# AI use cases for Phil (bennyjo/phil)

Repo: [bennyjo/phil](https://github.com/bennyjo/phil) | Use cases: 4 | Page: https://dedupedai.com/use-cases/bennyjo-phil/

Phil is an open-source Python project where an AI agent trades short-term prediction markets on paper, rewrites its own strategy after each settled bet, and can place small capped real-money twins through Pearl Connect.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Audit whether Phil beats the market price: https://dedupedai.com/use-cases/bennyjo-phil/audit-phil-calibration-against-market-price/
- Run your own paper-trading agent from a fork: https://dedupedai.com/use-cases/bennyjo-phil/run-own-phil-paper-cycles-in-a-fork/
- Trace which strategy edits Phil kept or reverted: https://dedupedai.com/use-cases/bennyjo-phil/trace-phil-strategy-lessons-from-git-history/
- Place capped real-money twins of paper bets: https://dedupedai.com/use-cases/bennyjo-phil/run-capped-real-money-twins-via-pearl-connect/

## How to audit whether Phil beats the market price with Phil

Repo: [bennyjo/phil](https://github.com/bennyjo/phil) | Difficulty: Intermediate | Time: 15 minutes | Basis: From the README

Run the offline score report to check if Phil's probabilities beat Polymarket's own prices.

**Who it is for:** A researcher or forecaster who wants evidence, not P&L, on whether an AI agent forecasts better than the market.

**The problem:** Profit 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 outcome:** A calibration and P&L report from the journal, with the brier_delta metric that compares the agent's forecasts to market prices.

### Steps

1. `git clone https://github.com/bennyjo/phil.git`
2. `cd phil`
3. `python3 core/score.py --skip-mtm` (inferred, check the README)
4. Read the calibration section and find brier_delta.
5. Open the newest file in `journal/retros/` and compare its conclusions with the score.

### Prompt to paste into your agent

```text
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.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/bennyjo-phil/audit-phil-calibration-against-market-price/

## How to run your own paper-trading agent from a fork with Phil

Repo: [bennyjo/phil](https://github.com/bennyjo/phil) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README

Fork Phil, run a few paper cycles with Claude Code, and watch the agent edit its own strategy.

**Who it is for:** A developer testing self-improving agents who wants a paper-only copy of the loop without touching real funds.

**The problem:** Reading about a self-editing trading agent does not show how its cycles, commits, and strategy edits behave in practice.

**The outcome:** Your own fork with a few cycles of agent commits to its strategy, journal, and ledger, all on paper.

### Steps

1. Fork bennyjo/phil on GitHub and clone your fork.
2. `cd phil`
3. `python3 core/validate.py`
4. `./loop.sh 10 45`
5. Check the new commits in `strategy/` and `journal/`, then run `python3 core/score.py --skip-mtm`. (inferred, check the README)

### Prompt to paste into your agent

```text
Fork bennyjo/phil, clone the fork, and run `./loop.sh 10 45` for paper cycles. Report how many cycles committed, which files under strategy/ changed, and the output of `python3 core/validate.py`. Verify that no file under core/ or config/protected.json changed.
```

### Caveat

loop.sh pushes to main on origin and commits every cycle publicly, so run a fork without origin (`git remote remove origin`) if you want the run kept local.

Risk, credentials: Each cycle uses your Claude Code account, and loop.sh pushes commits to origin, so keep any tokens outside the repository.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/bennyjo-phil/run-own-phil-paper-cycles-in-a-fork/

## How to trace which strategy edits Phil kept or reverted with Phil

Repo: [bennyjo/phil](https://github.com/bennyjo/phil) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Read the strategy commit history and daily retros to see what the agent learned and reverted.

**Who it is for:** A researcher studying agent self-improvement who needs concrete examples of edits, reverts, and lessons.

**The problem:** Claims about agent learning are hard to check without seeing each change the agent made and why.

**The outcome:** A list of strategy edits with the retro reasoning for keeping, sharpening, or reverting each one.

### Steps

1. `git clone https://github.com/bennyjo/phil.git`
2. `cd phil`
3. `git log --oneline -- strategy/` to list strategy commits. (inferred, check the README)
4. Open `journal/retros/` and find the daily deep-retro entries that audit those edits.
5. Compare each kept or reverted edit with the settled bets it cites.

### Prompt to paste into your agent

```text
Clone https://github.com/bennyjo/phil, list commits that touch strategy/, and match each to a retro in journal/retros/. Output a table of edit, decision (keep, sharpen, or revert), and the evidence cited. Verify that every row cites a retro file that exists.
```

### Caveat

The README says the git log is the record of lessons, but a single edit's effect on results is hard to isolate from the other changes made in the same period.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/bennyjo-phil/trace-phil-strategy-lessons-from-git-history/

## How to place capped real-money twins of paper bets with Phil

Repo: [bennyjo/phil](https://github.com/bennyjo/phil) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README

Set up Pearl Connect and run one real-mode cycle with per-bet caps in place.

**Who it is for:** An experienced operator who has checked the rules where they live and wants a small, capped real-money test of the agent's edge classes.

**The problem:** Paper fills are simulated. Some operators want to see real fill quality against the simulated cross-the-spread model, with strict caps.

**The outcome:** One real-mode cycle that places capped Polymarket orders through Pearl Connect and records real fills in the journal.

### Steps

1. Download Pearl at pearl.you/connect and set up the local signer.
2. `export PEARL_CONNECT_STORE=<path to your Pearl Connect workspace>` (inferred, check the README)
3. `python3 core/real.py doctor`
4. `./loop.sh 1 45 --real`
5. Check `config/protected.json` caps and the journal for the real fill records before you run again.

### Prompt to paste into your agent

```text
In a clone of https://github.com/bennyjo/phil, with PEARL_CONNECT_STORE set to <workspace path>, run `python3 core/real.py doctor` and report whether it says ready. Then run `./loop.sh 1 45 --real` and report the real fills and caps applied. Verify that per-bet stakes do not exceed the cap in config/protected.json.
```

### Caveat

Real bets can lose money, the README says prediction-market trading is restricted or unlawful in some jurisdictions, and the research is not financial or betting advice.

Risk, credentials: Signing keys stay in Pearl Connect's local signer, but the workspace path and any tokens must be kept out of the repository.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/bennyjo-phil/run-capped-real-money-twins-via-pearl-connect/