> A list of nonconformities for each R7RS implementation would be better
> and simpler, I think, than trying to reduce it all to a single number.

True.

Best of all would be a de facto standard test suite, and a well-defined
way to run it on each implementation.

What's the closest thing we have to that -- the Larceny test suite?

> In principle I think this should be included with the implementation
> itself, like the BUGS section on man pages.

This is a good idea.