Translation Algorithm for CILP model (Garcez, Lamb, Gabbay) - why difference in answer between my implementation and the answer from the book?

Question

I have implemented the Translation algorithm for CILP (Connectionist Inductive Learning and Logic Programming model by Garcez, Lamb and Gabbay) from their book "Neural-Symbolic Cognitive Reasoning"

The paper which describes the example Neural Network can be found here:

https://www.staff.city.ac.uk/~aag/papers/cilp.pdf

My implementation can be found here:

https://github.com/nravindranath10/CILP_Translation_Algorithm

While the book claims that the 2 iterations are required for the Neural Network to converge to the final answer which is {B}, in my code the Neural Network seems to take only 1 iteration to converge to the same answer {B}. I have highlighted the "m = 2" line in my attached image from the book.

I would like to understand if I am going wrong somewhere in my implementation.

Thanks, Ravi