Well, I looked at the circuit and shorting pins 4&5 definately does not seem like the proper thing to do. That will always put 0.95V at the reference pin. The circuit then divides this by 10 and each led represents on of those values. I didnt read the article, but I guess this is what they found from their filter by testing the circuit.
Try to take out the short between 4 and 5 and let us know what happens. That puts a higher voltage at the input than your circuit was supposed to handle. Though I thought the 3914 could handle that, maybe it is safe, but does not behave properly.
Perhaps you could measure the voltage at pins 8 and 5.