I like this framing. Scientific workflows probably shouldn’t depend on closed black boxes.

Exactly, a reproducible research workbench should itself be reproducible.

Everything here is open source and MIT license. The model may still be a black box, of course you can use an open model if you prefer.