Hi @hpant would you consider testing the new VARIANT type for your JSON data? I appreciate it will require rewriting the next step in your pipeline, but should be more robust wrt errors. 

Disclaimer: I haven't personally tested variant with Autoloader. It should speed up some reads, but I don't think it yet creates file statistics.

Alternatively, you can partially define a schema and put the remainder of the columns in a 'rescued' column, but it is a hassle splitting it out afterwards.