I really haven't seen a definitive answer on this.  I am wondering if, when I see 'create a sandbox', this is what you are doing (as I'm doing).
1) create a new RTC install that's in isolation to other RTC (typically, on your own laptop that uses localhost) using 10 user license and derby db
2) create fake users (dev, test, etc.) with appropriate license.
2) If necessary, import template you are going to modify
3) create project area using the desired template as your baseline, and add appropriate dummy users to test (dev, test, etc.)

4) start playing. This could mean creating testing, then abandoning attributes if you try something then need to abort it because you tried one type of attribute, but found it didn’t work like you wanted, but you can’t change the type after you created it (at least I can’t, ‘type’ is greyed out).

5) once you get close to a working template, you export it in your isolated instance (but give it a non-official name/ID since it’s likely this isn’t the final version), then create a new project area that is based on that new instance, add the same users, and test your template.  Consider this your unit test of the template.

6) repeat steps 4 and 5 until you get a good working template that seems ready to roll-out.  Give that semi-final name of *-beta.1  (since you likely aren’t done yet). Then export the template you want to test to a test instance where others can see it. Create another working instance off that beta.1 template, then let others test it.  Each template beta implementation should increment the beta number (beta.1 to beta.2 to etc.)

7) When you finally get done, the final beta looks good, then save/export it as the official template version (drop the beta name).

8) destroy the isolated sandbox instance (so that gets rid of all the abandoned attributes) by uninstalling RTC/JTS and deleting the workareas.

Is this what people are basically doing?

Another question is, for large enterprises, is there a possibility of ID collisions (if my exported template has an attribute ID of defect.newfield.attribute  for my final template, and some other RTC server also has the same defect.newfield.attribute, especially if that other server has a different type (maybe medium string, may be enumeration)?  What happens then if you try to import your template that has the colliding ID?

And a final question, I haven’t seen this (may be I missed it), but is there a way to upgrade a project area with a newer version of the template (assuming existing project areas didn’t modify anything so there’s no risk of colliding ID’s?)?