Hi @Vladimir Ryabtsev​ 

It looks like it is actually the performance obtained by pruning.

As an answer to your previous query, I was not referring to the DFP based on joins. I was referring to the pruning by literal filters. I just used a common term DFP as I don't exactly remember what type of pruning they call this (filter pruning maybe lol).

Do check out this blog I found in my old bookmarks for further info.

https://www.databricks.com/blog/2020/04/30/faster-sql-queries-on-delta-lake-with-dynamic-file-prunin...

Cheers..