Discrete Transfer Function > Difference Equation Implementation Problem

I have a discrete two pole, two zero filter that simulates pretty well in Simulink using the discrete pole-zero block. The system is a little pathological in that one pole is at z = 1 (dc, pure integrator) and the other is close to the Nyquist frequency.

I did the algebra to convert this to a difference equation. I built a Simulink model using gains, summers, and integer delay blocks. This was done in the standard Direct Form I.

When I simulate the difference equation model, I get results that are very different from the pole-zero block model.

I have gone over the algebra many times and am pretty sure it all correct. Any suggestions to help me find out the source of the difference?

Thanks in advance, Bob White