The review gate

Yesterday you made something that could genuinely go live. Today is the skill that decides whether it does. Anyone can produce — that stopped being scarce two years ago. Knowing precisely what is wrong with a piece of work before a customer sees it did not, and it is the most hireable thing in this program.

1. Editing and reviewing are two different jobs.

They get treated as one thing, and confusing them is why most "reviews" produce a slightly nicer version of the same problem.

	Asks	Produces
Editing	How do I make these words better?	A cleaner draft
Reviewing	Should this be published at all?	A decision

A piece can be beautifully edited and still not be shippable, because the claim at the centre of it cannot be backed. Tidying the sentence does not fix that — it makes the unsupportable claim read more convincingly.

Review comes first. If the answer is "not ready", editing it was wasted work.

2. The author is the worst-placed person to do it, and the reason is mechanical.

You already met this twice. Day 4: the Fresh-Session Review works because the model has no memory of writing the thing. Day 6: an agent auditing its own work catches less, because the whole build is sitting in its context.

It is the same limitation in people, and it is not a character flaw:

When you read your own draft, you read it with everything you know about the product loaded. The reader has none of it. You are physically unable to experience your own sentence the way a stranger will — you can only simulate it, badly.

Which is why "I'll read it again in the morning" is a weak substitute for someone else reading it, and why a founder reviewing their own launch post is not a review.

3. A rubber stamp is worse than no review at all.

This sounds like an exaggeration. It is not.

An unreviewed piece is known to be unreviewed. Someone might still catch the problem. But a piece marked reviewed by someone who skimmed it now carries a record saying it was checked — so nobody looks again, and the mistake travels further with a note of approval attached to it.

You saw the same shape on Day 4: a human in the loop is worth nothing if they skim and click approve. Real review means you can say what you checked and what you would have caught.

So the honest options when you cannot review something properly — no time, no access to the evidence, you wrote it yourself — are to say so, or to hand it to someone else. Not to pass it.

4. What a review checks, in order — and why wording comes last.

Order matters, because reviewers naturally start with what is easiest to see and run out of attention before reaching what matters.

Voice and correctness go last on purpose. They are the most visible and the least important. A reviewer who fixes six typos and misses an unprovable claim has done the job backwards — and will feel productive doing it.

5. Is the central claim provable, and with what?

The first check, and the one that does most of the work. "With what" is the part people skip.

The second contains its own evidence. Anyone can check it, and if it is wrong someone will say so — which is exactly why it is believable.

This is yesterday's claim inventory doing its job. Every claim marked assumption or unsupported is a review finding waiting to happen, and the three honest responses are the same as Herald's: find the evidence, weaken it to what is true, or cut it.

6. Name the weakest part. Every piece has one.

A review that says "this is good" is not a review. Something in there is weaker than the rest — that is what "weakest" means.

If you cannot find it, you have not read carefully enough.

And be specific enough to act on. "The second paragraph overstates it" is useless. Quote the sentence, say why, and suggest the fix.

Two habits that make this land:

7. Three parts to any review, and an ambiguous one reads as approval.

Every review delivers three things:

A verdict	Ready · ready with the changes below · not ready. Never ambiguous.
Findings, ranked	Most serious first. Each one: what the problem is, why it matters, and a concrete fix.
What you could not check	If you had no access to the evidence behind a claim, say so rather than passing it.

Two rules attached to those.

An unclear review gets read as approval. The author wants to ship. A review that lists some thoughts without a verdict will be treated as a yes, and you will both be surprised later.

A finding without a fix is a complaint. Naming the problem is half. The other half is what to do instead, in a line or two.

And the review has to survive — attached to the work, not said in a chat that scrolls away. The value is the record of what was challenged and what changed before it shipped.

8. What a reviewer must not do.

Two boundaries, and both exist for the same reason.

A reviewer does not rewrite the piece. Suggest the fix and hand it back. The moment you rewrite it, you are the author — and there is no longer a reviewer. This is the most common way review quietly stops happening on small teams: the reviewer is faster at fixing than explaining, so they fix it, and nobody independent has now read the final version.

A reviewer does not decide strategy. If the positioning itself looks wrong, that is a finding to raise with the team, not an edit to make. Changing the message during a copy review means the decision got made by whoever happened to read it last.

That is not the same as staying quiet. Say plainly when the audience, the positioning or the objective looks wrong — and say it as a finding for the team rather than fixing it yourself. "This is well written and I think it is aimed at the wrong person" is one of the most valuable things a reviewer can return.

Editor enforces both, and one more that you should hold yourself to: it never reviews its own work. If it wrote or substantially rewrote something, it refuses the review and asks for a different reader.

Which is why yesterday's page went through Herald but not Editor. Herald wrote it. Today Editor reads it cold — and that is the entire point.

9. What this lesson did not cover, and how to go deeper.

Covered: editing versus reviewing, why authors cannot review themselves, the rubber-stamp problem, the six checks in order, provable claims, naming the weakest part, the three-part review, and the two boundaries.

Not covered:

Where to go. The fastest way to get good at this is to review other people's work and have yours reviewed. Offer to read something a batchmate posts in #daily-log and give them a real verdict with ranked findings. You will learn more from one uncomfortable review than from this page.

Tomorrow — Day 12: Content that earns attention. You have a page and a standard. Next is how someone finds their way to it in the first place — and why content only compounds when it contains something a competitor could not have written.