in this table :

Overall correlation: ?0.901 unused: ? ? ? ? - 16.7088 [ ? 0/9 ? ] bin ?1: 16.7088 - ?2.0456 [5236/5928] ?0.858 bin ?2: ?2.0456 - ?1.6241 [5610/5946] ?0.744 bin ?3: ?1.6241 - ?1.4189 [5539/5925] ?0.641 bin ?4: ?1.4189 - ?1.2892 [5357/5873] ?0.516 bin ?5: ?1.2892 - ?1.1969 [5405/5972] ?0.497 bin ?6: ?1.1969 - ?1.1263 [5241/5939] ?0.464 bin ?7: ?1.1263 - ?1.0699 [5067/5903] ?0.408 bin ?8: ?1.0699 - ?1.0234 [4960/5934] ?0.416 bin ?9: ?1.0234 - ?0.9840 [4808/5941] ?0.376 bin 10: ?0.9840 - ?0.9500 [4700/5931] ?0.348 unused: ?0.9500 - ? ? ? ? [3339/6874] ?0.330

the average CC is 0.508 +/- 0.17

if y is the correlation, and x is bin number, and 0.330 is excluded, that line fits with a CC=-0.941 (therefore R^2=0.885)

is the median resolution per bin used for the overall CC?

No, the resolution isn't taken into account at all for the overall CC. It simply sets x=f-obs-1 and y=f-obs-2 and then applies the standard formula for the correlation coefficient. (Phil Evan's idea to use E^2 seems very useful; maybe we'll apply it in the future.)

how is "unused" treated?

Those reflections are not taken into account. It really is a very simple procedure. The command predates xtriage. Ralf