Semantic Memory (L2)

Status: Prototype (TRL-3)

The Semantic Memory derives simple rules from the Episodic Memory records and applies them to adjust the champion strategy before publication.

How it works

• Every 200 steps, the system analyzes the collected episodic records and computes average optimal parameters for different market regimes (low/high volatility, low/high DQ).
• Before publishing a new champion genome, it adjusts max_risk_per_trade and phi_llm based on the current market volatility and DQ using the derived rules.

Derived rules example

• High volatility → reduce max_risk_per_trade
• High DQ → reduce phi_llm

Implementation

• src/intelligence/semantic_memory.py – rule derivation and application.
• Integrated into SwarmNode and applied every 50 steps (champion adjustment) with rule updates every 200 steps.