UNF Version 5

Important Update:

UNF Version 5 has been in use by the Dataverse project since 2009. It was built into every version of the DVN, starting with 2.0 and up to 3.6.2. However, some problems were recently found in that implementation. Namely, in certain cases data normalization is not implemented fully to the spec. UNF signatures it generates are still reasonably strong statistically; however, this means that at least some of our signatures are not independently verifiable. I.e., if somebody fully implements their own version of UNF calculator, for certain datasets it would calculate signatures different from those generated by the DVN. Unless of course they implement it with the exact same bugs as ours.

To address this, the Project is about to release UNF Version 6. The release date is still being discussed. It may coincide with the release of Dataverse 4.0. Alternatively, the production version of DVN 3.6.3 may get upgraded to use UNF v6 prior to that. This will be announced shortly. In the process, we are solving another problem with UNF v5 - this time we’ve made an effort to offer very implementer-friendly documentation that describes the algorithm fully and unambiguously. So if you are interested in implementing your own version of a UNF calculator, (something we would like to encourage!) please proceed directly to the Version 6 documentation.

Going forward, we are going to offer a preserved version of the Version 5 library and, possibly, an online UNF v5 calculator, for the purposes of validating vectors and data sets for which published Version 5 UNFs exist.