Comment by jononor
Code using pandas is testable and reusable in much the same way as any other code, make functions that take and return data.
That said, the polars/narwals style API is better than pandas API for sure. More readable and composable, simpler (no index) and a bit less weird overall.
Polars made the mistake of not maintaining row order for all operations, via the False-by-default argument of maintain_order. This is basically the billion-dollar null mistake for data frames.