Downsampling shrinks the candidate search area
With the same search distance (R_buffer = 10 m), the full survey data produced 36,110 candidate rows, and the thinned data 22,484. That is 1.6 times more candidates for the same 907 stays.
This is expected. Candidate selection draws a circle around every GPS point of a stay. The full data has about 9.9 times more points, and more circles cover more ground, even with the same radius. So thinning the data also makes the search area smaller, not only the number of points. Keep this in mind if the search distance is tested again (R_buffer: 10 m vs 15 m).
Part of Raw-density refit.