Re: candidate keys in abstract parent relations

OK, upon rereading Codd's definition of a surrogate key, I see that I
was using an overly restrictive definition (although the literature I
have available to me is not at all clear on what, precisely,
constitutes a surrogate key, and how it might or might not differ from
a typical auto-incremented sequence. Celko's long tirade on
auto-incremented keys as "exposing physical state" is particularly
unhelpful in this regard.). It still bugs me that my table, as
currently defined, has no candidate keys that don't include a generated
identifier. But I guess that's real-world database design!

(I'd still like to know if there's a general method for defining a
parent row's identity in terms of a set of child rows.)

.