What don't you understand, or consider fragile, about the "routine"? That's a very small part of the automation I use, which all worked perfectly under Jaguar. Nor was it a flame toward Apple: I reported it as a bug in RADAR before writing the article, waited for their response before finishing the article ("it's been reported before, we're still looking into it"), and talked to a bunch of other people concerning possible fixes before revising said article. It's "top-billing" solely because it's the latest published article. Monday, they'll be a new "top-billing".
Concerning pppd, I didn't know about the plugin architecture (save for the /etc/pppd directory). You think that the "if the user has manually disconnected, no longer dial automatically" is a simple turn-off-able flag in a ppd script? My impression was that the decision was made somewhere in a GUI library, which just stopped pppd from being called ever again (which can be confirmed by disconnected manually, then attempting to have an autoconnect occur, and NOT seeing pppd's fleeting absence in the logs or the ps).
As for dialing purely with the pppd daemon, I've tried that. Capturing the output of the pppd with ps, as well as some CCLEngine script didn't prove very helpful: running the exact command shown in the ps didn't work, nor did passing STDIN or command line flags to the "real" pppd from my "fake" version. As the final paragraphs attest to, it's my ignorance of pppd that heightened my failure.