I don't mean to be rude, but this sounds like user error. I don't understand why anyone would use an LLM for this - or at least, why you would let the LLM perfo

I don't mean to be rude, but this sounds like user error. I don't understand why anyone would use an LLM for this - or at least, why you would let the LLM perform the transformation. If I was trying to do something like this I would ask the LLM to write a Python script, validate the output by running it against the first handful of rows (like, `head -n 10 thing.csv | python transform-csv.py`). There are times when statistical / stochastic output is useful. There are other times when you want deterministic output. A transformation on a CSV is the latter.