PADP v1.0 — 13-step agentic forecasting protocol
Powered by Claude Sonnet 4.5 via Claude Code
Can an AI agent be consistently profitable on Polymarket prediction markets? This dashboard tracks a live experiment to find out.
PADP (Polymarket Agentic Decision Protocol) is a structured 13-step methodology for systematic forecasting. The agent researches markets, estimates probabilities, stress-tests its reasoning, and calculates optimal position sizes.
Each analysis takes roughly 30 minutes of autonomous research. The protocol includes Fermi decomposition, Bayesian updating, pre-mortem analysis, and Kelly criterion position sizing.
The agent generates a detailed thesis letter for each trade, explaining its reasoning and confidence level.
Claude Sonnet 4.5 via Claude Code, $1,000 to play, and a simple question:
Can systematic AI forecasting beat the market?
The agent writes the thesis, the trade gets placed on Polymarket, and we wait—sometimes days, sometimes weeks, sometimes months—for the market to resolve.
If it works, that's a new capability. If it doesn't, the next question is how long until it can. Maybe this could be one more benchmark to help define AGI.
With enough budget to run multiple LLMs and give each one a bankroll, we could also explore whether there are statistically significant differences in performance across models and providers.