"The problem becomes efficiently soluble if we take away the per-cell preference..." Ok - so we can make it simpler by saying that these plants are only sown in 'rows', rather than cells. That's an easy fix, but the 'Neighbour Row + 1' still makes it too messy to be 'efficient'? From what I understand then, the best way to go about this would be to grid all plants against each other, and where they intersect, assign a score (-2 to +2) based on their interaction, then choose the highest scoring interactions and work down through the interactions? Not 'elegant', but logical, I guess?