Normal people measure the reliability of a return-receipt system as the
chance of finding out whether the message was successfully delivered.
An end-to-end design is much better for this than a link-level design:
it's much easier to implement, and it doesn't need support from relays.
provided, of course, that the delivery notification facility is
part of the design from the beginning, and that the recipient
end points are mostly willing to generate delivery notifications...
neither of which is true in the real world.
DSN proponents have a different notion of reliability. They think
they've done their job if the user receives a bounce saying ``This
message was relayed to a system that won't tell you what happens.''
not necessarily what the sender wants, but it's the best that MTAs
can reasonably do given the above constraints.
In theory, a link-level system can guarantee that the user will always
receive some sort of bounce, while an end-to-end system can't. So DSN
proponents say that link-level systems are ``reliable'' and end-to-end
systems are ``inherently unreliable,'' even though users actually find
end-to-end systems more reliable.
users do find end-to-end systems more reliable if the recipient ends
actually honor the requests. but it's a big "if".
Keith