Have attached some screen shots of the output so you might get a better idea of the problem.
Frustratingly, it is some very slight discrepancy between the way GeoConvert is calculating grid square coordinates and the way I'm calculating grid square coordinates.
I tried increasing precision to 8 decimal places and it made worse, there were more masks.
I tried rounding down in all cases, but there was still the mask you got
My calculation seems to be correct, some / most of the time because I haven't noticed this before.
Without porting the exact code that IPACS uses, or having their list of coords for each grid square, it's hard to know how to improve this one.
Is the FSET conversion tool you use an IPACS tool, or from someone on here?
In the meantime, remember that in the settings you can turn off the use of masks, which will do the job apart from shore line regions.