> loaded into a staging table from a flat file, performing the normalization and inserts into the destination tables. This is how I've always done things, but

> loaded into a staging table from a flat file, performing the normalization and inserts into the destination tables. This is how I've always done things, but I'm starting to think decoupling normalization from the entire process might be a good idea. This would have simplified design on a few projects (especially the ones that grew "organically") and seems like it would scale nicely. I'm starting to look at normalization as a completely separate concern and this has led to some fun ideas. E.g. if a system operated on flat files that were already normalized it could save time writing tools; you're at the command line and everyone involved can get their grubby mitts on the data.