# Feed Hunter

Automated X/Twitter feed intelligence pipeline. Scrapes → triages → investigates → simulates.

## Architecture

```
Scrape (CDP) → Triage (claims/links) → Investigate (agent) → Alert (Telegram)
                                                                    ↓
                                                             Spawn Project
                                                                    ↓
                                                          Simulate / Backtest
```

## Pipeline Stages

1. **Scrape** — Extract structured posts from X feed via Chrome CDP
2. **Triage** — Identify verifiable claims with actionable links
3. **Investigate** — Agent follows links, verifies claims with real data
4. **Alert** — Telegram notification with findings + inline action buttons
5. **Simulate** — Paper trade the strategy, track P&L without real money
6. **Backtest** — Where historical data exists, test against past performance

## Simulation System

Every viable strategy gets a simulated portfolio entry:
- Virtual bankroll (configurable, default $1000)
- Paper positions tracked in `data/simulations/`
- Daily P&L snapshots
- Performance metrics: win rate, ROI, Sharpe ratio, max drawdown

## Project Spawning

When a strategy passes simulation thresholds:
- Auto-scaffold in `projects/<strategy-name>/`
- Working bot code
- Risk parameters
- Go/no-go recommendation

## Schedule

- Feed scrape: Every 2-4 hours during market hours (8am-10pm CST)
- Investigation: Triggered by triage hits
- Simulation updates: Hourly for active positions
- Daily digest: 9am CST summary of all active simulations

## Files

- `config.json` — pipeline settings, thresholds, bankroll
- `data/simulations/` — active paper positions
- `data/backtests/` — historical backtest results
- `data/investigations/` — investigation logs per post
- `data/alerts/` — alert history