I think it's interesting to note how this is usually handled in Haskell. Thanks to the Curry-Howard isomorphism, the (most general) type of any expression in Haskell can be inferred, and thus type declarations are essentially not required anywhere, with a few exceptions; for example, sometimes you deliberately want to limit the type to something more specific than would be inferred.
Of course, what is required and what is recommended are not the same thing; in practice, the convention seems to be that top-level definitions always have type declarations, while localised definitions have the type declarations left out. This seems to strike a good balance between explicitness-for-readability of the definition as a whole, contrasted with brevity-for-readability of the local "helper" or "temporary" definitions. If I understand correctly, you can't use var for "top-level" definitions (like a method or global function) in the first place, so I guess this translates to "use var everywhere you can" in C# world. Of course, typing "int" is the same number of keystrokes as "var", but most examples will be longer than that.
