How are you handling memory provenance in persistent agents — verified vs. inferred facts?

In three linesDeveloper highlights the challenge of distinguishing verified facts from inferences in persistent agent memory. Old inferences get promoted to facts over sessions, breaking auditability. He manually implements provenance tagging (verified/inferred/speculative) and asks whether existing solutions (Zep, Mem0, Cognee) address this epistemic layer problem.

Summary generated by Claude — human-verified