That's interesting, but I think I'm missing something.

To define the schema as json, you would probably read it from a bucket (example), let the spark/autoloader define the schema and save it in a file for validation. In this case, you are also relying in schema inference, so why don't you just use it directly in your ingestion?

I saw a lot of discussions up to schema enforcement. Some just store raw data as string in bronze, making the data type conversion for silver only.

At the moment, in my job, we just read all from the source and write with mergeSchema option. However, we do not define the data types and usually it leads to some strange errors. There are a lot of tables to manage and sometimes I feel that I don't have too much control over it.

If it don't get you in trouble, I would really appreciate a sample of your setup.

Again, thanks for the attention Brahma!