Which exploration policy finds robustness
cliffs fastest under fixed budgets? We ablate –focus
∈ {boundary,runtime,robustness, balanced} atop our
agentic sweep framework [1], measuring (i) time-to-first-cliff,
(ii) area-under-uncertainty, and (iii) policy ranking under fixed
wall-clock and evaluation budgets. Results show that explicitly
boundary-seeking acquisitions dominate low-budget discovery,
while balanced policies win under tight runtime constraints
when end-to-end system effects (ghost cost [2], scheduler
overhead [3], and SLA tails [4]) are included.