Skeptical Entry: Decentralized Social Media Does Not Measure Up. Yet.

A man who only audits the systems he dislikes is not an auditor; he is a partisan with a clipboard. The convictions on this site get the same treatment as the dependencies, so today I am grading the thing I most want to succeed: decentralized social networks. I have run accounts on two federated platforms for 19 months, logged the experience, and pulled the public numbers.

The architectural argument is correct and I will not spend words re-making it. Identity you can move, no single company owning the town square, protocols instead of platforms. All true. The question a measurer must ask is different: is it working? Five metrics, graded.

Metric 1: Retention. Grade: F

The migrations tell the story. Each major exodus from centralized platforms produced a spike of federated signups, and each spike decayed the same way. Public server statistics after the largest one showed monthly active users falling more than 30% from peak within a year. People arrived on principle and left on experience. Principle is a terrible retention mechanism; I say that as someone retained by principle.

Metric 2: The reply-guy ratio, or discovery. Grade: D

My own logs: a post with equivalent content reaches roughly 8x fewer readers on my federated account than it did on the centralized platform I left, adjusted for follower count. Federation without shared discovery means a thousand village squares and no market town. Some networks are fixing this with shared relays and composable feeds, which is why the grade is D and not F. The trend line matters, and this one points the right way.

Metric 3: Actual decentralization. Grade: C-

The dirty secret of every federated network: the node distribution is a power law. On the largest federated microblogging network, the top instance alone has historically held on the order of 20% of active users, and the top ten a majority. One popular new protocol launched with essentially all traffic through infrastructure operated by one company, with federation arriving later and thinner than the marketing implied. Measured by where the users actually sit, these systems are less decentralized than email, a 1982 protocol nobody brags about. Decentralization is a distribution, not a checkbox on the protocol spec.

Metric 4: Cost per user, honestly accounted. Grade: B

Here the numbers surprise in the other direction. Mid-sized volunteer instances publish their budgets, and they cluster around $0.30 to $1.50 per active user per year. Centralized platforms spend an order of magnitude more, then recover it through an advertising apparatus that is itself the thing we are trying to escape. Federated economics are genuinely viable at the co-op scale. The problem is not money. It is moderation labor, which is unpaid, burns out node operators, and does not show up in the budget figures. I interviewed the admin of my own instance: 6 to 10 hours weekly, unpaid. That line item is load-bearing and invisible.

Metric 5: Exit rights. Grade: A-

The one metric where federation simply wins. I tested it: migrated my account between instances and kept 100% of my follows and followers in 25 minutes. Post history did not transfer on that platform (hence the minus), but identity portability is real, working, and unavailable on any centralized network at any price.

The scorecard

Metric	Grade
Retention	F
Discovery	D
Actual decentralization	C-
Cost structure	B
Exit rights	A-

Verdict

The architecture is right and the product is not yet good, and confusing those two claims is the standard failure of people like me. I hold at 0.81 the belief that a federated or protocol-based network reaches 100 million monthly actives by 2030. That is a real bet on the architecture. But my confidence that the current experience retains a median user is about 0.25, and pretending otherwise would make this log advocacy.

I remain on the federated networks, posting into the smaller room, because someone has to be the early data point. But I keep the logs precisely so I cannot lie to myself about which room is smaller. It is mine, by 8x, measured.