Thanks for the quick response and information, George and Radu. I had not realized the process was so complex (although I am not surprised). I'm wondering if it might be a lot easier to detect "this is complicated" and just not perform autocompletion if it is. RD> Is your schema an exercise or is it meant for a real situation? Heh-heh. Not only is it a demonstration schema, it is a demonstration of how NOT to write a schema. So not real at all. GB> For performance reasons we use as context information for content GB> completion the path to the root, including attributes and eventual GB> previous sibling elements, including their attributes. In this case, the GB> context is like this: GB> GB> GB> <poem> GB> <head> GB> <sonnet> GB> <pLine num="02"> GB> ??? GB> GB> where ??? is the place we try to find what content is allowed by the schema. GB> GB> We also use Jing, and in this case it seems it will give an error on the GB> num="02" attribute and it recovers ignoring that value, probably using GB> the first pLine pattern, so then when we look into what values are GB> available for that element, we get the value of the first line. GB> GB> Usually the context information that we use is enough, but in some cases GB> that use co-occurrence constraints that is not enough and more context GB> information is needed, in some cases the entire document may be needed GB> to decide what to insert next in the document. Unfortunately in this GB> situation the context information we pass along is not enough and the GB> recovery that Jing does when we feed it with the context fragment gives GB> an unexpected value. GB> GB> I cannot think of a true fix for this other than passing the entire GB> document content as content completion context. A simpler fix, that will GB> solve this specific situation, may be to extend the content completion GB> with previous siblings of the parent.