Asynchronous draft run-ahead for hiding proposal latency

Investigate whether allowing the action draft to run ahead of the target can hide part of the proposing cost in speculative action rounds, thereby improving the latency of synchronous propose–verify–commit systems.

Background

The latency framework assumes that proposing and verifying occur sequentially, so the draft’s proposing cost is paid in full before target verification begins. The paper identifies asynchronous draft run-ahead as an unmodeled execution strategy that could overlap drafting with target computation and reduce the effective proposal cost. The authors explicitly defer this issue to future work.

References

Proposing and verifying do not overlap, as in Sec. 3.1; a draft that ran ahead of the target could hide part of the proposing cost D, which we leave to future work.

— LEAP: Learning Efficient Action Proposals For LLM Agents  (2610.02670 - Xu et al., 2 Oct 2026) in Appendix C.1, “Assumptions”