I wanted to understand the big difference between the absorption peak in the configuration where the pump is entering in the sample by the absorbing surface and the other configuration where the pump is entering the sample by the not absorbing surface (flipped sample). So I made again a calibration and also a calibration with the flipped surface reference sample and I found a calibration factor R=17 1/W (flipped) and R=13 1/W (not flipped). For the not flipped sample I made a calibration history in this post and I found a mean calibration factor R=11 1/W with fluctuatoins of 10%. I don't understand why R changes so much. 

If we calculate the LMA15033 absorption with those R factors we get 22ppm for the flipped surface and 14ppm for the not flipped surface.