How does RAM compute the transformed MOI for AISC's effective MOI?

Hello,

I am wondering how RAM Steel computes the I_{tr} for composite beams. For example, I have a composite beam with the following properties as computed by RAM:

But by hand, I am computing I_{tr} = 941.86, slightly less than RAM. I am guessing that this discrepancy arises from the way in which the transformed width of the concrete, b_{tr}, is computed. In my calcs, I used the modular ratio multiplied by the effective width, i.e.,

b_{tr} = b_{e} * E_{c} / E_{s}

where the concrete Young's modulus I compute from AISC 360-10, Section I2.1b : E_{c} = w_{c}^{1.5} * sqrt{f'_{c}}. Then, with the steel and transformed concrete properties, I find the elastic neutral axis, and then use the parallel axis theorem to compute I_{tr} with {b_tr}.

Where exactly am I, and RAM source code, diverging? The discrepancy is only about 10-in^4, but this actually makes a significant difference once you compute I_{eff} and then use that to compute the post-composite deflections. This couldn't be a numerical error for such a simple calculation. There is clearly a difference between what RAM is calculating and what I am.