Good thread. One note: Polymarket switched from LMSR to a CLOB in 2022 so the softmax formula doesnβt directly price contracts anymore β but modeling implied fair value against the order book mid-price is basically the same idea. Kelly + Bayes stack is solid. What are you using for your probability
Ran the receipts on your thread:
β "Data from 2006β2018" β Polymarket founded June 2020. Impossible.
β Image 3 is a public ML tutorial notebook running on mnist_train_small.csv β handwritten digit classification. Paste that filename into Google. That 0.8282 is recognizing the number "7", not trading Polymarket.
β Equity curve shows dates 2010β2016 on the x-axis. That's a stock/forex chart. Polymarket didn't exist.
β Your XGBoost config: max_depth=1. That's one binary split per tree β the simplest model possible. "IF price < 0.5 β predict YES". 330 of those barely beat random.
β The 2x entry threshold in phase 6 is a real idea (it's Kelly Criterion + EV, not yours)
β XGBoost gradient boosting is real (and worth using)
Everything else is recycled ML education content dressed as alpha.
This post is labeled "paid partnership" at the bottom. Someone paid for this.
@atorres42 We are Ukrainian content creators with 450k fans. We would love to create content for Coca-Cola brands. Are you the right person to speak to about this?