I don't know whether you'll find this useful but perhaps it might start to explain what I was on about with windows and so on, I'll use your measurement as an example. This is intended as a laymans guide to windows in REW hence the annotations within each pic. Let me know if this makes sense or is just as confusing as the REW help
Lets start with your C measurement, here's the IR view zoomed out to show the window. By default, REW uses a 125ms left (hann) window with a 500ms right (hann) window. The details aren't important right now, just that it's a v wide window. What does it mean to say we have a wide window? It means we are including a lot of sound that arrives at the mic at many different points in time, i.e. we have the actual sweep itself played by the speakers plus all that sound that then bounces around the room for the next 500ms.
The FR is just another way of looking at that IR, it looks v ragged and becomes increasingly hard to interpret at frequency rises. It's just very noisy.
The usual reaction is therefore to apply some sort of fractional octave smoothing to make it less noisy/easier to read. This is the same measurement with 1/6 octave smoothing.
However there is another way to interpret this data and that's by changing the shape of the window, i.e. by changing what data (recorded sound) we include in the analysis. We can look at what this does in practice.
Lets first zoom in on the first 25ms or so of the IR so we can see in a bit more detail what is going on
As we shall see, all that low level fuzz is what makes the full range FR so noisy. Lets change the window to see what happens.
This is a 1ms left (hann) window, 20ms right (blackman harris 4) window. Notice the light pink line and how it is closer to a flat line as time passes (the graph moves to the right). The blue line shows the shape of the window, if you think of the previous straight line as 100% to mean "include all of the content in the windowed response" then that line sloping down means "take less and less of the actual recorded content into account when analysing the response".
Look at what happens to the FR as a result, not so hairy but we lose LF information. Basically the shorter window means we have reduced frequency resolution *and* eliminated some of that late arriving futz.
We can continue and make it even shorter, this is a 10ms right window.
and the resulting FR
The lost LF resolution is a problem. We can use frequency dependent windowing (FDW) to try to recover this but that's for another post (as well as explaining how the spectrogram or waterfall view uses windows).