Also due to population selection method, on small datasets, package generate single-item iterations

Right, single-item iteration will happen in two situations:

• for very small taxonomies (when the number of leaf categories is smaller than the size of the requested sample)
• in the further iteration, when the number of categories left to disambiguate is relatively small.

Ideally, the number of returned items in interaction should converge to 0. If it does not (i.e. stays above 0 for many iterations), then there is no way to disambiguate specific categories and create a complete mapping. The ultimate mapping and annotation will be suboptimal.