You managed to get the right product... but through the most roundabout way. (and actually, the methyl ester stays intact in the product, so that part of the mechanism is wrong). Whether or not the ester oxygen is a better Lewis base than the aldehyde oxygen, the carbonyl carbon of the aldehyde is more electrophilic and more reactive than the carbonyl carbon of the ester. Thus, the alcohol OTMS (treated as OH) will form the oxocarbenium ion with the aldehyde as the first part of the mechanism.
Interestingly, the diastereoselectivity changes as a) the geometry of the double bond changes from Z to E, and b) the stereochemical relationship between the OTMS and the SiR3 changes from anti to syn. For example, in the E case with a syn relationship, the BOAT transition state is actually favored, due to favorable overlap of the C-Si σ bond with the C-C π* orbital as Dan suggested. We'll call this one complete. Movies, care to add anything? Did you work on this project or in the lab?
The reaction is from the total synthesis of bistramide A:http://dx.doi.org/10.1021/ol901801h
Background papers, including stereochemical outcome of all 4 alkene/diastereochemical geometries:http://dx.doi.org/10.1021/ja002087uhttp://dx.doi.org/10.1021/ol050982i