bennyjo/phil ? reverse-engineered prompt

Build me a Python agent that acts like a self improving prediction market trader.

I want it to run on short term Polymarket markets, start with paper trading, and keep a journal of what it learned after each settled bet. After every cycle it should review the results, rewrite its own strategy when the evidence suggests it should, and keep a clean record of those changes in git style history. It should focus on markets with the biggest gap between its own estimate and the market price, ignore bad fit markets like simple crypto up or down bets, and include a daily review step that checks whether the strategy changes were actually useful.

Please also add a careful real money mode that only places tiny capped trades through Pearl Connect when the evidence is strong enough, with the right safety limits in place. Make it easy to run on a schedule, include watcher checks for new listings and big price moves, and add a simple score report for calibration and performance. If you need current docs or API details, look them up online first.