The Wikipedia page has a good set of links for details, but basically REA is an ontology. Ontologies describe things that actually exist, so REA records real things that actually happened with a set of real economic actors.
DEB is a fictitious abstract model. "Accounts" and "ledgers" aren't real, they are fictions, artifacts of a model we use to indirectly track events. DEB doesn't even have a notion of economic actors that take part in an exchange. As such, it breaks down with multiparty transactions, for instance. DEB can of course be extended to handle such notions, but it's no longer just DEB and starts encoding something more like REA, just within a less meaningful foundation, eg. there is no such thing as a "normal balance" because this need results from a fictitious accounting model.
The article also mixes concerns that are not actually part of accounting but of different ontologies, eg. pending->discarded|posted is recording what may AND did happen, accounting is only supposed to record what actually happened. Which isn't to say that tracking what may happen isn't necessary, but mixing it into your accounting records is dubious, and simply muddies what's supposed to be a simple and reliable system that records what actually happened.
Just look at the sample REA pattern involving a cashier, customer and sales person. The information that this exchange involves 3 parties is not even captured in DEB. This is why I said you can reconstruct DEB from REA because REA is richer, but not the other way around.
DEB is a fictitious abstract model. "Accounts" and "ledgers" aren't real, they are fictions, artifacts of a model we use to indirectly track events. DEB doesn't even have a notion of economic actors that take part in an exchange. As such, it breaks down with multiparty transactions, for instance. DEB can of course be extended to handle such notions, but it's no longer just DEB and starts encoding something more like REA, just within a less meaningful foundation, eg. there is no such thing as a "normal balance" because this need results from a fictitious accounting model.
The article also mixes concerns that are not actually part of accounting but of different ontologies, eg. pending->discarded|posted is recording what may AND did happen, accounting is only supposed to record what actually happened. Which isn't to say that tracking what may happen isn't necessary, but mixing it into your accounting records is dubious, and simply muddies what's supposed to be a simple and reliable system that records what actually happened.
Just look at the sample REA pattern involving a cashier, customer and sales person. The information that this exchange involves 3 parties is not even captured in DEB. This is why I said you can reconstruct DEB from REA because REA is richer, but not the other way around.