Balanced budget between LIGHT and HEAVY.
upper bound on optimization trials for this mode.
number of instruction/demo candidates generated per module.
number of validation examples used to score candidates.