When comparing against a known latent space with known topology, focus the improvement in inference assessments on just the Markov boundary (or Markov blanket?) of the latent space from the true network and make the f1, pr-auc, AUC, and other graphs on just that subset of nodes.
When comparing against a known latent space with known topology, focus the improvement in inference assessments on just the Markov boundary (or Markov blanket?) of the latent space from the true network and make the f1, pr-auc, AUC, and other graphs on just that subset of nodes.