Surrogate-guided: a response-surface model of the predictor is fit and then exploited, balancing exploration of unknown regions against exploitation of promising ones. Strong when the objective surface is smooth and evaluations are cheap.
Matflow Evolutionary Search
A genuine multi-objective genetic algorithm. A population of candidates evolves through selection, crossover and mutation, with Pareto dominance driving selection — the result is a diverse trade-off front rather than a single scalarized optimum. Best with 2+ objectives.
Ranking with TOPSIS
TOPSIS ranks the found front against your objective weights: each candidate is scored by its distance to the ideal solution and the anti-ideal solution, giving an orderable list you can hand straight to the lab.
Constraint handling
Compositional constraints (columns summing to 100%) and per-feature bounds are applied during candidate generation. The optimizer never proposes an infeasible formulation — which is why the top candidates are lab-ready by construction.
Both searches find good fronts, not provably optimal ones — the global Pareto front is not guaranteed, only a good approximation of it.