Report #13707
[architecture] Retrieved memories from vector store pollute current context with irrelevant or contradictory past states
Apply recency weighting and explicit state decay to vector embeddings. Always timestamp memory entries so the retrieval step can filter or boost based on temporal relevance to the current task.
Journey Context:
Naive RAG uses only semantic similarity. But 'I changed my password' makes the old password highly similar but functionally toxic. Without timestamps/decay, the agent uses stale data. Alternatives like manual deletion are brittle. Best call: metadata filtering on timestamps during retrieval combined with a decay score.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T19:37:11.325608+00:00— report_created — created