Attune · the architecture we use

The Pre-Publish Gate

Five questions before anything goes out.

Takes two minutes. Catches the kind of mistake that's embarrassing precisely because nobody disagreed with the rule.

Why this exists

Every organization has rules about what goes out. Nearly none of them are checked by anything.

The rule sits in a style guide, or a brand document, or somebody's head. Everyone agrees with it. And then a piece goes out missing it — not because anyone decided to skip it, but because at the moment of publishing, nothing asked.

A rule that lives in a document and runs nowhere will rot.

That's the whole diagnosis. The fix isn't a better document or more discipline. It's a checkpoint at the one moment that matters: just before the thing becomes public and expensive to take back.

Five questions. Every one of them below caught a real error in our own publishing — the examples are ours, and they're recent.

The five

One

Has this already gone out?

Catches: the accidental repeat.

Check the live archive — not your drafts folder, not your calendar. The public record. Compare the title, the URL, and the opening line, because the same piece travels under different names as it moves between systems.

What it caughtA draft in one folder, a scheduled row in a calendar, and an already-published post turned out to be one article under three different titles. It was one click from being re-sent to the whole list six weeks after it first went out. Nothing in the workflow compared them, because nothing was looking at the live archive.
Two

Does it point at something that still exists?

Catches: the retired name, the dead link, the old price.

Every product name, offer, link and price in the piece, checked against whatever holds your current facts. Especially the call to action — it's the last thing written and the first thing that goes stale.

What it caughtThree of our four most recent published pieces were sending readers to products that no longer existed under those names. Not one of them named the course we actually sell. Every author knew the current name; the pieces were just written before it changed, and nothing re-read them after.
Three

Can a reader check every name?

Catches: the citation nobody can follow.

Every researcher, author, company or source named in the piece is linked to a primary source — not a summary of it, not a news write-up about it. If you can't find the primary source, that's information: you're less sure of this than the sentence sounds.

What it caughtA published piece named a researcher and a paper in plain text, with no links. One of those papers had been flagged in our own notes as needing verification before publishing. It went out anyway. Unverified and unlinkable is the worst pair — no reader can check it, and neither can you.
Four

Can you check every number?

Catches: the statistic that drifted.

Every figure traced to a source you've actually read, quoted at the precision that source supports. Keep them in one file so the check is a lookup rather than a research project. And prefer the specific verified number to a vague gesture at it — "50% more than humans" is the same length as "far more than humans" and carries ten times the weight.

What it caughtA draft softened a precise finding into a vague one, losing the strongest fact in the piece. The exact figure was sitting in our verified-sources file the whole time.
Five

Does it carry the line you always carry?

Catches: the standing rule everyone agrees with and nobody applies.

Every organization has one — a disclosure, a credit, a required attribution, a legal line. Ours is "edited and researched with agentic AI help; the opinions and judgments are ours." Yours might be different. The question is whether anything actually verifies it before publishing, or whether it depends on someone remembering.

What it caughtThree published pieces missing it. Nobody had objected to the rule. It simply lived in a facts document and ran nowhere, so it was applied when someone happened to think of it.
And the rule that makes the gate real

If a check fails, say so and stop. Do not publish with a caveat attached, and do not hand it to someone else with a note about the problem. That's how the problem ships. A gate that can be waived under time pressure is a suggestion, and it will be waived exactly when it matters — because the moments you're most rushed are the moments you're most likely to be wrong.

Where to put it

Not in a document. Documents are where rules go to rot — that's the failure this whole page is about.

Put it where publishing happens. If a person publishes, it goes in the publishing checklist they actually open. If an AI assistant drafts for you, it goes in that assistant's instructions, phrased so the assistant knows the checking is its job:

Before you hand me anything to publish, run these five checks and report the
result of each one. You are the check — nothing else is looking.

1. HAS THIS ALREADY GONE OUT?
   Compare the title, the URL, and the opening line against the live public
   archive — not my drafts, not my calendar. The same piece travels under
   different names.

2. DOES IT POINT AT SOMETHING THAT STILL EXISTS?
   Every product name, offer, link and price checked against [WHERE YOUR
   CURRENT FACTS LIVE]. Check the call to action first — it goes stale first.

3. CAN A READER CHECK EVERY NAME?
   Every person, paper, company or source named is linked to a primary
   source. If there's no link, say so rather than publishing it bare.

4. CAN I CHECK EVERY NUMBER?
   Every figure traced to [YOUR VERIFIED SOURCES FILE], at the precision the
   source supports. If a figure isn't there, verify it and add it — don't
   soften it into a vague phrase to avoid the check.

5. DOES IT CARRY [YOUR STANDING LINE]?
   Last line, every time.

If any check fails, tell me which one and stop. Do not hand me a draft with a
known problem and a note about it — that is how the problem gets published.
The test of whether it's working: the gate should fail sometimes. If it has passed every single time for a month, it isn't checking — it's decorating. Go and look at something you published recently and see whether it would really have passed.

What this is actually teaching

The checks are useful, but they're not the point. The point is the shape.

Most people trying to be careful write more rules. The rules accumulate, nobody can hold them all, and the ones that matter get lost among the ones that don't. Then something goes wrong and the response is to write another rule.

Put each rule where it runs, not where it's remembered.

Facts belong somewhere one person owns and dates. Checks belong inside the thing that does the work. Patterns belong in whatever you read before you start. A rule with no runtime is a wish — and you find out which of your rules are wishes the first time one of them quietly fails in public.