Tag
The author discusses fixing the reflection loop in their trading agent to separate thesis evaluation from PnL to avoid learning from noise, and identifies the need for better context management instead of force-feeding past reasoning.