Feature table

For Identifiers (just the complete URI and local part, not about mechanisms)

	HDL	PURLz
permanent	Yes	Yes
permanently resolve to a digital object	Yes	Yes
unique within our PID system without the DNS name portion of the	Yes; very flexible naming, including pseudo/paths	Yes, but
URL	?	?
works with VIVO, arXiv, OAIS (CUL), Voyager Catalog, and WorldCat etc.	Yes	does nothing in and of itself to not work with existing systems;
resolvable through a web browser	Yes	?
Simplicity	Yes	Yes
support for a local namespace prefix and an identifier part	Yes	Yes
support for opaque identifiers	Yes	Yes
Local part should not be Cornell branded	completely flexible	completely flexible
may be surrogate for physical object	Yes	Yes
should be short (so use 26 letters plus numbers)	Yes	Yes
should be easy to copy by hand	depends on the length	depends on the length and/or layout

	HDL	PURLz
Supports billions of identifiers	Don't know; limited to the practical capacity of the underlying database.	ditto
Robust architecture	Yes	?
Robust implementation	Yes	?
ability to request metadata about the identifier	Yes	?
lightweight understanding of identifier equivalence	Yes	?
should be easily discoverable by Google	Yes; the resolvers can't be crawled without providing a dump	?
integrate well with the "web architecture"	Yes	?
vitality checking	?	?
Need to avoid unbounded generation of surrogate persistent identifiers	?	?
PID corresponding to every Cornell NetID	?	?

	HDL	PURLz
Requires payment to external organization	one-time $50 payment per named resolver (not mirrors)	no
Can continue resolve IDs in absence of external organization	Yes	Yes

HDL is Corporation of National Research Initiatives's Handle system.
PURLz is Zepheira's and OCLC's new persistent URL work.

We think PURLZ will reject the creation of a duplicate PURL, but we're not entirely sure how it deals with duplication when a PURL name component overlaps a PURL domain component – e.g., if we inadvertently created a domain /a/b/ with resource 1234 and someone later used the name b/1234 in domain /a/. Maybe pure string comparison for non-uniqueness is all that's required, but conventions need to be established to avoid confusion.