Hi, I have a graph-backed asset defined by

@graph_asset

. It is self-contained in a sense that it loads the necessary data in the first op and materializes the results in the last, but it still have some logical dependencies on other assets. I wanted to map these in the manner of

non_argument_deps

for “normal”

@asset

‘s but couldn’t find a nice way. I ended up “swallowing” the graphs inputs in the first op using

In(Nothing)

for each input, but it feels dirty, especially because it will invoke

load_input

on the upstream assets’ io managers for nothing. Am I missing something here?