It would be nice if fake records were directly and explicitly marked, as opposed to tracking them in an external file, so that at any point of time you can discriminate between fake records and real records.
Why:
1 - To not have to worry about 'up' and 'down' faker states; less worry of corruption
2 - For a richer app development situation, where there's a mix of faker data AND human inputted data, and to be able to discriminate between either at any time
3 - To be able to dynamically toggle fake data on/off downstream (e.g. in the app front-end)
Perhaps a system as such:
For any model which would have fake data generated, a column is deemed to be exclusively used to just track faker state. Perhaps strapi-faker generates this column ("faker_state") ?
When fake records are generated, they get a specific flag in this column (like "1"), marking it as fake.
Now, management and access of fake records is easy as just filtering for the faker flag.
Yes, this changes model structure, but during app development, model structures are changing anyways. Anyone using a faker plugin is presumably pre-production.