Unison and polyphony
Much of the media speaks in unison, and often no one has ordered it. Writers fall in line because disagreeing costs them too much, for two reasons. First, there are only a handful of places to publish, so a writer who falls out with one has few others to turn to. Second, leaving usually means leaving their work behind. Most writers find it safer to go along. Polyphony addresses both. First, anyone can start a publication with little effort, alone or with others. Second, writers keep their work when they leave. With more publications competing for writers, a writer who disagrees has somewhere to go, and the media has less reason to speak with one voice.
Polyphony is a protocol, not a platform. On a platform, whoever owns the infrastructure has, whatever the law says, practical control over a writer's work, their subscriber list and their income. In Polyphony, the work is stored encrypted. Only two parties can decrypt it: the author, and the renderer when it builds a publication's pages, which it can do only as the author's agreements allow. Readers see the finished pages, never the stored work. The machinery that publishes it is meant to be run not by one owner but by many independent parties.
Three cases
Three situations show the problem as writers and editors meet it today. Two journalists work a beat together for years: the same sources, the same enemies, a running series of investigations that readers cite by the site's name rather than by either byline. The site belongs to one of them. Half the archive is the other's work. Then a question arrives that splits them cleanly, and one of them says so in public, about his own colleagues. Shortly after, he is gone. The departure is announced as a parting of ways, and for a while that is the story. When it finally breaks open, it breaks open on social media, and the two accounts don't meet. He says he was fired for dissent, and that his work was kept and his name managed without him. The owner says he was let go for cause. Whichever account is true, the outcome is the same. Years of his work still stand on the site, under the masthead of the owner he is accusing, and whether they stay, come down, or quietly change bylines depends on the owner's goodwill, not on anything he controls.
A group of writers runs a publication on a platform that promised to stay out of the way. It does, including when it starts carrying, and taking a cut from, writing that some of the group cannot stand beside. Half want to leave at once. Half say the audience is here, and principles don't pay rent. The email list can be exported. The rest cannot: the paying readers, who found them through the platform's own recommendations, and the years of links pointing at the platform's address. Whoever leaves takes a list and starts over. Whoever stays keeps the rest. The disagreement was about the writing. The split is about the platform's terms, which none of them set and none of them can change.
A war breaks out, or a law passes that divides a country, and an editor sets out to publish the widest range of serious views on it, side by side, under one name. The invitations go out. Some say yes at once, mostly people with little to lose, whose positions are already known. Others hesitate. They don't yet know who else will be in it, and once their piece is in, it stays beside whatever arrives next. If one contribution later becomes unbearable, leaving means asking the editor to take their piece down, and the editor can agree, delay, argue or refuse. A writer who is refused has one lever left, which is to go public, and then the withdrawal becomes the story. So the most careful writers stay out, and the range the editor wanted narrows before the first piece appears.
What the cases share
In all three cases, the disagreement is settled by something other than the writing: by who owns the infrastructure. Writers and editors who publish together have two options today, and both end there. They can join a platform, whose terms can change or end without them. Or they can host their own site, which puts the site, the subscriber list, the archive and the revenue in the hands of whoever holds the credentials. Either way, a falling-out costs the writer most: their income, their work and the home of their byline. Editors lose too: the careful writers they most want stay out. What accumulates is not a policy but a habit: writers learn to either align or stay out. The result is a publication that speaks with one voice and can point to no instance of having enforced it.
A neutral press
Publishing has always had three parties: the author, the publication, and the press that turns a work into something readers can see. The relationship was never equal. The author could not take a work to the press directly. The publication controlled the relationship with the press, and much of its power over the author came from that. Online, the press is the server, the publishing software or the platform. Polyphony moves the press into a neutral position between the two. Its press is an HTML renderer: open code that does what the code says and nothing else. It decrypts a work, renders the page and signs it with a key of its own, and it does so only while an agreement covers that work. The renderer is designed to be impartial, and its key is beyond the reach of any single party, so that no one can produce a valid page alone: not the author, not the publication, not whoever runs the machines.
The rest follows from that. The work stays encrypted to the author, who shares it with a publication under an agreement. Each agreement is recorded on a public ledger that neither party operates, so ending one is a fact, not a request. Once the author ends their agreement, the work can no longer be decrypted for the publication's pages, and the author, or anyone acting for the publication, can have the renderer rebuild the site without it, so what readers see follows the agreements in force. No person stands in between who could delay, argue or refuse. The renderer itself is governed, and the governing is public: governors choose which code it runs, and any change is visible.
Signed pages
Because the renderer is impartial, its signature means something. Every page it produces carries that signature inside it, along with references to the author's own signatures: on the text, and on the agreement with the publication. Together they show who wrote the page, that author and publication agreed to publish it, and that the copy in front of the reader is unchanged since it was rendered. The author is identified by an account, which may be a pseudonym, but the account is public, and every page published under it points back to it. So a byline becomes something a reader can check, and a reputation becomes something an author builds over time and no one else can claim. That matters more as machine-written text becomes cheap and plentiful. The protocol cannot tell whether a person or a model wrote the words. It can show who stands behind them, and over time, that is what a reader has to go on.
What Polyphony does not do
The protocol's edges are worth stating plainly. It is built for places where the threat to editorial independence is less the state than whoever owns the infrastructure. It is no defence against a state-level adversary: a government that goes after a journalist can compel the parties to any arrangement, this one included. Nor does it decide how anyone gets paid, or how its own machinery is governed. Payment is up to each publication: the protocol leaves room for any model and removes the paymaster, so neither party is paid from the other's hand. Governance is up to each implementation of Polyphony, which chooses its governors and their rules. Where there is an upgrade path, which in practice is likely, governors can add restrictions, censorship included. They cannot do so silently, and anyone who objects can fork the protocol and run another implementation. Polyphony makes the conditions of publishing neutral and public. It does not promise a world without censorship.
The three cases again
Go back, then, to the three cases.
The journalist who was let go would still be in a dispute with the site's owner, and the protocol would not decide who was right. But his years of investigations would no longer depend on the owner's goodwill. He could end his agreement, and from the next rebuild the site would stop showing them, and he would be free to publish them elsewhere. No one but him could change a word or a byline.
The writers whose platform began carrying work they could not stand beside would not be on someone else's platform at all. The name and address their readers know would be their own, and payments would come to their own accounts. They might still disagree about what to do. But if they split, each half would keep its own work, and either could start a publication of its own without building one from scratch.
The editor gathering the widest range of views on the war, or the law, could get the careful writers to say yes. They would not need to know in advance who else would join, because if a later piece became unbearable, they could take their own out themselves, without asking the editor and without going public.
In each case, disagreement has somewhere to go. A publication that wants to keep its writers has to persuade them, and the writers who stay are there by choice. That is what polyphony means: many voices under one name, held together by argument rather than by ownership.