Polymarket Millionaire Review: 40 addresses, 100,000 transactions, only three ways to make money

Core Viewpoint

First Type: Directional, Waiting for the Right Outcome

The most profitable strategy in the sports sector is so simple that I initially didn't believe it.

Among 18 effective addresses, 14 only buy and do not sell. They hold until settlement, redeem when winning, and go to zero when losing, without making swings.

Even though they only buy and do not sell, the ways of making money are completely different.

swisstony: $494 million in trading volume, 1% return rate, net profit of $4.96 million. Fully automated, placing 353 orders in 30 minutes, covering the five major leagues. Each game earns just a little, but the volume is massive.
majorexploiter: 39% return rate, with the largest single bet of $990,000. Over 600 transactions almost all bet on two Arsenal matches. Daring to place heavy bets, winning means several million.

One focuses on volume, the other on betting, both made millions. The methods are opposite, but they share a common point: they have an information advantage on the events they bet on.

The Top Ranked is Losing Momentum

kch123, ranked first in sports, has accumulated a profit of $10.35 million.

However, as of mid-March analysis, in the last 30 days, they lost $479,000. In the last 7 days, the win rate was only 31% (15 wins, 33 losses). All 14,303 transactions were buys, with 0 sells. An average of 493 transactions per day, with 74% of transactions occurring in less than 10 seconds.

The machine that made ten million is losing momentum. You wouldn't know this just by looking at the rankings; you need to dismantle the on-chain data to see it.

I Deceived Myself with My Own Label

fengdubiying, ranked 13th in sports, with a profit of $3.13 million.

When I did batch analysis, I labeled them as "sell-dominant," making it seem like they were making swings.

Dismantling the data: 93.6% of the cash flow comes from redemptions, with sells only accounting for 6%. The real strategy is concentrated betting on LoL esports. The largest single market bet was $1.58 million (T1 vs KT Rolster), with a win rate of 74.4% and a profit-loss ratio of 7.5 to 1.

Selling is their stop-loss tool, not the main strategy. Just looking at the buy-sell ratio on the dashboard, you would completely misjudge what this person is doing.

Second Type: Structural, Making Money Without Predictions

The crypto rankings are a completely different species. In sports, it's about betting direction; in crypto, it's about market making.

Digging deep into the Crypto Top 5: three are market-making bots running binary options on price fluctuations, one is a price threshold market maker managing inventory with MERGE, and one specializes in public milestone event arbitrage (return rate of 43.3%).

Retail investors are betting on price movements, while top players are the market makers.

How Market Makers Make Money

0x8dxd, a BTC 5/15 minute price fluctuation market maker.

94% of transactions are symmetric orders, buying both up and down. Operating all day, with a median single transaction of less than $6. The buying price rises + falls < $1, and the difference in between is profit. At least three independent addresses are running the same model.

Another market-making address is even more extreme: it almost monopolizes liquidity supply in the Economics category. 982 buys, 0 sells, six-figure PnL. They earn from maker rebates plus liquidity premiums.

Good Code Doesn't Equal Making Money

At this point, you might think market making is risk-free? There’s an open-source Polymarket market-making bot on GitHub, with well-engineered code, real-time WebSocket data, a three-piece risk control system (stop-loss + volatility freeze + sleep period), and automatic position merging. The author admits: it’s not profitable.

The reason is that the pricing logic is penny jumping, inserting a penny in front of the existing optimal quote. In plain terms, it’s just following orders, lacking independent pricing capability.

No matter how refined the code is, it’s useless; whether market making is profitable depends on whether your pricing model can be more accurate than the market.

Another noteworthy data point: based on on-chain transaction timestamps, over 70% of arbitrage profits in the Polymarket crypto price market are taken by bots with delays of less than 100 milliseconds. Less than 8% of wallets in the entire market are profitable. If a bot has a delay in seconds, it’s basically providing liquidity to high-frequency players.

Third Type: Cognitive, Betting Less but Judging Every Bet

The third type of address is completely different from the first two. The trading frequency is very low, possibly only making two or three trades a month, but every trade is backed by research.

Here are a few examples.

An address in the weather category uses publicly available data from the meteorological bureau to model, only entering when the win rate exceeds 0.77, possibly making only two or three trades a month, with single profits in the tens of thousands of dollars.
Another address has 89% of transactions as buying NO, with holding periods measured in months, win rates not high, but average payoffs over 9 times, covering all small losses with a few big bets.
There’s an even more extreme case: in the FDV (full result) market, they only do one thing, buying NO at 50-55 cents, waiting until settlement to get $1. Win rate 100%. It’s not luck; it’s that others haven’t noticed this pricing deviation.

But cognitive types don’t mean "the deeper the research, the more you earn." I dismantled a case where someone used 1.37 million lines of historical data to create a probability matrix for BTC price deviations, and the backtest performance was perfect, but when rolling validation was done, it collapsed immediately. Market efficiency improves quickly; patterns that were useful last month have already been arbitraged away this month.

The true edge of cognitive types is that your understanding of a certain category is deeper than the market pricing, not that the model is more complex.

Comparison of the Three Types of Strategies

Comparison Table of the Three Types of Strategies

What Am I Doing

Having talked about others, let’s talk about myself.

I’m running several lines at the same time: crypto market making (structural), sports probability pricing (directional), and weather data modeling (cognitive). Each line is not large, not having the scale of 493 transactions per day like kch123, nor the trading volume of $494 million like swisstony.

After dismantling these 40 addresses, the thing I thought about the most was: figuring out which game you are playing is more important than optimizing any parameters.

If you play directionally but lack an information advantage, no matter how good the execution is, it’s still guessing. If you play structurally but can’t keep up with delays, you are the one being harvested. This isn’t motivational talk; it’s what I told myself after looking at the data.

Now, each line is running small-scale validations, confirming the edge exists before scaling up. I’m not rushing to expand; first, I’ll get one or two categories running smoothly.