solution:
see my fork here for an implementation of target generation with option to chamfer corners by n micrometers
https://github.com/nargetdev/kalibr_w_static_focal

background:
I noticed that particularly sharp images fail on AprilTag recognition .. or that on a slant there will be an absence of tag recognitions specifically at the focal plane.

i believe AprilTags are particularly sensitive to having a clean full white border - and that the separator corners intersecting are problematic.

I've implemented a solution and tested using the AprilTag3 iOS app. Screenshots attached.

I think it's quite clear that the grid with the chamfer is performing better.

note the aliasing (subtle grey stripes) of computer screen pixels against camera sensor pixels in the center of each image indicating focus.. i tried to hold the phone still (same distance) while switching the target.

Two images follow

1. no modification - sharp corners (perfectly in focus, very few tags recognized)
2. 500micrometer chamfer on separator corners (substantially more tags recognized)