> This attitude makes sense only if you are used to very bad “persistent
>> URL” systems. A URI is an identifier. Making it persistent is our job.
> Using a different identifier scheme won’t make our job easier.
I totally agree with all these statements as well as with the sentiment that
the approach I advocate is far from optimal.

My basic philosophy is that: 1) the greatest weakness in any system can be
found in the carbon-based liveware it depends on (i.e. people act like
people) ; 2) you can totally count on the second law of thermodynamics (the
entropy of a closed system always increases); and 3) there is too much work
to go around.

Translated for the case at hand, this means: 1) people will inevitably not
have enough time to do it right; 2) Data get more complicated and less
consistent; 3) The problems aren't going to be fixed. As a result,
methods/systems need to be engineered accordingly. This makes our job hard,
but that's employment security for us as that's where we contribute value to
the equation.

> can you give a practical example? I can see embedding an id somewhere in a
> digital file, and then creating a link to it as part of the indexing
> process, but what about external content that we have no control over... yet
> are expected to reference in a consistent way?

> As you observe, reality is messy. With regards to externally referenced
content, the options are limited. Ideally, the provider embeds their own
identifier either because they just do it, or they were convinced of the
value of doing so.

The reason I favor not being too prescriptive of syntax is that identifiers
are insanely useful and if you ask people to do anything they don't
understand or want to mess with, you'll inevitably find they ignore you
because they have too many other things to worry about. For maximum
compliance, barriers need to be low as possible.

But to get back to the example, let's suppose they don't provide any kind of
identifier no matter how much you bug them. Guess what the resolution
service provider's chances are of being informed if they move all the
content or even worse, change the system that serves the content?

Has anyone thought through, or put into practice, using Apache mod_rewrite
> tables for this simple "redirect one URL to another" use case?

Unless the URLs being directed to can be predicted from the source URLs (an
assumption that is only safe in certain types of closed systems), this is
just a different type of resolution service that suffers from all the same
issues as purls and handles.

To summarize this long email into a single sentence, you'll notice the ideas
that work the best and prove the most adaptable in the long run are simple
and compelling.