ONLINE
Β·scanning 14,000 wallets
Β·tracking 47 smart wallets
Live Feed
Performance
βbalance
βprofit
βtrades
βROI
βwallets scanned
βsmart wallets
>2.0sharpe ratio filter
ΒΌ Kellyposition sizing
How It Works
$ claude scans 14,000 wallets
$ filters 47 that never lose (Sharpe > 2.0)
$ copies their trades in real-time
$ buys at $0.48 β sells at $0.52 β pockets $0.04
---
who wins doesn't matter. the spread pays.
Start Here
The Guideβ 3 formulas, 12 tools, the 20-line Claude brain.
Toolkitβ Every open-source weapon. Free. Tested.
Securityβ 664 malicious repos on GitHub right now. Read this.
The Numbers
Polymarket hit $7.94B volume in February 2026. Weekly volume broke $2.1B in March β new ATH.
87% of wallets are in the red. The top 20 wallets captured more profit than the bottom 13,000 combined.
This isn't a casino. It's a math exam. Most people showed up without a calculator.
Formula 1 β Expected Value
Market says 40%. You believe 60%. Edge per dollar: $0.20
Rule: EV < 5% β SKIP. No exceptions. This eliminates 90% of losing trades.
Formula 2 β Kelly Criterion
Full Kelly says bet 33%. Never do this. Use Quarter Kelly. Always.
$1,000 bankroll β bet $83. Not exciting. Won't blow you up either.
Formula 3 β Bayesian Updating
Prior on Fed rate cut: 55%. After inflation data: 88%
Certainty is a bug, not a feature.
The 20-Line Claude Brain
import anthropic, json
def claude_probability(market_question, market_price):
client = anthropic.Anthropic(api_key="sk-ant-...")
response = client.messages.create(
model="claude-sonnet-4-20250514",
messages=[{"role": "user", "content": f"""
You are a calibrated prediction market analyst.
Market: {market_question}
Current price: {market_price}
Estimate TRUE probability (0.00-1.00).
Return JSON: {probability, confidence}
"""}]
)
return json.loads(response.content[0].text)
50 lines of custom code. Everything else is open source.
5 Mental Bugs That Cost More Than Bad Code
01Base Rate Neglect β 99% accurate test on 0.1% event = 9% true positive
02Sunk Cost Fallacy β bought at 70Β’, dropped to 40Β’. Would you buy at 40Β’ right now?
03Survivorship Bias β 87% of wallets are red. You never see their screenshots
04Copying Without Filtering β 91% WR on crypto, 15% on politics. Copy only dominance
05Overfitting β "Every time X happens..." Based on 3 examples. That's noise
Where to Start Tonight
"I want data first"
poly_data + polyterm β feed to Claude β find 47 wallets with Sharpe > 2.0 in 4 minutes
"I want to copy smart wallets"
polyterm --type smart + copy-trading-bot β filter by category β Quarter Kelly sizing
"I want to build a bot"
py-clob-client + Claude API + 20-line brain β paper trade 1 week β 200+ trades before live
"I want market making"
poly-maker β both sides β collect spread $0.02-0.05 per fill β thousands per day
Layer 1 β Data
Layer 2 β Intelligence
Layer 3 β Execution
Layer 4 β Infrastructure
Polysights / Insider Finderβ 24,000 users. $2M funding. $25K Polymarket grant.
$ polyterm wallets --type whales
$ polyterm wallets --type smart # >70% WR
$ polyterm alerts --type insider
$ polyterm alerts --type arbitrage # vs Kalshi
Top Performers
beachboy4 β
$6.12M profit in a single day sports Β· was deep in red before
majorexploiter β +$2,416,975 in March 2026 geopolitics Β· elections only Β· laser focus
CemeterySun β $36.6M volume market making Β· tiny edge Γ thousands of trades
What They Share
Not insider info. Not luck.
In traditional finance, careers are built on 2% edges. On Polymarket, 20% edges exist daily β if you can find them.
The Incident
December 2025. A GitHub repo called polymarket-copy-trading-bot contained malware. Professional README. Working code. Real API connections.
Hidden inside a dependency: code that read your .env, extracted your private key, and sent it to a remote server.
The bot worked. Your money disappeared. 14,285 people downloaded malware before anyone noticed.
Rules
β NEVER use your main wallet. Dedicated wallet, minimal funds.
β Audit every dependency. pip list. Google suspicious packages.
β Repo created after Feb 2026 with 500+ stars? Likely star-farmed.
β Use
Revoke.cash to limit USDC approvals. Never unlimited.
β Start with $100-300. If it works for 2 weeks, scale gradually.