There are models generating raster outputs bleeding out of the dateline on both sides, typically by half a pixel (the model uses a pixel is center approach, and samples both dateline ends as if they were separate data set).
When rendering these the code might end up missing some bits when one end of the image touches the dateline, basically:
The code sees the coverage might be a 0-360 case since it goes beyond the dateline, and reads a tiny portion off the second dateline
The sliver image throws off rescaling since a 1 px or 2px image cannot be made 20-30% smaller
The mosaicking code putting them toghether ends up picking the grid to world of the sliver, resulting in an overall image that has wrong proportions (so the data is all there, but compressed in the horizontal direction)