> A text file may be inefficient as is, but it's perfectly compressible, even with primitive tools like gzip. I'm not sure the SQLite binary format compresses equality well, though it might.

I hope you’re saying because of indexes? I think you may want to revisit how compression works to fix your intuition. Text+compression will always be larger and slower than equivalent binary+compression assuming text and binary represent the same contents? Why? Binary is less compressible as a percentage but starts off smaller in absolute terms which will result in a smaller absolute binary. A way to think about it is information theory - binary should generally represent the data more compactly already because the structure lived in the code. Compression is about replacing common structure with noise and it works better if there’s a lot of redundant structure. However while text has a lot of redundant structure, that’s actually bad for the compressor because it has to find that structure and process more data to do that. Additionally, is using generic mathematical techniques to remove that structure which are genetically optimal but not as optimal as removing that structure by hand via binary is.

There’s some nuance here because the text represents slightly different things than the raw binary SQLite (how to restore data in the db vs the precise relationships + data structures for allowing insertion/retrieval. But still I’d expect it to end up smaller compressed for non trivial databases