Few comments on this PR!

• It's cool!
• I was not convinced that the output belongs into metadata rather than annotations/<set>, but I think it does not matter so much!
• I think the metadata file should contain information about the annotation set (annotation['set']) and the input annotation (annotation['raw_filename']). Maybe also the date of importation? (just like for the annotations' index).

What do you think?

I am in favour of putting the controlled vocabulary metadata inside the metadata folder. I find it simpler to have everything stored in one place rather than in several places.

I agree it's important to save the name of the annotation set in the metadata file. But I'm not convinced we should also save the raw_filename of all the files using this annotation scheme as all the files of a given set should be using it anyway...

For now the controlled vocabulary looks like this :

It's a good idea to add the annotation set !
I can add a column annotation_sets, with a list of the sets in which the tier is used.

I'm not sure about the raw_filename (and importation_date), the table would be difficult to read...

I agree that ideally all the files should share the same vocabulary within a set, but for debugging purposes it is still useful (I think) to keep track of where the information was retrieved. That does not mean keeping a duplicate entry for each file (so that should not pose any readability problem). Also this should allow to spot mismatches between annotations within a set if they occur. If you can't see a situation where this may be useful then forget about it! It's just an idea:)