Also I was able to get diffs working without requiring a pk argument using a virtual hash column So now its just

meltano invoke dbt-osmosis:diff -m MY_MODEL

And you will always know how your changes are affecting output data during development. Software cares about code diffs, data devs whose transformation code and data amalgamate to a single software product need data diffs. And they need them easy / relative-to the dbt models. (this is fundamentally different than datafolds data-diff since we are comparing the data output of a model as it is now vs how it was in your git history)