I think the way I'm using it: a reverse IndexedEmbedded / ContainedIn is probably a kind of edge case that was not conceived of when this algorithm was written.

Yeah, I'm not sure about that ExtraLazy setting. When I wrote up this Jira I remember noticing ExtraLazy not being honoured and now I'm seeing it happening anyways. So I may be looking at different manifestations here. I guess I need to try to isolate this into a test case for it to be addressed appropriately.