Steve,
The problem is that your "history" doesn't match the experience of any one else participating in this thread. You keep making assertions about testing being broken, sometimes with "hundreds of broken dependencies". Since one of the key criterion of packages entering testing is "dependencies are correct and fulfillable", this strikes most of us as unlikely.
Yes, and security upgrades never change behaviour of software and never break things. That's the way it OUGHT to be. The reality has its own turbulences.
I won't claim testing has never had a broken depends, but it's
very rare, and never hundreds of packages.
Well, I might have been out of luck. Maybe it hasn't been hudreds, just a "full screen of" (didn't count them and wouldn't remember anyway). That changes nothing on assertion, that using the testing routinely is not official, nor advisable way for ordinary users.
It's a basic point of science that the person making the unusual claims needs to provide the data to back it up.
My original intention was not, and still is not, to discuss capabilities of testing. I want to discuss possibilities, how could the stable be more attractive for ordinary user, how to make it usable on hardware newer-than-3-years-old, how could the user be blessed with fresh software rather than 2-years old, how to allow him to easily and effectively participate on bug reporting, and how to avoid the work of backporting security fixes to ancient software.
If You and several people claim they haven't met such problems with testing, I can live with that. I also heard people whose experience was different, and my personal one is closer to them. That's all.
Peter