adds Telegram alerts, a 25-wallet watchlist and CSV; Premium at 19.99 USD is the developer tier, with REST at 600 requests a minute, server-sent trade streams and an MCP server. Its terms forbid reselling the data and forbid using it to train or evaluate machine-learning models.

Convexly sells the part that actually costs, which is re-reading one address's full record on a schedule. The first check needs no signup; Watchlist at 9 USD a month covers three wallets re-read every 30 days, Researcher at 49 USD covers ten and adds on-chain reconciliation and an API, and Trader at 499 USD covers fifty re-read weekly. The distinctive artefact is the 19 USD one-time read: a single reconciled read of one named address, delivered with the block it was reconciled at, a hash over that subset and the command that re-derives it against any Polygon archive node, refundable within 14 days. Its methods index carries 27 methodology papers, 15 registered forward tests and 4 published negative results — including one about its own instrument.

Polymarket dashboards on Dune are the option where you stop taking anyone's word for the arithmetic. Reading and forking any public dashboard is free on a view-only account, and every chart can be opened as SQL, so when a number looks wrong you can find out what it was measuring instead of emailing support. The costs are freshness and licence: these are batch-indexed reads minutes to hours behind the chain, the free plan cannot schedule a query at all, scheduled queries expire after three months on Analyst, and the data licence is internal use only below Enterprise.

Where this breaks

A profit figure is a calculation with an exclusion rule behind it, and most of these do not publish theirs. The clearest evidence is an accident: four products in this category print a corpus counter on the landing page, and the wallet counts agree while the market counts do not. Polysights says 160,722 markets, OrcaLayer roughly 1.6 million, Airavat over 2.2 million — a factor of thirteen over one venue's history, and none of them publishes whether it is counting a question, a binary outcome pair or one leg of a multi-outcome set. Anything computed per market, a win rate first among them, is computed over a denominator these products do not define the same way. Who publishes the rule behind the label grades all five on exactly this.

Unrealised and realised are two different claims, and an open position is marked at somebody's price. A realised figure needs lot matching, and FIFO is a choice rather than the only one. An unrealised figure is the venue's current price applied to a position nobody has closed — fine on a liquid market, a quote rather than a valuation on a thin one. And splitting, merging or redeeming conditional tokens moves value in and out of a position with no trade to attribute it to, so the question to put to any product is whether its pipeline watches fills, balances, or both. A number without that answer is not comparable to another number without it.

The history you are reading may be a window rather than a record. This is the failure that looks exactly like success, and the only public measurement of it comes from a vendor measuring itself: Convexly's study of 8,656 Polymarket wallets reports that 3,268 of them — 37.8 percent — sat at exactly 2,000 fills, the cap on the trade history it had pulled, and that capped wallets lost a median of 46.9 percent of their resolution coverage against 20.8 percent for uncapped ones. It withdrew an earlier top-100 audit on the same ground, because 85 of its 100 published rows sat at the cap and therefore described a window. Convexly marks each row as a full read or a recent-trade window; nothing else here prints a completeness ratio, so on every other product the honest reading is "incomplete by an unstated amount".

One person is not one address, and one address is not one person. Polymarket separates the profile address that holds funds from the key that signs, splitting size across addresses costs nothing, and a fund, a shared account or a bot acting for subscribers all print as a single row. The population is not clean either: OrcaLayer's April 2026 audit of 470 million trades found that six of ten wallets with apparently flawless records were airdrop farmers, accounts buying near-certain outcomes to harvest a cent. What wallet tracking shows, and what it cannot is the page for where that boundary sits — and it is a boundary about people, which is why the sentence the record supports stops at "this account established this position at this time".

A deposit is not a profit. Collateral arriving at an address is a transfer; so is collateral leaving it. Either can move a balance without anything having been won or lost, which is why a balance-shaped number and a trading-result-shaped number have to be read as different fields even when a screen puts them in the same column.

And this page exists for Polymarket because the same task is not available elsewhere. A CFTC-regulated exchange publishes market statistics; the per-account layer goes to the regulator, under a rule that requires each futures commission merchant, clearing member and foreign broker to submit a report to the Commission for each business day. So there is no public per-account ledger on Kalshi for anyone to index, and no product can build one. What you can read there is your own account, through your own credentials.

If you outgrow this

If the question changes from "what did this address do" to "is this record distinguishable from luck", you have left this page: that is a comparison of methods rather than a lookup, and the five wallet analytics products compared is where the published exclusion rules, the error bars and the absences sit side by side.

If you need the record programmatically rather than on a screen, two of the five publish an interface: OrcaLayer's REST, server-sent streams and MCP server on its 19.99 USD tier, and Convexly's API from its 49 USD tier under a daily row quota. Predicts.guru and PolyWallet publish none, so their ceiling is what you can read off a page or, on PolyWallet's paid tier, export as CSV. Check the licence in the same sitting: OrcaLayer forbids resale and model training, Convexly's self-serve licence covers one named individual and forbids republishing its output to third parties, and Dune is internal use only below Enterprise. A research workflow that quietly becomes a product is a licence problem before it is a data problem.

If a position's value is what you are really chasing, the open half of a wallet page is only as good as the resolution behind it — who decides the outcome covers the process that turns an open position into a settled one, and it is where the surprises live. And if the address you are reading trades somewhere else as well, nothing above will see it: tools that cover more than one venue is the listing, and analytics and dashboards is the full set of products that read this record for you.