I've documented pretty much everything new in 0.4 now, with notes on what's new, for the most part, so 0.3 users won't be terribly confused.
Playbooks docs are now in 3 parts -- a basic section (mostly required knowledge), an advanced section (of which people can pick and choose what they learn), and a best practices page that talks about organization and so forth. Upgrades welcome.
This isn't super task oriented yet, but I expect that to grow over time, to explain why, for each feature, you would want to use it, and to provide context for examples.
While developing this, I realize there's a little bit of boilerplate involved in the new best practices approach.
As such, I think it could be said that if a playbook does not have a tasks section AT ALL, we could know to try to load tasks from tasks/setup.yml (or main.yml, whatever), and if a playbook does not have a handlers section, we could load from handlers/main.yml. The same could be said for assuming vars_files loading vars/main.yml if that existed.
As such, the best practices "acme" playbook would be nothing more than:
hosts: webservers
user: root
And I think that's pretty cool. Convention over configuration. I'm going to make a feature ticket for 0.5 to do this.
--Michael