I actually fully disagree. I find it much easier to "fake" the DB in the first stages (like just store shit in memory) where you explore and understand what yo

I actually fully disagree. I find it much easier to "fake" the DB in the first stages (like just store shit in memory) where you explore and understand what your system must do, the questions it should answer with data from the DB, etc. Then, just replace your fake repo with an actual DB and you're done. And it goes with what you're saying: changing a schema is not as easy as changing the code. So start by really understanding what you need to do, and then build the schema at the end.