This was just an early snapshot release, so for me at least, the canonical source is arcn.tar. I don't know much about the conventions for this sort of thing, but I'm still working on the foundations, and getting that right, especially, is something best done by very small (even n=1) groups. All the more so since CL suffered so badly from doing the opposite; the core of the language seems like it was designed by 20 different people, each with slightly different design philosophies, but all with slightly overlapping functional territories.
I plan to do releases frequently though, and I've been incorporating suggestions from this site.
Perhaps then the community can still forge ahead with a Git repository to share ideas and bug fixes and suggestions for improvements, which you can feel free to incorporate or ignore, and then the community should just baseline their Git repositories off of the arc<(max n)>.tar?
Yes, exactly. The Git repository is terrific resource as a place to conveniently store everyone's patches, but of course some patches may be "bad" (by one criteria or another), and that is OK (good even) for its purpose.
I'm running Linux, and so date was broken and thus so was the web server, I looked in the repository and was happy to see someone had a fix, and I didn't even have to read the date man page (lazy me!)
So, just like you say, Paul's canonical releases of arcn.tar and a community repository (for good patches, bad patches, experimental patches, weird patches...) are both useful.
I don't know, look at how GvR treats lambda: a weak version and he still tried to back it out of the language at one point. Even when benevolent the judgment can stumble -- hell, McCarthy still does not like sexprs! :) The good thing about Arc is that it has decent macros, so not even PG can dictate to users -- we'll just MAC an end run. :)