●builderYou can reduce context contamination in long-running agents by implementing per-memory decay rather than treating all history as equally valid.
●researcherThe new Temporal Generalization Test (TGT) provides a way to measure memory retention gaps over held-out intervals.