OrcaLayer, this week's featured listing on PM Wisdom, tracks Polymarket wallets, P&L, and smart-money consensus, but it starts from a distinction most leaderboards ignore: making money is not the same as being right. Its index covers more than 1.5 billion on-chain fills across 3.1 million wallets, corrects for NegRisk mechanics, filters airdrop farmers and bots, and powers tools for traders, researchers, and developers. The team also maps Polymarket's Ukraine territorial markets against the ISW frontline. We put our questions to OrcaLayer about what public data misses, how a wallet earns the Smart label, and why its sharpest product lesson came from a reader proving one of its own metrics wrong.
Disclosure: OrcaLayer is the current featured listing on PM Wisdom. This interview is editorial content. It is not financial advice, and a High trust rating is an editorial confidence label, not a safety guarantee.
What problem does OrcaLayer solve?
OrcaLayer: The public leaderboard tells you who made money. It does not tell you who was right. Those are different questions, and on Polymarket the gap between them is wide.
A wallet can post a 97 percent win rate by repeatedly buying outcomes that already trade above 95 cents. It collects a few cents of spread on events that are effectively decided. That is a real business and there is nothing wrong with it, but it is not forecasting, and if you follow that wallet expecting insight you will be disappointed.
We index the chain and separate the two. Among the wallets we track with a win rate of 90 percent or better, 64 percent turn out to be collecting that spread rather than predicting anything. As of the end of August 2026, we count 139,355 farmers, roughly one in eleven of the wallets with enough resolved markets to rank at all.
So the problem is easy to state and tedious to solve: a number on a leaderboard is not evidence until you know how it was earned.
Why did you build it?
OrcaLayer started as infrastructure for myself. I was trading on Polymarket and building agents around it, and everything I wanted to do needed the same thing first: a clean database I could actually query. Who holds what. Who has been right, and over how many markets. Which wallets are worth watching and which are noise. The public data would not give me that in a usable form, so I built the index.
It stayed private for a long time. I used it to test my own ideas, to find wallets, to run the processes I was interested in. The site came later, and only because the hard part was already finished. The API we publish is the one I work on myself, not a reduced version of it.
That order matters for what this is. It was built to be relied on with my own money before it was built to be shown to anyone.
What makes your data different?
Three things, and none of them are clever.
First, we compute from Polygon rather than reading a number off a page. More than 1.5 billion on-chain fills across 3.1 million wallets. Worth saying plainly: one order can produce many fills, so that is not a count of trades.
Second, we correct for how Polymarket actually works. NegRisk markets let positions be split, merged, and converted, so a naive profit calculation misreads them. Fills are matched FIFO. Hedged positions get flagged instead of being counted as directional wins, because a wallet holding both sides of the same market is running a structure, not a view.
Third, every threshold is published. Win rate, P&L, the farmer filter, all the formulas, at orcalayer.com/methodology. If you disagree with a definition, you can see exactly which one to argue with.
That last point matters more than it sounds. Most of this category asks you to trust a score. We would rather hand you the arithmetic.
How do you identify smart traders?
Three conditions have to hold at once.
A win rate of 55 percent or higher, measured per market rather than per trade. Per trade flatters anyone who scalps. Per market asks whether the call was right.
Positive total P&L across resolved markets, so being frequently right on tiny positions does not qualify.
And no farmer or bot classification. The main farming signal is average entry price: a wallet that keeps buying above 95 cents is harvesting spread, not predicting. Trade count and average trade size sharpen the read.
What this does not do is predict. It is a filter, not a forecast. It makes a win rate harder to inflate, which is a smaller claim than "these wallets will be right next time" and a more honest one. Past performance on a leaderboard stays exactly what it is, and we say so on the site.
Who is OrcaLayer built for?
Three groups, and they use it differently.
Traders who are about to follow someone. The usual sequence is that a wallet shows up in a screenshot with an impressive number attached, and the question is whether that number means anything. That check takes about a minute with us.
People writing about prediction markets. Journalists and researchers keep needing the same thing: who actually holds this position, how big, and since when. That is a query, not a research project.
Developers. Everything the site shows is available through the API, so if you want to build your own dashboard or alert bot, you are not scraping anything.
The common thread is that all three arrive with a specific wallet or a specific market in mind. We are not a place to browse. We are a place to check.
What can developers build with your API?
The API returns the same index the site runs on, so anything the site does you can rebuild.
Two things in that index do not exist in the exchange's public API. First, both
sides of every fill. Polymarket's public activity feed names only the taker, so maker
attribution for an arbitrary wallet is impossible there by construction. Since
large players tend to work with limit orders, they sit on exactly the side the
public feed cannot see. Second, mint and merge as first-class events. An
entry_type field tells you that a MINT with a SELL on YES means the wallet
entered NO, not that it exited. A naive parser reads that backwards.
There are three official ways in: a REST API, a Python SDK, and an MCP server for AI
agents, plus an agent skill for people who want the same capabilities with no
server to run. The MCP server exposes six tools, including smart-money consensus on
a single market and the whale alert feed. It installs with uvx orcalayer-mcp, and
it is in the official MCP registry. The client-side projects are public under MIT
on GitHub.
The leaderboard and wallet endpoints are open. Alerts and the live stream need a paid plan. The rate limit is 600 requests a minute on Premium.
What have users taught you so far?
The sharpest lesson came from someone tearing apart one of our own posts.
We had published a market observation and used smart-money consensus as the argument: this many tracked wallets on one side, that many on the other, therefore the price looks wrong. A reader pointed out that a head count weights a five-dollar wallet the same as a five-hundred-thousand-dollar one, and that on a cheap longshot the small tickets dominate by construction. He was right. The metric was descriptive, and we had used it as proof.
Two things changed. Head count is no longer presented as evidence that a price is mistaken. And the market consensus endpoint returns a capital-weighted split next to the head count, because the two disagree more often than you would expect. On one market last week, the head count said 60/40 for NO while the capital said 63/37 for YES.
The warning is written into the tool description itself now, so anyone calling it reads it before they read the numbers. That felt like the right place to put a lesson we had to be taught in public.
What is next for OrcaLayer?
The honest answer is narrower than a roadmap.
The gap we care about most is between what a price says and what actually settled. NegRisk mechanics let a position be converted rather than sold, which means a wallet's real exit is not always visible in the order book. Reading that correctly needs the CLOB and the chain reconciled, and that is the piece we are working on. It is also the piece nobody has fully solved, us included.
The capital-weighted view is done on markets and in the MCP server. It is not in the wallet profile yet, and it should be, so that is next.
Beyond that, wider coverage of the territorial and event markets where the underlying facts move faster than the price does.
We are deliberately not adding a signal product. The moment we start telling people what to trade, the filter stops being a filter and becomes another opinion.
What is the Ukraine territorial tracker, and why build it?
It is the piece of OrcaLayer with no equivalent that we know of.
Polymarket runs markets on whether specific Ukrainian cities change hands. Pricing those well means knowing where the front line actually is, and that information does not live in the order book. So we overlay the daily ISW frontline map onto those markets: 35 cities and 41 market subscriptions at the moment, each with a distance to the front and a threat level from SAFE through CRITICAL.
It started as an internal tool. Reading a territorial market without a map is guessing, and we were tired of switching between two tabs.
It turned out to be the most cited part of the whole product. Price alone does not tell you a settlement changed hands. The map does.
Published by PM Wisdom.
Read the full listing, trust rating and editorial review at pmwisdom.com/projects/orcalayer.