calibration

[nesquik011]

OS : Windows

During calibration as a checker, I keep encountering issues and can’t figure out the cause. Here’s the situation:
1- The image quality is excellent.
2- The patterns are clear.

NOTE: The entire pattern isn’t visible in the image because my lens is a micro type, highly zoomed in on the ceramic calibration board.

Should the entire frame be visible to capture the whole calibration board and its surroundings? Will losing 30% of it cause a problem?

Additionally, I have a dark overlay because my projector is a different version than the ones usually used, resulting in a 1.84% drop between the camera and the projector. I believe this causes the dark overlay. Would this issue break the calibration?
frame_10.bmp

I reduced the gain to 0 and captured another image with the same setting. Additionally, I got the hardware trigger to work, which helped improve the pattern and sync. I'll include two images: the very first one and the updated one for comparison to show the improvement.

frame_10.bmp

frame_00.bmp

NOTE : i did place the projector little far , some of the sequences worked and some did not , some did raw arrows , and sometimes drawChessboardCorners is away to make it stable robust and may you explain why two different drawings ?

update

i made a script do the following
it uses the shading as png and the up vp mask as tiff raw float32
Detects checkerboard corners in the shading image
For each corner, collects nearby pixels (within radius R=10) that are valid according to the mask
Fits a homography from camera pixels to projector coordinates using those neighbors
Calculates LMEDS inlier ratio to assess calibration quality at each corner
Accepts/rejects corners based on inlier ratio threshold (≥0.90)
Outputs statistics and a visualization showing accepted (green) vs rejected (red) corners

and the image of calibration passed -____- i really dont understant why it get low LMEDS inlier results in the C++ may you explain why ? if the images are bad the decoder wont work , the generated up and vp will be bad so the script will also fail , so what is the secret ! i will go crazy soon enough i cant spot it kindly help @jakobwilm

also i used many Aperture Ranges from 2 to 16 to not use any gain , sometimes it work in the C++ but mostly it fails and work in python