Buffer inconsistent and inexact

[WRingle]

I have a problem that is driving me crazy. (I realize there are some similar posts, but mine I think is distinct.)
I have a arcpython program to thin a large number of elevation points such that only the highest points within a particular radius (200 m usually) are kept. I do this by creating pairwise buffers around all the points as the first step.

arcpy.analysis.PairwiseBuffer(sorted_input_points, buffer_output, searchRadius,"NONE","#","PLANAR")

My problem is that the buffers are created slightly smaller than specified, so that in a few cases points that are less than 200 m apart are not selected (usually the distance is 199.97-199.99 m). I have visually inspected the buffers to confirm their size. All features and the Map use the WGS 1984 UTM 16N projection. I use the "PLANAR" option.

On the other hand, if I manually do a pairwise buffer, the buffer radius is correct and every point <= 200 m. from the focal point is selected, including the ones between 199.7-199.9.

Any ideas what is causing this?

[DanPatterson]

There may be something not apparent when using the tool's call in code, but try the geodesic option rather than planar.

Pairwise Buffer (Analysis)—ArcGIS Pro | Documentation

There is a lot said about things that can affect the results (eg coordinate system etc), it might be worth a read to see if you can narrow it down.  Make sure you are creating a featureclass in a local file gdb to rule some of them out.

[WRingle · Accepted solution]

That wasn't the problem and planar is more appropriate for UTM projections.

The problem is that buffers are actually multi-sided polygons rather than true circles. So if a point lies between vertices of a buffer polygon, if may fall outside the line between the two vertices, but in actuality be within a specified distance of the buffer center. Apparently the Densify command can help with this by increasing the number of vertices, but it will not give a true solution, since the same problem can possibly recur.

[DanPatterson]

strange, since featureclass buffers in a geodatabase are supposed to be represented by circular arcs not n-gons (which shapefiles use)