Hi George,
I found this in my IRC backlog from November 7, 2014:
20:34 #tor-dev: < asn> karsten: how should one read these graphs https://metrics.torproject.org/bandwidth.html#bandwidth-flags ? 20:34 #tor-dev: < asn> karsten: can the gap between two lines be interpreted as available (unused) bw? 20:35 #tor-dev: < asn> karsten: so the gap between the yellow lines is the available guard bw? 20:35 #tor-dev: < asn> karsten: but there is not that much available exit bw, right? so it's not entirely accurate.
These are all fine questions.
First of all, the gap between two lines is *supposed* to be interpreted as available (unused) bandwidth. Advertised bandwidth is the number of bytes that all relay together were willing and capable to handle. (It's the minimum of the three bandwidth numbers contained in server descriptors.) Bandwidth history is the number of bytes that all relays together actually handled. (It's the sum of byte histories contained in extra-info descriptors.) So, the difference of the two should be the number of bytes that was left unused, in theory.
However, I can't say how accurate this comparison is. If I had the time to investigate, I'd start by downloading the latest descriptor archives from CollecTor and comparing advertised bandwidths and bandwidth histories per relay. If you or somebody else wants to do that I might be able to give them a hand.
All the best, Karsten