The values I calculate manually for log2 fold change (using R or Excel) are always slightly different than the lfcMLE value, I'm curious why that would be (sorry, stats isn't my strong suit). For example (each condition has 3 replicates):

Condition 1 normalized counts: 18536.58274, 20600.98049, 22578.73105
Condition 2 normalized counts: 12705.48353, 8803.154338, 12085.95176

Condition 1 average: 20572.09809
Condition 2 average: 11198.19654

Calculated log2 fold change: log2(11198.19654/20572.09809) = -0.877422531
lfcMLE: -0.877250202

There are slight modifications to the fitting procedure which make it more robust.