1) The list of powernodes is derived from running a modified version of Eigentrust on the network, and selecting the top n nodes. This is run at the publisher/domain level.
2) After identifying the powernodes, they are used to rank news at the article level, based on how many powernodes are linking to a source.
Therefore the powernodes are selected based on the entire graph, and can change over time. Testing our ideas on the current media landscape seems to show the algorithm does a good job of identifying local maxima.
For example: Our current list of around 60 powernodes was generated from a graph of around 3.5 million articles from the last 3 years. It includes sites bitcoin.com and coindesk, two sites that are relatively new compared to several of the larger incumbents.
From the left, it includes sites like salon, vox, and msnbc.
From the right it includes sites like breitbart, the daily caller, and the washington examiner.
All trustless cryptography based systems have a threshold of malicious participants they can tolerate before failing. Bitcoin, for example, has a well known threshold of 51%. The media, however, is diverse enough that the idea they would all somehow collude with one another to game the system seems unlikely.