Looking AT the Blue Skies of Bluesky

"decentralized approaches come with their own trade-offs, and therefore multiple architectures exist... we conduct the first large-scale analysis of Bluesky... [which] decomposes and opens the key functions of the platform into subcomponents that can be provided by third party stakeholders".

The intros of Fediverse and Nostr provide a useful spectrum of decentralised social media (DSM):

• Fediverse "shifts control to instance administrators... The downside ... failure of an individual server can result in data and account loss"
• Nostr "overcome these problems by greater decentralized replication, but consequently lack important server-centric features such as content recommendations"

Then sets out how Bluesky works (available elsewhere), focusing on 5 key components (DIDs, PDS, Relays, Feed Generators, Labelers): "any competing operator can build the above components, and users can freely select between them... a radical shift from the walled-garden approach ... and proposes novel innovations with respect to other DSNs".

But are people taking advantage of this approach? They analyse "5,523,919 users, 225,461,969 posts, 40,398 Feed Generators, and 62 Labelers" to find that the 'decentralised in principle, not in practice' argument (see Adoption challenge) is true only in part:

• "the vast majority of users’ handles are linked to bsky.social...
• [but there's] a growing ecosystem of operators now issuing a majority of labels just two months after opening this system component...
• [and] Feed Generators... the most popular and competitive component

The body of the report introduces the components in more detail, describes the dataset analysed, and presents the analysis. It's dated April 2024, when "1,855 events (out of ≈ 280M)" on the firehose were non-Bluesky lexicon (Whitewind was the 2nd most popular) and "98.9% of subdomain name handles were under bsky.social". Of the remaining "57,202 handles... no operator exceeds a few hundred FQDNs per registered name", with only 1436 using domains from Tranco's 1 million most popular domain list, which "suggests limited engagement with the Bluesky platform" from domains associated with major organizations.

TBD