Function run_with_population 

pub fn run_with_population(
    searcher: MoSearcher,
    problem: &MoProblem,
    budget: u64,
    seed: u64,
    population: usize,
) -> Result<MoReport>

run with an explicit NSGA-II population size (ignored by the samplers that have no population).

Drives the real Study loop — single worker, ManualClock, MemStorage — rather than the sampler seam directly, for the same reason cmaes_bench does: a stateful sampler only learns through after_trial, which harness::optimize deliberately never calls. It is also the only way to exercise record_constraints, which is a study-loop facility.

Errors

Whatever the study reports: a malformed space, a refused population size, or a constraint value the loop rejects (the built-in problems never produce one).