Either: (i) 263 is incorrect; or (ii) my logic as to frame height is incorrect. Given that the horizontal side of things is really well documented, I'm currently guessing (i). Research to do.
Specific issue: the repeated (start_column != end_column) test, which can no longer be correct if start_column has been incremented in the (x_&7) test.
The visible effect was to omit pixels from the output wave, which also affected observed sync timing.