gary,

Its a function of how neurals work. All they can do is retrain the neural network with a new set of training data. Therefore giving different weights to the new pathways. You will find the new ratings can have as many issues as the old ones (might be the same issues might be different issues). It sounds to me the data they input into the neural network was flawed. Its the most common issues with neural networks.

I don't know how many people have looked into neural networks and how they work but its not a bunch of rules to devise the ratings. Its all behind the scenes and is weighted by the training data. The actual weightings on the values passed in are unknown to the users. We just trust the data input was correct and therefore the output should be right.

Just because you don't agree with the patterns it uses doesn't mean its wrong. Logic though tells us it is wrong but then again maybe its smarter then us.