Olaf, have you looked at Jupyter http://jupyter.org/ at all? This now has a backend for Haskell and supports many of the things you mention.

Neil, I must have forgotten about Jupyter. That is exactly what I was takling about! I have become quite a fan of literate programming. Now I admit that Jupyter looks quite different from spreadsheets, so wont' fit the bill of what Trent was originally asking for. As I tried to express earlier, the strength of spreadseets in my eyes is the somewhat explicit data flow. (You need to highlight a cell to see its formula.) Hence I envision a Jupyter with data flow made explicit. You start by defining your input data. Single values, lists, matrices, whatever. You can define some functions, which will act as edges of your data flow graph. Then you combine these functions and apply to the data. For each (f.g)(x), all of x, g(x) and (f.g)(x) become nodes of a graph with (f.g)(x) hyperlinked to g(x) which in turn gets hyperlinked to x. Thus you can explore backwards from the result and see how the answer was derived. Of course a nicely written Jupyter notebook needs no hyperlinks. One can leave enough explanation in the markup cells. But not everyone is trained to read and understand such mixture of text and code. There might be an applicative functor (a variant of 'State SomeGraph') for what I'm after. Once I've defined it, I'll post it here. Olaf