Query-conditioned reuse beats replaying whole trajectories by 10.7 points at 48.9% fewer tokens.

Holding retrieval, target state, model, decoding and tool budget fixed, researchers compared how a retrieved memory gets used. A target-bound note recording a reusable procedure, bindings to recover, applicability conditions and verification requirements hit 62.3% average success across WebArena, WorkArena and AppWorld. Injecting the full trajectory did worse and degraded sharply as traces grew. arXiv 2608.12847 Store the lesson, not the transcript.