Extend MissPath-FM to downstream decision tasks and continuous-time irregular settings

Establish the effectiveness of MissPath-FM for downstream decision tasks and continuous-time irregular time-series settings, which remain unresolved extensions of the framework.

Background

MissPath-FM is evaluated primarily for imputation, forecasting, and unconditional generation on discretized time-series data. Although the framework incorporates time gaps and evaluates an irregular missingness mechanism, the paper does not establish its behavior in genuinely continuous-time settings or in downstream tasks that use imputations or generated trajectories for decisions.

The authors explicitly identify both downstream decision tasks and continuous-time irregular settings as unresolved directions, so this is an open extension rather than a research question answered by the paper.