Using "Alter Played By..." on a role works as expected for ordinary object and fact types, but fails when the role is moved from one generalized object type to another.

A generalization does not have a single expression standing for the whole type - it has one expression per specialization underneath it. When the role's new player is a generalization, CaseTalk does not ask which specialization is meant. It silently takes the first one it happens to find, so the role ends up attached to an arbitrary single specialization instead of to the generalization as a whole.

This only becomes obvious when the role was ALSO played by a generalization before the change. In that case the role already had one substitution for each branch of the old generalization. All of those branches are pointed at the same single expression of the new generalization, and the surplus ones are then discarded without warning. What was a role played by a generalization with several specializations comes back as a role tied to just one of them, so the resulting expressions in the before/after preview are wrong or incomplete and the change has to be cancelled.

In addition, when the two generalizations are unrelated (they usually are), the uniqueness and subset constraints that were on the role are removed as part of the change.

Workaround: adjust the substitution for the role by hand through expression input after the change, or rebuild the role rather than altering its played-by.