I've started to develop a real bias for self-documenting tests. Here's the regression test I just got working:
/**
* Tests Bug 57: when you are actively involved with a conversation, the
* new responses aren't accounted for as read in the list view.
*/
@Test
public void TestBug57() {
openCommYou(Mark);
// Start the conversation...
int convId = startGenericConversation(Mark);
checkOpenGenericConversation(Mark, convId);
// ... respond to it several times...
addGenericResponse(Mark, convId);
addGenericResponse(Mark, convId);
addGenericResponse(Mark, convId);
// ... and go back to the summary view. It *should* show as completely
// caught up. The bug was that it wasn't doing so.
goToSummary(Mark);
refreshConversationList(Mark);
assertNoNewResponses(Mark, convId);
}
You can kind of figure out what that's doing even if you don't know the system, and it's wholly obvious if you do. The comments are almost redundant. *That* is what a really good test should look like in my book: clear, concise, and trivially easy to write new one... 
