Dismantled the Polymarket leaderboard 40 addresses, only three ways to make money

The differences between the three types are not just parameters; they are playing completely different games.
Written by: Leo

How do the strategies of people who made $10 million on Polymarket look? Using Data API + on-chain data, I reverse-engineered the top 20 rankings in both sports and crypto tracks.
40 addresses, over 100,000 trades, analyzed one by one. Not just looking at dashboard screenshots. Every buy, sell, and redemption is restored to its strategic behavior.

Method: Use Polymarket Data API to pull transaction records per address, LB API to verify profit and loss, on-chain REDEEM/MERGE data to restore real cash flow.
Each address has between 2,000 and 15,000 trades. After breaking down the data, I found that regardless of sports or crypto, profitable addresses fall into three categories.

The differences between these three are not just parameters; they are playing entirely different games.

Type 1: Directional
Buy the right side and hold until the sports event settles. The most profitable strategy—so simple I initially didn’t believe it.
Out of 18 effective addresses, 14 only buy and never sell. Hold until settlement, redeem if they win, zero out if they lose, no swing trading.

The same “buy only” approach but with a completely different profit method.
swisstony: $494 million in trading volume, 1% return rate, net profit of $4.96 million. Fully automated, 353 trades in 30 minutes, covering five major leagues.
Earning a little per game but volume is huge.
majorexploiter: 39% return rate, largest single trade $990,000.
Over 600 trades mostly on two Arsenal matches. Willing to bet big—winning means millions.
One volume, one bet—both made millions.

The methods differ, but both share a common advantage: they have informational edges on the events they bet on.

Top of the leaderboard: kch123
Sports leaderboard top, total profit $10.35 million.
As of mid-March analysis, lost $479,000 in the last 30 days.
Win rate in the past 7 days only 31% (15 wins, 33 losses).
All 14,303 trades are buys, zero sells.
Average 493 trades per day, 74% of trades occur within 10 seconds.
This machine that made $10 million is slowing down.

Just looking at the leaderboard won’t reveal these details; you need to break down on-chain data.

My own label misled me: fengdubiying, 13th in sports, profit $3.13 million.
When I analyzed in bulk, I labeled it as “sell-dominant,” thinking it was swing trading.
Breaking down the data: 93.6% of the returns come from redemptions, only 6% from sales.
The real strategy is concentrated betting on LoL esports.
Max single market: $1.58 million (T1 vs KT Rolster), win rate 74.4%, profit/loss ratio 7.5 to 1.
Sales are just stop-loss tools, not the main strategy.
Just looking at buy/sell ratios on the dashboard can lead to complete misjudgment of what this address is doing.

Type 2: Structural
Not relying on predictions to make money—crypto leaderboard is a completely different species.
In sports, it’s about betting on direction; in crypto, it’s market-making.

Deep dive into the top 5 crypto market makers:

• Three are market-making bots trading binary options on ups and downs.
• One uses MERGE to manage inventory with a price threshold.
• One arbitrages milestone events in public offerings (return rate 43.3%).

Retail traders bet on ups and downs; top players act as market makers.

How do market makers profit?
0x8dxd: BTC market maker on 5/15-minute candles.
94% of trades are symmetrical orders, both buy and sell.
Operates all day, median trade size under $6.
Buy price increase + decrease < $1; the spread is profit.
At least three independent addresses run this pattern.

Another market maker address is even more extreme:
Almost monopolizing liquidity in the Economics category.
982 buy orders, zero sell orders, six-figure PnL.
Profits come from maker rebates plus liquidity premiums.

Good code doesn’t necessarily mean profit.
You might think market-making is guaranteed profit?
There’s an open-source Polymarket market-making bot on GitHub, well-engineered with WebSocket real-time data, risk controls (stop-loss + volatility freeze + sleep periods), and automatic position merging.
The author admits: it’s not profitable.
The reason: its pricing logic is penny jumping—placing bids just one cent ahead of the best quote.
Basically, it’s copying others, with no independent pricing ability.
Even the best code is useless if your pricing model can’t beat the market.

Another noteworthy data point:
Based on on-chain transaction timestamps, over 70% of arbitrage profits in Polymarket crypto markets are captured by bots with latency under 100 milliseconds.
Less than 8% of wallets are profitable overall.
If bot latency is in seconds, it’s essentially providing liquidity for high-frequency traders.

Type 3: Cognitive
Betting less frequently but with each decision based on judgment.

This third type of address is completely different from the first two.
Low trading frequency—maybe only two or three trades a month—but each trade involves research.

Examples:

• A weather-related address models using public meteorological data, only enters when win probability exceeds 0.77, maybe two or three trades a month, each profit several thousand dollars.
• Another address: 89% of trades are “NO” bets, holding positions for months, with a win rate not high but an average payoff over 9 times, covering small losses with a few big bets.
• An even more extreme case: in FDV (full outcome) markets, only doing one thing—buy “NO” at 50-55 cents, settle at $1.
  Win rate 100%.
  Not luck—others haven’t noticed this pricing bias.

But cognitive strategies are not just about “deep research equals profit.”
I analyzed a case where someone used 1.37 million lines of historical data to create a probability matrix for BTC price deviations.
Backtesting looked perfect, but when rolling forward, it immediately failed.
Market efficiency improves rapidly—what was profitable last month is arbitraged away this month.
The true edge in cognitive strategies is understanding a category more deeply than the market’s pricing, not just having a more complex model.

Comparison of the three approaches
I’m doing several things myself:

• Crypto market-making (structural)
• Sports probability pricing (directional)
• Weather data modeling (cognitive)

Each is small-scale—no daily average of 493 trades like kch123, nor $494 million in volume like swisstony.

After analyzing these 40 addresses, the most important realization:
Figure out which game you are playing—this is more important than optimizing any parameter.

If you pursue a directional approach without an informational edge, even perfect execution is guesswork.
If you rely on structural strategies but can’t keep up with latency, you get exploited.

This is not just motivational talk; it’s what I tell myself after reviewing the data.

Right now, I’m validating each approach on a small scale, confirming the edge before scaling up.
No rush to expand—first, get one or two categories running smoothly.

Data sources:
Polymarket Data API + LB API + Polygon on-chain data
Analysis period: January-March 2026

Thinking of trying Polymarket?
First, clarify which game you want to play.