When Sequoia camera is triggered, a capture per each sensor it is taken. These captures are assigned to one rig instance to calibrate the rig successfully. If some images are missing from a capture, this will be discarded. In case more than 5% of captures is incomplete then the software does not further proceed with the calibration.
Unfortunately, the issue is related to the camera, and the only solution for the current dataset is to process it with the previous version 4.2.17.
In 4.2.17 the images are assigned to a capture based on the capture time while on 4.2.25 the images are assigned to the capture based on capture id. The last algorithm allows consistency in the processing of case images are missing. But since in your case the wrong tags are written by the camera than 4 images instead of 8 are assigned to a capture.
Below you can find the version 4.2.17.