> First there were a large number of different types of facts that needed to be stored. ... So additional provenance information was required about who put the fact into the database.
Wikibase (the underlying datastore of Wikidata.org, based on a MediaWiki extension) has had to include many of these features, if probably not all of them. It's organized as a document wiki with full history for each "subject" entry, and claims about "subject" entities can include references or other sorts of justification, including "deduced from". Multiple entries and disagreement are explicitly possible, but statements can also be editorially "deprecated", so that unambiguous errors are clearly identified as such and not reproduced as fact or relied upon in any way. Uncertainty, calendar systems etc. are supported for every "date+time" mention, including "start" and "end" indications for any statement.
It's worth mentioning though that querying is done on a separate system where only the "current" version of data is imported (ignoring all "deprecated" statements); it's not a true temporal DB in any real sense. But that's enough for their usecase, and the full history of entries is always available on the underlying "wiki" document store.
Wikibase (the underlying datastore of Wikidata.org, based on a MediaWiki extension) has had to include many of these features, if probably not all of them. It's organized as a document wiki with full history for each "subject" entry, and claims about "subject" entities can include references or other sorts of justification, including "deduced from". Multiple entries and disagreement are explicitly possible, but statements can also be editorially "deprecated", so that unambiguous errors are clearly identified as such and not reproduced as fact or relied upon in any way. Uncertainty, calendar systems etc. are supported for every "date+time" mention, including "start" and "end" indications for any statement.
It's worth mentioning though that querying is done on a separate system where only the "current" version of data is imported (ignoring all "deprecated" statements); it's not a true temporal DB in any real sense. But that's enough for their usecase, and the full history of entries is always available on the underlying "wiki" document store.