well this is the thing I've been talking about on last thursday. Look at these two pictures. ;) Maeybe this kind of analysion is useless but surely it shows what problem we are dealing with here. On my plot (green dots) I chop off all negative values, all the values above 10000 for multipoles greater that 300 (which isn't a big crime), and throw out all points of which "errorbars" were bigger that the values itself. The red dots are raw data as it goes from LAMBDA.
Well perhaps (I hope ;) somewhere among green dots are those we see at the picture from NASA.
bartek