A document goes wrong in its order.

By eleven the night before it is due, the deck has 40 finished slides, aligned and on brand, built with real care. Someone senior reads it end to end for the first time and says: I don't think this is the story.

Nothing is wrong with the slides, and every sentence in them is competent. The problem sits above them, not in them, in an argument nobody actually agreed to before the deck got built.

I have watched this happen to press releases, board memos, white papers, my own essays, and once to a wedding speech, rewritten past midnight because the draft from a week earlier turned out to be answering the wrong question. It is almost never a writing problem. It is an ordering problem.

The work got done back to front. The cheap decisions, the ones that cost 30 seconds to get right, were made last, and the expensive ones went first.

The research and the finished pages are made out of the claim and the headers. Revise the claim and everything built on top of it stops making sense, because it was never really about the words on the page.

Change your claim while it is still one sentence and the fix costs 30 seconds. Change it after you have written the headers and you are redoing an outline. Change it after the research and you are binning the research. Change it after the pages are finished, and that is the eleven o'clock rewrite: you are not editing a sentence, you are throwing away the work that sentence was quietly holding up.

The flaw is identical at every stage. Only the moment you catch it changes what it costs to fix.

The four-level shape is a borrowing, though not the authority behind it. Simon Brown's C4 model diagrams a software system at four levels, system, container, component and code, an abstraction-first approach to diagramming rather than a claim about the order design decisions get made in.

In that model a "container" is not a Docker container. It means the separately running parts, a mobile app, a website, a database, worth saying plainly since the word invites the narrower, more common reading.

The ordering idea itself is not new, and the closer prior art predates C4 by decades: Barbara Minto developed the Pyramid Principle at McKinsey in the 1970s, arguing for the same top-down order in business writing, out of consulting rather than engineering.

Those four moments, the claim, the headers, the research, the finished pages, have names. This framework calls them D1 through D4, and each one has one job and one test, and you go down a level only once the level above survives its test.

Some people genuinely write to find out what they think, and that is the standard objection to working this way. Fair, but D1 is a decision you commit to, not a discovery you wait for. If you have no claim yet, writing an ugly, throwaway page to find one is itself a D1 activity, not an exception to it. The framework says polish comes last. It does not say never write to think.

The honest admission here is that D3 and D4 are the softest boundary of the four, and they collapse into each other wherever a piece has to make a prediction rather than simply state what happened.

A house makes the same four levels concrete.

Easier shown than described, so here is this article at each of its own levels. Step through them.

You do not need software for this. Indented bullets are the whole apparatus, and depth on the page maps to depth in the argument, so a structural problem becomes something you can see rather than something you have to feel.

The pricing memo above has the tell in it already. One section carries 4 reasons, the other carries 1, and that gap is invisible in flowing prose and obvious the moment it is written as an outline. A D3 point that will not sit under either header is the other kind of tell: a gap in the D2, not the D3.

This is also where redrafting is cheap, because it happens before the expensive levels. You can sketch 4 different D2 structures in the time it takes to write one good paragraph, and pick the strongest before committing to any of them.

The framework earns most of its keep once a document has more than one author. Agree the claim and the headers before anyone drafts a sentence, or expect the eleven o'clock rewrite.

It also lets people come in at the level that is theirs. A subject expert can check the D3 without reading a word of prose. A designer can start on the D4 once the D2 is locked, instead of waiting for a finished draft to reverse-engineer the structure out of.

Polish has stopped being evidence of thinking. A weak argument now arrives beautifully dressed, with confident transitions, because a model will happily supply all three.

Language models are very good at D3 and D4. They will draft your supporting sections, give you 5 versions of a paragraph, and polish the rhythm until it sings. They are much weaker at D1, and not because they cannot form a sentence that sounds like a thesis. D1 is a decision about what you are willing to claim, built out of what you have seen and who you are trying to move, and that decision was never theirs to make.

Your D1 and D2 are the prompt. Hand a model a topic and you get something generic, because you gave away the only decision it could not make for you. Hand it a claim and a set of headers instead, and it gives you back something that sounds like whoever meant it, because in the part that matters, you did.

That cheapness creates a contradiction worth naming, because cheap drafting looks like it should flatten the curve from earlier, and it does not. The 40 slides can be regenerated in 20 minutes. What a model cannot cheapen is the senior person's read of them, or the credibility spent when that read goes badly. Cheap D3 and D4 do not lower the cost of being wrong. They move the expense from hours of drafting to a much smaller stretch of attention, spent by whoever is trusted to catch the claim before it ships.

Everything below the claim and the headers is craft, and craft is recoverable: a paragraph can be redrafted in a minute. What cannot be redrafted is the meeting where everyone found out, at the last possible moment, that the deck was arguing the wrong thing.

Which is where the reviewer's job actually moves. The most useful question in any document review is what level are we reviewing at. If the answer is D1 and someone is fixing commas, the meeting has already gone wrong.