Northwind decides its Christmas range in October. Four coffees, a gift set, a landing page, changes to the front page and the catalogue copy.
Not publishing it is free — write it, leave it as drafts, publish in December. That much needs nothing from this part.
What that does not give you is a way to look at it. An unpublished draft can be previewed a page at a time, which answers "is this page right" and never answers the question that actually matters in October: does the site hang together with the range on it? Does the catalogue read sensibly with four more coffees in it? Does the front page still make sense? Can somebody who is not the author click around it for ten minutes and find what is wrong?
That needs a whole site, standing up, that is not the public one.
A second environment
So there is a second authoring environment — Staging — with a delivery environment of its own. The Christmas range is promoted into it and published there, against a delivery key that is not the public site’s key.
The result is a complete, browsable, private copy of Northwind as it will be in December, at a URL you can send to somebody. The public site is untouched and has never heard of any of it.

The whole pipeline on one canvas: a solid line publishes, a dashed one promotes, and Staging reaches the public site through neither.
The one step that needs Pro
This is the second and last time this series mentions a plan, and it is the honest one.
Starter allows exactly one authoring and one delivery environment. So this part — and only this part — genuinely needs Pro.
What is worth being precise about is what is gated, because it is not what people usually assume:
No Content capability in this series is restricted by tier. Localization, personalization, contextual values, workflow, Adapters, streams — every one of them runs identically on Starter. What a plan buys is scale and environments.
There is no feature flag on any of the last five parts. Northwind on Starter is a smaller site with the same abilities. Northwind on Pro is a bigger site that can also stand up a second copy of itself.
If you are on Starter, read this part rather than following it. Nothing later depends on it.
Everything that talks to Content takes the environment as an argument, so pointing the same tooling at staging is a flag rather than a second copy of anything:
node content-model/apply.mjs \
--base https://api.ebitex.io \
--cookie-file ./session-cookie.txt \
--environment <authoring environment id> \
--delivery-environment <delivery environment id>
Content flows downstream
This part took a wrong turn worth inheriting, because the wrong design is the intuitive one.
The first attempt had Staging inherit the library from the authoring environment and promote back into it — stage the work downstream, approve it, push it up to production. That is how a lot of deployment pipelines are shaped, so it feels right.
It does not work, and the reason is structural. Environment inheritance covers the library — Contracts, Templates, Components — but not the Experience tree. A child environment has no site and no ancestor nodes, so there is nowhere for a page to land: promotion needs an aligned parent at the target, and there is not one.
The shape that works is the other way round:
Author in one place, and promote outwards. Content flows downstream, away from where it is written.
Promotion preserves identity, and the target’s copies come from the source. Which also rules out the other tempting shortcut: do not build the second environment by applying your content model to it a second time. That produces two different coffee Contracts with two different identities and nothing anywhere that can reconcile them. One authored copy, promoted.
The root goes first
One thing to get right, and it is worth knowing what the API is doing about it.
Ask for a promotion plan and you get an ordered list of everything that would move — schema first, then content. The obvious next move is to echo that list straight back to the execute endpoint.
Execute will not take it on its own. A promotion is rooted at one thing, and the closure is re-planned server-side from that root, so the request has to name it — root, beside the items — or it is refused with promotion_root_required and nothing is written. Name the page you are promoting. The items are the include list, and their order carries no meaning at all.
That reads like ceremony until you notice what it refuses to guess. The root used to be taken from the first item, positionally, with nothing in the request shape saying so — which made echoing a plan back promote a Contract, one item, successfully and silently. An ambiguous request is now answered instead of interpreted, which is the trade worth making every time: a named error costs you a minute, and a plausible guess costs you the afternoon you spend wondering where the page went.
Where this leaves us
A Christmas range nobody can see, standing up on a site somebody can walk around.
One part left with a mechanism in it. Next: a cross-sell rail that shows coffees as cards, without a card field anywhere on coffee and without copying a word.
