Necessity of proposal and history inputs in RTX trajectory reproduction

Determine whether proposal and history inputs are necessary for reproducing the RTX reserved live trajectories by evaluating a target-only UB2 competitor on the same reserved prompts.

Background

The RTX replay study reproduced the native live trajectories using an independently orchestrated draft-only proposal process and a target-only replay process, but the replay still relied on sealed proposal and history inputs. Because no target-only UB2 competitor was tested on the same reserved prompt set, the observed agreement does not establish that those inputs are necessary. A same-set target-only UB2 comparison is therefore required to resolve whether proposal generation and prior-history information contribute essentially to the reproduction.