With adequate light you should be able to do an indoor test; though it is better if you can utilise daylight through a window as it tends to be much stronger than indoor lighting.
Assuming you haven't already, try your tripod based focusing test using Liveview. Although the Liveview AF on the 5D4 is also phase based (like through the viewfinder) the AF is performed "on sensor" so MFA calibration isn't an issue. If that's failing repeatedly then I would suggest the lens was defective.
However, if that works, but AF through the viewfinder is unreliable, then look into using Sigma's USB dock.
I have the 35A (and a 5D4) and found that I could MFA it for one focus distance, but that it was unreliable at other distances. I went through the DotTune method with the lens mounted on the camera, and the sensor plane-to-target distance at the four different distances that can be calibrated using Sigma's Optimization Pro software. For each distance I found the camera's MFA value that got me the best focus.
Although Canon's internal MFA values don't necessarily correspond to Sigma's, I found the MFA values I got from DotTune to be pretty close (it only took a couple of repeated test cycles to nail it down). Now the lens pretty much never misses - even wide open.