It has been rather convolved trying to look into this in an effective way. You shared the traces of the problematic cell as an SNT exported
.swc file when we asked for and needed the
.traces file. The problem here is that SWC files do not contain any metadata. It is for instance, quite cumbersome to translate those center coordinates you reported into the swc space. Your cell is rather simple and you may thing these requests are overzealousness, but the time we spend guessing what may be wrong, is actually time we could spend looking into what may be actually wrong.
Anyway, you should consider the possibility that one of the neurite extends tangential or quasi-tangential to the sampling circle. In those cases, and specially in a discrete matrix of pixels, a single process can contribute to multiple intersections at the intersection point.
This seems to be an actual possibility:
(BTW, This is actually, one of the reasons why the Sholl plugin implemented multiple samples per radius). It could also be that some rounding errors in the source code contribute to the exacerbated counts you find, made obvious by the simplicity of your cell, and the fact that you seem to be using a radius increment larger than your pixel size.
If you want us to look further, please ensure the issue exists in an updated Fiji installation subscribed to the Java8 update site, have a look at Bug_reporting_best_practices, then include in the shared zip file, the actual image used for tracing, the .traces file, and any other information that may be relevant.