Memory Worth: Agent Memory Governance

What Changed

Introduces MW: two counters tracking success/failure co-occurrences per memory.

Why It Matters

Provides a principled, feedback-driven way to manage agent memories amid task shifts, potentially improving long-term reliability. Low-cost integration could become standard in LLM agent architectures.

What To Do Next

Implement MW counters in your agent's retrieval logger and suppress memories below MW=0.2 threshold.