suppose there is an electron beam of a few MeV and Gaussian cross section propagating along Z and passing through some medium. Beyond this medium there is a USRBIN set with type X-Y-Z and part PHOTON. The USRBIN is suitably wide in XY and positioned so that it can collect pratically all the photons. If spatial integration (e.g. with Mathematica) of the ASCII data file (“x y value value_err” rows) that is generated when plotting in Flair gives a number say equal to 0.011, is it correct to understand it as “for every primary electron 0.011 photons are created”?

not really.
The fluence value [in cm^-2] you get from USRBIN is the tracklength [cm] density [cm^-3] over the bin, i.e. the traveled path inside the bin volume divided by the bin volume.
When you plot its 2D distribution in Flair, keep in mind that the resulting value [always in cm^-2] depends on the considered interval along the third dimension (Z), over which the fluence is averaged (it can be - by default - the whole Z-interval of USRBIN, or a single Z-bin you specify in the plotting interface, or whatever other sub-interval you indicate there, and the result is obviously not the same).
Now, if you integrate this 2D distribution, i.e. you sum each bin value multiplied by its XY area, you will get an a-dimensional quantity, which shall coincide with the USRBDX integral fluence scoring on the whole XY surface (not divided by the latter!), provided that the considered Z-interval is small enough.
If instead you want to count particles, you should rather implement a USRBDXcurrent scoring.
The two give identical results only if the photons are crossing perpendicularly the scoring surface.
As regularly stressed, normally fluence is a more appropriate and meaningful quantity than current, since it is proportional to a detector response.

thank you very much for your kind reply. My bad, I forgot to mention that the Z size of the USRBIN (or of the Z-range set for the plot in Flair) is assumed to be small enough so that one can assume the value along Z for each (X,Y) to be approximately constant. Said that, I’m fine with obtaining the a-dimensional integral fluence scoring on the whole XY surface as representing the signal that a real detector would provide – thank you for pointing out once more this latter point!