Even for small variants, there are many cases when there are “simple context” + “multiple variants” clustering together, any sequence aligner could have hard time to get consistent alignment outputs. In such case, the variants will be “complicate”. Even for more accurate reads, this can be still a problem. This paper has a summary of a couple of good examples: https://www.biorxiv.org/content/biorxiv/early/2018/05/24/270157.full.pdf