Right, I'm taking a brief ill-deserved break from writing up reviews of WINE submissions to write this down.
When you take a hike on unfamiliar territory with the aid of a map, and you get lost, this is not something that happens suddenly. Uncertainty creeps in concerning one's geographical location, the facts are twisted to support an ailing theory, maybe the visibility worsens, and one is reluctant ever to backtrack. At some stage, one is reduced to a very crude strategy like "go uphill whenever possible" in the hope of finding a landmark.
Reading a badly-written paper is similar. The problem papers are not weak papers that one can reject with reasonable confidence, they are papers that may be significant, but can't be understood. And it starts out all innocently - a symbol is introduced whose meaning is not specified precisely, so you form a hypothesis about its meaning and try to continue. Terminology may be used in a non-standard way - you think you know what is being said, but something else entirely is intended. This is a problem in particular when the authors are not really in one's own community, authors who usually write papers for conferences you have not attended. Another problem is the usage of a technique that is attributed to a pre-existing paper, for which various properties or performance guarantees are being claimed. How seriously can I take these claims? At some stage, you realise you have no understanding what you are reading. The time has come to go right back to the very first point where the meaning was not made 100% clear, and write up a query on that as part of the review. Then try to do the same at the second point of uncertainty, although that second query would ideally be informed by the answer to the first query.