Precedence of NA trait-level metadata over dataset-level metadata

[yangsophieee]

Currently, trait-level metadata is overwritten by dataset-level metadata when the trait-level metadata specifies NA, when the goal is for dataset-level metadata to be overwritten by trait-level metadata in all cases. For example, if the dataset-level replicates field is '3', but the trait-level replicates field for plant_growth_form is '.na', we want the '.na' to ideally overwrite '3' since plant_growth_form is a categorical trait with no replicates.

[yangsophieee]

Thoughts from @ehwenk:

For replicates at least this seems like a "garbage in, garbage out" scenario we shouldn't try to engineer. As in, I don't really feel like people should ever assign replicates: 1 at the dataset level if they plan to overwrite it with replicates: NA for some traits. But of course there are many studies where there is a dataset value like field field_experiment which is then overwritten by location as either field or field_experiment - but never by NA's. The initial idea was that one either reported at dataset or at traits, not that there was a partially incorrect value at dataset that was overwritten as convenient in other fields. We adopted it to make that possible, but I don't think people should be encouraged to input inaccurate dataset level values, just so they can fill in less information at the trait level.

Closing this issue now.