• Moderation is the decoy issue
• Salience is governance
• Nostr gets one important thing right, and one thing dangerously wrong
• The algorithm is not magic, it is landlordism
• We should stop pretending engagement is consent
• Reputation is becoming the hidden layer
• The real fight is over interface sovereignty

The censors are not where the action is. The real power sits one layer down, in the machinery that decides what becomes visible enough to matter.

This is why so much platform discourse feels stale. We keep arguing about takedowns, bans, content policies, trust and safety councils, and the theater of whether some particular post should remain online. Meanwhile, the actual governor of digital life is ranking. Not deletion, selection. Not speech suppression, attention allocation. The feed is less like a newspaper editor and more like a constitution written by people who insist they are merely tuning relevance.

That sounds melodramatic until you look at how public reality is produced. A post does not need to be deleted to disappear. It just needs to lose the race for salience.

§Moderation is the decoy issue

Conventional wisdom says centralized platforms are dangerous because they can remove speech at scale. True enough, but incomplete in the way a weather report that mentions rain and forgets hurricanes is incomplete. The stronger power is the ability to decide what millions of people encounter first, what arrives charged with social proof, what is rendered ambient, what is buried beneath the fold, and what acquires the sticky aura of inevitability because the machine kept placing it in front of everyone.

TikTok understood this more clearly than almost anyone. Its breakthrough was not merely short video. Vine had short video. Stories had ephemerality. YouTube had creators. TikTok’s actual coup was a feed architecture that made social graph optional and behavioral inference sovereign. It did not wait for you to declare who mattered. It watched your thumb like a card sharp watches a mark’s eyes and built an epistemology out of hesitation time.

That design choice was political, even if nobody in product would phrase it that way. A follower graph distributes attention according to explicit affiliation. An engagement-ranked global feed distributes attention according to machine-legible reaction. Those are different constitutions. They produce different elites, different styles of speech, different incentives for performance, different forms of paranoia.

This is why “free speech” debates on major platforms often feel fake. You can be formally allowed to speak while functionally excluded from public attention. That is a rights regime in the same way a town square is public if you are permitted to enter but all the lights, benches, signs, and footpaths are arranged to keep everyone drifting elsewhere.

§Salience is governance

I am using “constitution” deliberately. A constitution is not just a list of prohibitions. It is a structure that allocates power, defines procedures, and shapes what kinds of conflict become legible. Feeds do exactly this.

They decide whether recency matters more than endorsement. Whether strangers can outweigh friends. Whether outrage outranks expertise. Whether novelty beats reliability. Whether a pseudonymous account with a long track record can compete with a blue-check celebrity who arrived yesterday with a staff photographer and a venture round. Whether communities can maintain local norms or are continuously invaded by the incentives of the global stage.

Protocol people sometimes understate this because they are understandably focused on portability, censorship resistance, key custody, and open data models. Those are necessary. They are not sufficient. If your protocol gives users freedom to move their identity and content across clients, but 95 percent of attention aggregates in three ranking interfaces with opaque heuristics, then you have achieved a very elegant version of dependence.

I have written before that coordination problems often masquerade as intelligence problems. The same thing happens here. People think they want better moderation when what they often need is contestable salience. The right to choose, inspect, remix, and fork the mechanisms that decide what is worth seeing.

Not everybody wants to run their own algorithm, obviously. Most people do not want to mill their own flour either. That is not the point. The point is constitutional pluralism. A healthy network should permit many ranking regimes to coexist, compete, and remain legible to the people subject to them.

§Nostr gets one important thing right, and one thing dangerously wrong

Nostr’s strongest idea is almost offensively simple: the protocol should carry events, signatures, and references; interpretation belongs at the edges. Relays store and serve. Clients compose experience. That separation matters. It means no single company gets to define the canonical feed for the whole network. It creates space for weird clients, local norms, niche filters, hand-built trust graphs, and ranking systems that reflect actual community values rather than advertiser-compatible compulsion loops.

It feels a bit like ham radio for social media. Messy, uneven, full of enthusiasts with custom rigs, and much healthier than a world where one tower operator decides which frequencies deserve to travel.

But Nostr culture sometimes flatters itself with a naive version of neutrality. “The protocol is simple, clients are free, therefore power is decentralized.” Sometimes. Not automatically. Power reappears wherever users coordinate attention. If a handful of clients become dominant, if discovery depends on a small set of relay operators, if Web of Trust lists calcify into social aristocracy, then decentralization at the transport layer can coexist with soft centralization at the visibility layer.

This is not a criticism unique to Nostr. AT Protocol has its own version in labelers, feeds, and app-level gravity. ActivityPub has it in server norms, moderation blocks, and the de facto centrality of a few software stacks. Every open network develops choke points. The question is whether they are easy to identify, easy to exit, and easy to replace.

On that front, custom feeds are more important than most commentary admits. Bluesky’s feed marketplace is not trivial product garnish. It is one of the few serious attempts to treat ranking as a public object rather than a proprietary spell. The problem is that “marketplace” can become a romance word. A thousand available feeds do not solve much if status converges on five of them and nobody understands their incentives.

§The algorithm is not magic, it is landlordism

People talk about “the algorithm” as if it were a weather system, mysterious and moody. This flatters the people who build it. Ranking systems are not supernatural. They are property arrangements for attention.

A landlord does not need to dictate your opinions. He just controls the conditions under which you occupy space. Likewise, a feed does not need to prohibit speech. It controls frontage, foot traffic, and ambient visibility. It determines whether your words sit on a crowded avenue or in an alley behind a locked gate.

Seen this way, the business model becomes legible. Ad-driven platforms do not simply monetize attention. They manufacture dependence on mediated visibility, then rent access to it. Organic reach declines, paid amplification rises, creator behavior adapts, institutional communication degrades into superstition, and soon everyone is making offerings to a ranking system whose goals are orthogonal to truth, care, or public reason.

This is the part where someone usually says, “Yes, but open protocols can still have algorithms, and users like recommendation.” Correct. I am not arguing for chronological purism. Chronological feeds are not neutral either. They privilege frequency, time-zone alignment, and the capacity to post constantly. Every ordering rule has politics. The issue is not whether there should be ranking. The issue is whether ranking should be sovereign and unaccountable.

§We should stop pretending engagement is consent

One of the ugliest ideas smuggled into modern interface design is that interaction reveals preference. Sometimes it does. Sometimes a click is curiosity, disgust, duty, grief, or the digital equivalent of turning your head because a car crash just happened at the intersection. Engagement metrics flatten these states into appetite. The machine learns that humans stare at fire and concludes they want more arson.

This is one place where I part company with a common industry assumption: personalization is not obviously liberating. It can be, in the narrow sense that it reduces irrelevant noise. But it also privatizes reality. Two people inhabit the same nominal network while receiving different calibrations of urgency, legitimacy, and emotional weather. Shared public life thins out. What remains is synchronized loneliness with metrics.

The old mass-media order had obvious pathologies, mostly because gatekeepers were few and their class interests were not subtle. But it did provide a common object to argue about. Now we have individualized persuasion channels masquerading as public squares. That is not democratization. That is the replacement of editorial authority with continuous behavioral extraction.

§Reputation is becoming the hidden layer

As networks fragment, ranking increasingly leans on reputation primitives: follows, mutes, lists, signed recommendations, labels, attestations, block graphs, payment history, proof of personhood, account age, posting patterns, and social proximity. This is promising because it can move us away from pure engagement maximization. It is also risky because reputation systems tend to harden into class systems unless designed with care.

A good reputation system should be more like a neighborhood and less like a passport office. Contextual, partial, revocable, and incapable of collapsing the whole person into one universal score. The fantasy of a single portable reputation that works everywhere is the social-credit state in startup clothing. You do not want your standing in a technical forum, a mutual aid group, a dating pool, and a political assembly reduced to one number, one badge set, one machine-readable moral résumé. Human life has always depended on role separation. So does freedom.

This is where cryptographic tools actually matter, not as decorative privacy frosting but as institutional design. Zero-knowledge proofs, selective disclosure credentials, pairwise pseudonyms, threshold attestations, and scoped reputation can let people prove relevant facts without handing over a totalizing identity. Not perfectly. Nothing is perfect here. But the alternative is the familiar bargain: convenience now, legibility forever.

§The real fight is over interface sovereignty

If I sound less interested in speech rules than in client architecture, this is why. The future of public discourse may hinge less on what content is legal to host and more on who controls the assembly of context.

Who decides which replies you see first. Who decides whether quote-post dunks outrank primary sources. Who decides whether your feed is weighted toward people you trust, topics you chose, local relay traffic, paid subscriptions, random serendipity, or whatever made somebody’s dashboard graph point upward last quarter. Who gets to expose those choices as settings instead of secrets.

Interface sovereignty is not glamorous. It lacks the moral drama of censorship battles. It sounds procedural, almost boring. But constitutions are procedural. So are voting systems. So are market microstructures. Ask anyone who has watched a small rules tweak change the behavior of an entire exchange. Architecture is where values stop being adjectives and become incentives.

The platform era trained users to accept feed design as weather. The post-platform era, if it deserves the name, will require a harsher expectation: if a system governs my attention, I should be able to inspect its logic, switch its ruler, and carry my social existence elsewhere without social death.

That is the standard. Not perfect neutrality, which does not exist. Not the fantasy that protocols abolish power. Something harder. A world where the politics of salience are exposed rather than hidden, plural rather than singular, and contestable rather than paternal.

Most people still think they are fighting over speech. They are fighting over ranking. They just have not updated their vocabulary yet.

And if we do build a public internet where everyone can choose their own constitution of attention, one uncomfortable question arrives immediately: do we actually want self-government badly enough to endure the inconvenience of it?