Edition
News, but calm.
The synchronization layer between reality and understanding.
The news didn't break.
Its delivery system did.
If you stopped paying attention, you didn't fail the news. The structures around it failed you. Reporting is still being done well by people who care about it — but it reaches you through a machine built on three things that work against you: publishing that thinks in documents, economics that pay for attention rather than understanding, and distribution with no concept of completion or significance. Headlines withhold because withholding is what the machine rewards.
Walking away was rational. But a world where careful people opt out of knowing — while the loudest, least careful sources fill the silence — is one everyone pays for. What's at stake isn't an industry, and it isn't journalists. It's whether staying truthfully informed stays possible for a person with a life.
You're not behind.
Your news is.
Here's the tax nobody names: every article re-explains the world from scratch, because no publication knows what you already saw. You re-read four paragraphs of background to find the one sentence that's new. Multiply that by every story, every day. "I can't keep up" isn't a personal failing — it's the compound interest on that tax.
And the feed never ends. There's no moment where it says you're caught up — go live your life. A product with no bottom never intended for you to finish. Calm isn't the absence of news. Calm is news with an edge you can reach.
News is a function,
not an object.
Staying informed should mean keeping your understanding synchronized with meaningful changes in reality, at the lowest reasonable cost to your attention. Written down, that's not a philosophy. It's a subtraction.
relevant(world_state(now)),
reader_state
)
Maintain two models: what is known about the world, and what a particular reader has already seen. The news is the meaningful delta between them. Everything else — articles, apps, briefings, alerts — is a way of delivering that difference. That's why subtraction is the product, and why no chatbot bolted onto a feed can fake it: it's an operation between two state models, and almost nobody maintains either one.
Reality — what Edition has reason to believe · Reader — what you've already been shown · Delta — the news · Projection — an edition, a briefing, a card, or nothing at all.
Five questions,
collapsed into one alert.
Every item of news raises five separate questions. Today's products answer them all at once, with the same red badge, at the same volume. That collapse is most of what makes news feel unbearable.
| Question | What it actually asks |
|---|---|
| Truth | How well established is this? Reported, corroborated, or confirmed? |
| Novelty | What materially changed since you last knew? |
| Significance | How consequential is that change — to anyone at all? |
| Relevance | How much does it pertain to you in particular? |
| Urgency | Do you need to know now, or is Thursday fine? |
These come apart constantly. A thing can be true, new, and relevant to you but trivial. It can be true and enormously significant without touching your life at all — and still belong in your edition, because civic breadth is worth something. It can be all four and simply not be urgent.
Calm isn't typography, or fewer words, or a nicer tone. Calm is information with its hierarchy restored.
A system that models these as five distinct properties can structurally refuse to shout. It can hold something for the morning. It can include what matters to the world and mark it as such. It can say this is real, but small. No product built on a single engagement score can make those distinctions, because it has thrown away the axes they live on.
We know it isn't fantasy.
We accidentally
built a version once.
Circa News rebuilt the article from first principles: stories became stacks of atoms — single verified facts, quotes, and events, each a complete thought carrying its sources. You didn't just read a story; you followed it. When something changed, Circa showed you only the new atoms. Never the rerun. It was a crude, hand-operated version of the subtraction above — and it worked.
A newsroom of about twelve ran it, staffed with editors who'd come from Reuters and the Wall Street Journal. The press called the format a breakthrough. Users called it their primary news app, unprompted, for years. The industry's memory of it — case studies, pattern libraries, citations that still recur a decade later — outlasted the company by a wide margin.
Twelve people, an hour faster than the Times
When the first US Ebola case broke, Circa had been quietly keeping the story for months. The newsroom wrote one atom, placed it on top, and hit send. Followers got a pure update; newcomers got a complete, verified story. A WIRED reporter was in the room with a stopwatch.
WASH. POST 2:07 PM
N.Y. TIMES 2:57 PM · +70 minutes behind
Boston Marathon bombing. Zero retractions.
Through the most error-riddled breaking story of the era — while major outlets misidentified suspects on live air — Circa published without a single retraction. Not because it went slow, but because its rules were structural: nothing entered a story unverified; the unconfirmed lived in clearly-marked editor's notes, separated from the known.
Then it died — and the argument didn't finish
Circa shut down in June 2015. On the cause, every post-mortem and both founders agree: the business failed. Twelve salaries were the cost floor, and the model that fit the product — loyal readers paying for calm — barely existed as a category yet.
What's genuinely unsettled is whether the format had a ceiling. The most serious scholarly treatment of Circa raises exactly that: atomized cards may have asked more of readers than most were willing to give, and may have stripped away narrative pleasure along with the redundancy. That critique has never been resolved, and this project doesn't need to resolve it — because atoms aren't the interface. The whole architecture that follows exists to separate what's known from how it's told. If cards were the limitation, then the honest response is a system where the cards were only ever one rendering among many.
Circa is provenance here, not premise. The idea stands or falls on the equation above. Circa is just the reason to believe a small team can actually run one.
The conditions that made Circa
impossible have moved.
Not "solved" — none of these is finished, and two of them are hypotheses this project exists to test. But all three have changed enough that the question is worth reopening.
Then: twelve salaries capped coverage; a three-follower story couldn't earn its keep.
Now: agents could make the one-follower story affordable — if they can maintain the state at the quality bar. That's the first thing to test, not to assert.
Then: 2015 had no consumer-subscription culture; calm was unfundable and ads demanded bait.
Now: paying for software that respects you is ordinary. Whether enough people would pay for this is still unknown.
Then: "just trust our newsroom" was the only answer available.
Now: with machine text indistinguishable from reporting, provable process is worth more than assurance. Trust isn't solved — but there's finally a way to earn it that isn't reputation alone.
Editions,
not feeds.
Your news arrives as an edition: finite, finishable, with a beginning and — radically — an end. When you're done, it says so and stops. Badge to zero. And some mornings, the honest edition is the one below. No manufactured urgency. The system asks one question before it asks for your attention: do you actually need anything from us today?
Nothing important on your follows has materially changed since yesterday. Go live your life. We're watching.
✓ CAUGHT UP · 0 ITEMS · READ TIME 0:00
It's more than
a nice idea.
We've written the whole thing down — a constitution its agents can't override, a rulebook for the hard cases, honest explainers of the machine underneath, and an inventory of the ways it might be wrong. Tap around. Every underlined term° opens a card. Every card is a node. Every related term is an edge you can follow. The site is dogfooding the product.
The Constitution
What Edition believes and refuses to break — from "append-only, version everything" to "reality is shared; presentation is personal."
Governance
Agents run the newsroom; a constitution runs the agents. Three rings, an intervention counter, an advocate for the reader, and the risk math that scales scrutiny.
The Protocol
A numbered rulebook for the matters that draw heat no matter how carefully you cover them. Judgment calls cited like case law.
Why a Graph
Six ingredients, and why each one keeps a promise an article can't. Honest about the technicality — clear about why it matters.
The Architecture
Two states, one delta, many faces. Why the presentation layer is walled off from the truth — and what the reader model does and doesn't know.
Where It Breaks
Six ways this could be wrong, written down before anyone else has to point them out. The most useful page here.
I'm not trying to
convince you
this works.
I'm trying to find out where it breaks.
This isn't a pitch and it isn't recruiting. Every claim here is either a receipt from the historical record or a hypothesis with a way to be proven wrong — and those two are labeled differently on purpose. The most valuable response isn't "cool idea." It's "that assumption is doing more work than you think," or "I lived through that part and it wasn't like that."
If you worked on the original: your memory is a primary source, and it outranks my reconstruction wherever the two disagree. Start with where it breaks →