Who it is forA researcher or forecaster who wants evidence, not P&L, on whether an AI agent forecasts better than the market.
The problemProfit alone does not show skill. You need to know whether the agent's probability is a better forecast than the price the market already showed.
The outcomeA calibration and P&L report from the journal, with the brier_delta metric that compares the agent's forecasts to market prices.
What you need first
- Python 3 (CI runs on Python 3.12); the engine uses the standard library only
- git and a clone of the repository
- A Claude Code account and the `claude` command on your PATH, for running cycles
- A fork with write access to origin, or no origin remote for local-only runs
- Pearl Connect installed from pearl.you/connect and a funded Safe, only for real-money mode
Steps
git clone https://github.com/bennyjo/phil.gitcd philpython3 core/score.py --skip-mtm(inferred, check the README)- Read the calibration section and find brier_delta.
- Open the newest file in
journal/retros/and compare its conclusions with the score.
Prompt for your agent
Clone https://github.com/bennyjo/phil, run `python3 core/score.py --skip-mtm` offline, and report the calibration results and the brier_delta value. Verify that the report states whether the agent's probabilities beat the market price, and say so if it does not.
Caveat. The score is from paper trades that Phil made in its own repository, so the result describes this one experiment and does not show general trading skill.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for bennyjo/phil and "Audit whether Phil beats the market price"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer