But then there are things like GPT-3 , which means stashing everything in a rigid schema isn't as hard-core of a requirement as it used to be.

OTOH, facilitating:

    1. access to the raw data
    2. access to the metadata
    3. access to the source code of whatever software was used / created to run the experiment
    4. making sure everything is computer readable (i.e. not a 256x128 graph as a PNG embedded in a bloody PDF)

should be a requirement for any scientific publication worth its salt.