See table on slide 10, line 3. Neven Simicevic (LaTech) got good agreement with independent JLab and LANL test results after including losses from the bevels. I think a 1" thick bar with quartz window pmt's and negligible glue losses should yield ~100 pe's summing the two pmt's. (The final Q-weak bars were half as thick with twice as many reflections, the glue cutoff wavelength was 250nm, but we couldn't afford 16 quartz windows on 5" pmt's so we ended up with only 16 pe's in Q-weak without pre-radiators.)