What actually consitutes a "regression?"
Good question, and thanks for asking JimDrake.
The basic idea is this: a test "fails" when something about it changes from the way it was when it was set up.
The notion is based on having a test be a sequence of actions that starts at a known state and moves the web application to a new final state, which can be documented as the "baseline" against which future runs of the test can be compared.
Here is a very simple example of doing this, testing a Windows application "Notepad" [which is kind of like a simplified Word process]: http://e-valid.com/Products/Documentati ... tepad.html
Here the baseline is simple enough because the application that you see is only really capable of showing text passages.
You can also -- in keeping with the simpler Windows example -- compare and later differentiate captured screen selections: http://e-valid.com/Products/Documentati ... tepad.html
For full-blown web pages you probably would be relying on looking into the DOM for differences, as described here: http://e-valid.com/Products/Documentati ... bject.html
-- eValid Support