Data Scrub Skill

Garbage in → backtest lies. Before ANY strategy runs on new data, audit it.

Ask for

• File path
• Expected timeframe (1m, 5m, 1h, 1d)
• Expected timezone (UTC recommended)
• Expected date range
• Symbol(s) covered

Run these checks — each must PASS

1. Schema

• Required columns: timestamp, open, high, low, close, volume
• Dtypes correct? (timestamp = datetime64, OHLCV = float)