Here is the problem. Our Clients can be grouped in several ways - Master Client/Related Client: Different offices can enter the same clients twice - each one generates a different client number and there is no unique ID that is applied to both of them to indicate that they are, in fact, the same entity. The Master/Related client relationship was an attempt to identify these clients; the master is the client that all related clients depend on, i.e. if the master client were lost, then the others would be also.
However, we also have control groups - which is a way of grouping similar clients, and all clients in a control group must have the same master client - however, the master client does not need to be in the control group! So, if you assign a client into a control group he must be assigned as related client to the master of the control group. If he was already assigned to a different master he must be unassigned and reassigned to the control group master. Now, that still doesn't identify which clients are in reality the same entity, so our steering committee decided that all new clients have to have their SSN or tax ID (TIN) entered when entered into the system and that any duplicate clients must have a single Master client and any duplicates will be related clients. This layer of complexity does not replace the existing rulers - it has to be added to them. So now if a control group has a set of master/related clients and you add to the group a client that is in a master client relationship based on the TIN, the clients in the control group must be assigned as related clients to the TIN groups master client. Considering that the master clients of a control group don't have to be in the control group, and the related clients dont either - how do you set up a test plan or matrix that would cover all scenarios involved?
Draw it! I'd try drawing it out first to try and visually discover the scenarios. This would be my first step. I think you could probably extract the majority of scenarios from a good relationship diagram.
Next, I would create some sort of mini-application that would take a data sampling and ensure that my relationship diagram accounts for all clients in the database (ie, assign all the clients based on my relationship diagram and see if I have any leftovers).
Finally, ask a subject matter expert. These people are great sources for information.