How does Smart Render really work?

I know what Smart Render allegedly does: If it was enabled the first time you rendered, then you make a few small changes and re-render, the parts that were unchanged will just be copied over and not-rerendered, with only the changed portions re-rendered, thus greatly reducing rendering time if the changes were only on a small part of the timeline.

But there is a lot unexplained in the rather brief statement in the documentation (which is not much more than what I wrote above). Specifically:

1. Where does it copy over the previously rendered output from? Do I have to specify (and hence overwrite) the same filename for the re-render, so it copies over the unchanged blocks from the same file and just inserts the new blocks?

2. Or can it be (or must it be?) a different filename but the original filename must be in the previously saved location since that's where it will expect the previous render?

3. Or does it save this information in a temporary file somewhere? And if so, is this retained between invocations of VS? Or if I close and reopen VS will the cached information disappear?

4. Is there some indication that Smart Render is actually being smart? Other than the box being ticked and your estimating how long the re-render is taking? I.e., some clear indicator like "Skipping xxx seconds of previously rendered output ..." or similar information in a logfile somewhere?

I am seeing cases where only 10% of the total time was changed, output coding parameters were unchanged, and smart render was on for both the original and new render, but there did not seem to be any noticeable shortening of render time.

(As a secondary question, is smart render "smart" enough that if only the audio content were changed, that it would retain the more time-consuming video render and just re-render the audio and remux in the audio track?)