"When we refer to the texture in terms of how the viewer code sees it, the first row (00 - F0) is the top row of your texture. For the columns they are left to right."
... and...
"The plane requires a 33x33 grid, which it samples at pixel positions 0, 2, 4, ..., 60, 62, 63 in each direction. "
I had trouble getting the right vertices showing in a sculptie, so i checked this, and it seems that along the columns the sampled rows are
0, 1, 3, .... , 59, 61, 63
where the shorter sampling interval is the first interval, instead of the last one as in sampling the columns.
Below is a pic demonstrating the effect. I made 64 x 64 sculptie maps with simple x and y gradients and random noise in [0,50] in those locations that should not be sampled.

In the right sculptie z-values of 128 are set in all those pixels that are sampled according to wiki (rows and columns 0,2,4,....60,62,63), giving a flat sculptie if the sampling is from those points. On the flat background there is ridge, with vertices 2 and 30 eleveted to value 250 along rows and columns, corresponding to rows and columns 4 and 60 in the texture. The resulting sculptie has only the edges correctly sampled, with all the interior points sampled from the random noise background.
In the left sculptie the flat background pixels are set to rows 0,1,3,.., 59,91,63, and respectively the ridge is in the rows 3 and 59 in the texture. The resulting sculptie is sampled exactly with the desired values.
The sculptie maps are here:
Wiki locations:

My suggestion:

The conclusion is that either SL counts the rows starting from the bottom of the texture or the column and row directions are not sampled in the same way, and when using any standard image processing software to prepare the texture, the sampled rows are 0,1,3,...,59,61,63.
If anyone could repeat the experiment with any indepedent test, I think this should be added in the wiki also.
--
LindaB