The reliance on filenames to define (parametric) dependencies was among the reasons I later adopted nextflow. The model fit the type of computation dependencies better for my case. In the mean time snakemake grew and many DAG that were hard to describe back then are now expressed directly with snakemake primitives