[erezsh]

Great question, this is something we put a lot of effort into.

For both dates and numerics, we format them in a normalized format, and round them to the lowest mutual precision. It gets even more complicated, because when databases reduce the precision of timestamps, some of them round it, and some of them truncate. We implemented both, and we either truncate or round both timestamps, according to which database has the column with the lower precision.

We haven't got to bytes and strings yet, but it's on our list, and I imagine we'll use a similar approach.

For now, we print a warning whenever we don't have special handling for the column type. If you see a value mismatch where it shouldn't be, let us know and we'll implement it next.