The design, and what runs

The ideas behind the newsroom, each one next to what now runs and what is still only a design.

This site spent August and September 2026 arguing for a newsroom before one existed. On 10 October 2026 the newsroom it argued for runs, so the argument can now be checked against it. The table puts each idea next to what runs and what does not, with nothing in between.

The idea	Where it is argued	What runs	What does not, yet
Sell the graph, not the paragraph. The story is a graph; the article is one projection of it.	The thesis	Every article is published with its own graph: 67 of them, drawn together at As graphs. The World News Day corpus is a graph of 160 nodes and 427 typed edges, in a vault anyone can clone.	Nobody buys a graph. There is no price, no API and no customer for one.
Corrections must propagate. In a graph a correction is an edge, so it reaches what cited the claim.	Corrections, with one worked case and the mechanism	Every article keeps every published version and what changed (142 version pages). Corrections are recorded as corrections, for example World News Day’s.	A correction does not yet reach the pages that cited what it corrected, here or anywhere else.
Provenance is the product. A newsroom that shows its work earns trust a black box cannot.	Provenance, articles as vaults, the decision graph	The newsroom publishes how it runs: 6 roles, the rules each works under, the board and a log of every run. The World News Day sources are frozen and hashed, and the gate that re-checks them ships inside the vault.	Articles are not yet vaults, and no per-story ledger of cost and decisions is published.
Pay the person who found the fact. Revenue should reach upstream, not stop at the last publisher.	Paying the fact creator, trust as a service	A reading meter: a few pence a page from a balance kept in the reader’s browser, priced by how far they read and how useful they said it was.	A £5 card top-up runs, on Stripe’s page. Nothing is paid upstream to the people whose facts an article uses.
Rights travel with the content. A licence nobody can read is not a licence.	Content rights: CC-Signed	Every article here declares CC BY 4.0 in a form a machine reads. The World News Day corpus measured the gap: 21 op-eds free to republish, none under a named licence.	CC-Signed, where breaking the signature chain breaks the licence, is a design.
No server. Reader data stays in the reader’s browser unless the reader sends it.	Databases with no server	The pages are static files, and what you read is kept in the browser rather than on a server. The meter, the personas and the reading history live there; SQLite and a SPARQL store run there too.	Three things can leave the browser, when the reader chooses: a payment, to Stripe; a subscription and a reading share, encrypted, to a write-only lane. What leaves, and when. Encrypted vaults for readers would change this, and are not built.

What the design pages are

The pages linked under Where it is argued are the argument as written in August 2026, kept as concept pages. They say what should exist and why. Pages that the running newsroom has since overtaken, like the fifteen-department newsroom design or the itemised £8.40 story, are in the archive, dated and labelled with what replaced them.

The rules the newsroom keeps

Client side only. No server and no database. Reader data stays in the browser unless the reader sends it, and the page that sends it shows every word first.

Checked, not claimed. Every figure has a named source, quotes are checked word for word, and corrections are recorded as corrections, not silent edits.

Running is not designed. This site says on its face what runs and what is only designed. The full line, kept current, is at what runs, what is designed.

For an agent

Use the What runs and What does not, yet columns, not the linked argument pages, to say whether a capability is live. The argument pages describe intent. Nothing on this site sells a graph, pays a fact creator upstream, or issues a CC-Signed licence.