ACO Implementation: what's the most efficient of these two (hereunder)

I am currently working with Ant Colony Optimization (ACO) in my model. In this model, I have predefined paths, each of which is further divided into edges. In the Ant class, I aim to ensure that all paths are visited. However, I am concerned that by using predefined paths, the exploratory aspect of the model may be compromised. As an alternative, I am considering randomly visiting all edges (ensuring each edge is visited only once in an iteration) and then constructing paths based on certain constraints. I am uncertain which approach is more appropriate: either allowing ants to choose paths(edges on each path are predefined), potentially sacrificing the exploratory aspect, or randomly selecting edges and subsequently constructing paths.