Inconsistent site list depending on map extent

In the process of dividing up a large CONUS map run into latitude stripes (such that the output files could be concatenated to reconstruct the full grid), I noticed that, depending on the latitude bounds of the stripes, a small number of grid points would either be omitted or added. So far, all of these points have been located right along the region boundary - so offshore or in Canada, e.g.

"map region" refers to the detailed polygon that, for example, mirrors the border of the U.S., and "map extent" refers to the rectangular polygon that generally encloses the "map region"

TLDR: It appears that this issue is benign, in that 1) only a handful of grid points are changed (omitted or added), and 2) these appear to be located along the boundary of the map region and may not be critical grid points.

I created sites files for the latitude stripes by modifying the latitudes in the map extent polygon. For example, where the full CONUS map extent is from 24.4 to 50.0 N, I create a number of stripes that go from 24.4 to 33.21 N, 33.21 to 36.71 N, etc. to cover the full extent (ideally, the selected latitudes ensure similar run times for each stripe).

The omitted or added grid points seem to be located along edges of the map region that straddle the map extent of a given stripe. It's as if nshmp-lib is creating a new region (java area?) that is the intersection of the map region and the map extent, resulting in very small shifts in the boundary (floating point issues?), rather than testing each grid point against both polygons.

A solution may be to test each grid point against both original polygons (map region and extent) rather than creating a union polygon.