When an additional variable (for example, float *test) was added to the grid object. It would produce a different answer. Accessing an array out of bounds and not initializing this array was causing the problem.
A new gold standard needs to be generated as this changes the answer.
I verified this earlier today comparing a local result generated on enzo-dev and a local result generated on enzo-3.0. Before this fix, the answers were different. Now they are bitwise identical.