On a rooftop in Brooklyn, a castle investment intern had a few drinks and told me some things he probably shouldn't have.

I casually mentioned that I was thinking about predicting markets. He paused for a moment.
“We have an internal model, specifically for this. Each contract is scored from four dimensions, and if all four match, we enter; if one doesn't, we withdraw. It's that simple.”
I quickly asked what the four dimensions were.
👉 Polymarket official website entrance:
He glanced around, lowered his voice, and quickly said, as if revealing a secret:
“Cross-market divergence, disposal coefficient, capital speed, pairing network correlation.”
I didn’t understand half of it at the time, but I memorized it anyway.
At 11 p.m., back home, I opened Claude.
“These are four scoring factors from a quantitative fund. Can you help me build a terminal that can run these four things in real-time on prediction markets?”
Claude replied simply: “Where do you get the data?”
I handed it an open-source library:
It contains 86 million transaction records. Every wallet’s every deposit and withdrawal, every outcome, all inside.
Three weeks later, I sat in my room, staring at a screen I didn’t fully understand the principles of, watching it spit out profits.
Just the “disposal coefficient” factor alone changed everything. It doesn’t care how good your entry was; it looks at how you exit.
The top-tier wallets can lock in 86% profit, and if losses exceed 12%, they cut the position decisively.
Normal wallets? Only about 58% profit, but they hold losses up to 41%.
Same entry point. Different exit strategies lead to vastly different outcomes.
And then there’s “capital speed,” 49 times. My money has already gone back and forth 49 times while others are just one round.
This terminal uncovered 42 pairs of correlations across 11 markets. For example, the contract “Microsoft Q3 earnings beat expectations,” market price at 80 cents, but the model calculates a 93% probability, so it enters. Two hours later, the price gap narrows, and it exits.
No subjective judgment, no news analysis. The screen just shows four numbers: aligned or not aligned.
His team runs this system with an entire floor of PhDs managing 800 million dollars.
And me:
Claude – $20 a month
VPS – $5 a month
poly_data repository – free
Polymarket API – free
Monthly cost: $25. No team, no office, no Bloomberg terminal.
So far, 280 trades executed, with a 70% win rate. Starting capital: $800.
Each bot keeps its own records:
pulse_alpha +299
arb_hunter +558
trend_rider +337
cal_engine +719
Total profit: $11,514.
Copy trading address:
Last week, he sent me a message.
“Delete everything I told you that day.”
It’s already too late.