It is a teaser at first because so many pure or functional-leaning languages have REPLs: sml, F#, Haskell, even Python.

Is there a common way to capture the "right" program from the REPL at the end of all the small changes? The final program doesn't need the experiments along the way.

Now building regression testing to sample data from the whole population in hopes it will catch problems when they’re introduced. May start declaring dtypes on pandas columns next.

Type inference / strong but undeclared types in pandas seem more the current pain point. With current data inference goes one way, but with the next set of data it goes another.

I’m dealing with this currently in Jupyter + pandas, and a notebook intended to be shared and reusable. Too easy to make a change that works with the current data of interest, and commit back.

Sounds like QUIC