ETL vs ELT
Whether data is transformed before loading into the target or after it, which decides where the compute happens and how much raw history you keep.
ETL transforms in a dedicated processing tier, then loads the clean result. It was the norm when warehouse compute and storage were expensive and coupled — you did not load anything you were not going to use.
ELT loads the raw data first and transforms inside the warehouse or lakehouse using its own engine. Cheap object storage and separated elastic compute made this practical, and it has two real advantages: the raw data is retained, so a transformation bug is fixed by re-running rather than by re-ingesting, and new questions can be asked of history that the original transformation would have discarded.
The cost of ELT is governance. Raw everything, transformed by whoever needs it, is how a lake becomes a swamp — which is what the transformation layer, lineage and schema enforcement in modern lakehouse tooling exist to prevent.