Can You Beat Polymarket? — The Reproducible Study (Code + Methodology)
by Marvoproduction
$19
⚠️ Read this first — this is CODE & METHODOLOGY, not a dataset.
No Polymarket data is included in this download. Instead, a one-command script rebuilds the datasets for you, fresh from Polymarket's free public APIs. That's deliberate: you reproduce the study from source instead of trusting someone's CSV. Requires Python 3.11+ and an internet connection. (No paid data, no API keys, no account needed.)
I spent weeks testing ten different ways to beat Polymarket — an eight-agent AI swarm, the classic betting edges (favorite-longshot, momentum, illiquid markets), calendar arbitrage, a negRisk Dutch book, and copying the on-chain "smart money." Every single one failed once you account for trading costs and confidence intervals.
This bundle is the complete, rigorous code and methodology behind that study — so you can reproduce or falsify every number yourself.
📖 Read the whole series for free at research.marvoproduction-apps.de — this bundle is the code and methodology behind those articles.
What you get
The analysis library — bootstrap confidence intervals, favorite-longshot, calibration, momentum, base rates.
collect_data.py — rebuilds the datasets from the public APIs in one command.
reproduce.py — the market-structure analysis.
negrisk_depth.py — the negRisk Dutch-book order-book-depth test.
smartmoney.py — the on-chain smart-money walk-forward persistence test.
GUIDE.md + METHODOLOGY.md — a step-by-step manual and every experiment, sample size, and confidence interval.
How it works (3 commands)
pip install -r requirements.txt
python code/scripts/collect_data.py
python reproduce.py
The result, in one line: ten independent approaches, one conclusion — Polymarket is efficiently priced. No effect survived a bootstrap 95% CI above zero and the ~2% round-trip cost. What you're buying is the rigorous, reproducible method.
Paper research, for education — not financial advice. No data included; the code fetches public data from Polymarket at run time. Personal research & reproduction license. Python 3.11+.