Hi, on the web page where I posted results of systematics due to randomization of S/P: http://www.slac.stanford.edu/~petrella/systsp.html you can see that this systematic error is not stable, for example when cutting on integrated purtity. Now I'm trying to look at other results from these jobs to see if I can find what makes this errors so large, but probably this is also due to the S/P ratio and its error. For example if you look at the correction factors for IP > 0.50 (http://www.slac.stanford.edu/~petrella/tmp/SP_allrew/SPallweights/ip050_allrew/corrallwip050pol1.eps) you can see that the first bin has a large error (the exact value of the correction factor for this bin is S/P = 5.67 +- 5.34) These numbers (they're on the spreadsheet at http://www.slac.stanford.edu/~petrella/tmp/SP_allrew/SPallweights/SoverPFullRew.sxc) come from the double ratio of S/P on MC (0.74 +- 0.13) times the S/P ratio on data depleted sample (http://www.slac.stanford.edu/~petrella/tmp/SP_allrew/SPallweights/ip050_allrew/data_depl_AC_intp0.50_0.001.55.eps) On data depleted sample the signal component (fitted) is 291 +- 31 and the background component (fitted) is 38 +- 35, so the error on the final S/P ratio is driven by the background component on data depleted sample... and cutting on purity (and having less background) will give roughly higher errors on background component (at least the statistical error). For the data depleted sample we get S +- dS and P +- dP as they come out from the fit and then we compute the quantity S/P +- d(S/P). But these errors are correlated, aren't they? ciao, Antonio