Hi Mark: from what I understand, the incompatible mesh bonding writes a set of constraint equations to match displacements of nodes on each component at the common interface as closely as possible. This information was gained using the GeoSTAR/COSMOSM interface: we used to do this by hand before they wrote software to do it. You would identify a node on the first component at the interface, (say, node 7) that is very close to a node on the second component at the interface (say, node 32), then you could write a constraint equation such as UX (node 7) = k*UX (node 32), where k is some constant you'd have to estimate depending on the original distance between the two nodes. We would do the same for UY and UZ on the nodes to match interface displacements. I'm fairly certain that's what they mean when they wrote "...using multi-point restraints..". Their automatic algorithm should be far more accurate at writing the multi-point restraint equations - however it makes sense that you could see stress concentrations at the interface. To avoid that, it might be best to plot stresses at the element level (an option in the stress plot) instead of the node level. Then then errors should go down as you refine the mesh at the junction because at some refinement level the original location of nodes from each component at the interface will approach each other more closely - so the multi-point restraint equations will have less and less error.