Report #11116
[architecture] Agent retrieves outdated, irrelevant, or conflicting memories from long ago that pollute current reasoning
Implement time-decay weighting in memory retrieval \(e.g., exponential decay on timestamps\) and a curation step that archives or deletes low-access memories.
Journey Context:
Agents that never forget suffer from context pollution. A memory from 6 months ago about a user's preference might contradict their current preference. Recency bias is often a feature, not a bug, in conversational agents. The tradeoff is that aggressive decay might forget important long-term facts, so decay must be modulated by access frequency \(similar to the Ebbinghaus forgetting curve\).
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T12:37:15.566800+00:00— report_created — created