Could someone maybe explain what the difference is between 3D and LRUD envelope?

I get how 3D probably works based on the rendered 3D view of the cave. But how can the LRUD thing work when I only use splays instead of LRUD? Does it just pick some particular or average splay and use it as a LRUD measurement? The results are so vastly different. Sometimes much bigger (one of my simple caves) sometimes much smaller ( a pothole).