It seems to me that this first convert the csv stream to a JS array, which is further loaded by DuckDB. So the specific DuckDB velocity when dealing with (parsing) CSV files is probably shunted?
Likewise, a semicolon delimiter won’t be automatically recognized, although it is with FileAttachment().
I guess that the DuckDBClient needs to now the file’s type (.csv), which it can guess from the filename extension in “FileAttachement”, but not from the Input.file stream