Thanks for your inquiry.
The issue occurs because Aspose.Words calculates floater vertical position differently from MS Word when floater anchor is wrapped around another floater.
The problematic shapes do not actually disappear; they are positioned behind another shape in the previous table row in your document.
In the attached simplified 13347s.docx, the anchor line of the problematic shapes is moved down because of wrapping around an image. The shapes are positioned relatively to the anchor paragraph. MS Word uses the actual anchor paragraph position after wrapping to calculate shape positions. Aspose.Words uses the previous line bottom which is above the wrapped image. Because of that, the shapes are positioned higher in Aspose.Words layout.
The logic was specially introduced in order to fix some other customers’ issues. The tests for WORDSNET-4883 confirm that behavior.
It seems that MS Word behaves differently in this case because the wrapped image is anchored in the same paragraph as the problematic shapes. If the image is moved to anchor in another paragraph, MS Word positions the shapes the same way as Aspose.Words.
We have made some experiments to confirm the above condition but we quickly came to a document where we could not explain MS Word behavior. In the attached 13347sWeird.docx we introduced another shape anchored in another paragraph below that also moves down the anchor line of the problematic circles. We found that increasing the new shape height moves the circles anchor down, but it does not affect the circles positions. So it is calculated from some intermediate anchor line position which is neither original (before wrapping) nor final. This is bad enough, but there is more. In the same document, if the new shape is moved down, it affects the circles anchor line position in the same way as resizing. However, in this case the circles move down as well. So for the same anchor line position we get different anchored shape positions.
We need to understand the sequence of steps MS Word uses to arrive at the final layout and how to determine which intermediate anchor position should be used for the anchored shape positioning. This seems to be a complex issue.
We are postponing for now in favor of priority issues. You may make the problematic shapes page (cell) relative as a workaround.