← the whole review · The twelve shelves · The names · The labels not on your list · Retail & Commerce · Candidates from your old lists · Brand & Identity · The open calls

The mechanism

What sits above the label on each piece — the one word that says what kind of thing it is — and the eight proposals that follow from it

P-03The family nothing was ever approved under · full

You never approved a single kind of piece under the family called “Social-native & community”, yet it sits over the two biggest groups of pieces we hold. Do we name the two things that really are born social, and let the family go?

Every piece we save carries one word that says what kind of thing it is — a brand film, a case study, a press release. Call that its label. You approved the 64 labels name by name, and nothing on this page touches them. Above the labels sit ten family names, and one of them is the odd one out: under “Social-native & community” you approved nothing at all.

Empty on your list, and the busiest thing in the database. That family sits over the two labels stored as social post and social clip — 2,518 pieces and 943 pieces, 3,461 between them, more than any other group we hold. The reason it filled up is a mistake nobody made on purpose: “social” was written down as what a piece is, when it is really where the piece ran.

Look at what is actually under there. One is a written announcement — “GPT-5.5 is here! Have been daily driving this model…” — which reads as an announcement whether you meet it on LinkedIn or anywhere else. Another is a clip whose caption reads “This video was generated by our text-to-video model, Sora, without modification.”, which is a demonstration of a model that happened to run on TikTok. The third piece below is the same shape with the problem already solved: “Our open models are here…” ran on TikTok exactly like the second one, and what is written on it is product demo — a real kind of piece — rather than the name of the room it ran in. That is all any of these needed.

So the question narrows to something better: is anything genuinely born social — a thing that would not exist if the platform did not? Two answers hold up, and each of them needs a name you can actually write on a piece. Call the first a creator collaboration: a piece the brand made with a creator, where the creator's own voice is the piece rather than the wrapper around it. Call the second a meme format: a piece whose whole point is a format that is already going round, borrowed and refitted. Hide the platform from an editor and they will still say “that is a meme”. That is exactly the test every one of your 64 labels had to pass, and these two are the only things under this family that pass it.

Nothing here blocks anything. Those 3,461 pieces are already on their way to being split — “social” moves to the where-it-ran fields and each piece takes the label for what it actually is — and that happens whichever way you mark this one. What you are deciding is only whether two new names join the list, and whether the family name goes with them.

My take: Name the two — creator collaboration and meme format — and let the family name go. A group name you never chose, sitting over more pieces than any other group in the library, is the one thing here that is actively misleading.

GPT-5.5 is here! Have been daily driving this model for a few weeks and it's speed and capability to handle complex work is really impressive. Codex has quickly become the go-to coding agent for engin
text only — no image on this piece

What this call does to everything else written on these pieces

Nothing else written on these pieces moves. The brand each one belongs to, the crafts read out of it, the products it shows, the pile it belongs to (a pile is any named collection a piece belongs to — a campaign like Customer Stories, or the launch of a product) — every one of those lives in its own place and none of them mentions this family, so none of them has to be touched. The one knock-on is forward-looking: two new names mean two more choices offered to the software that reads new pieces as they arrive, and the sorting step that is currently switched off gains two more real names to sort by instead of the mixed old list it still points at.

The alternative, and its cost: Keep the family and approve nothing under it, which is where things stand today. The cost is not theoretical: the largest group of pieces in the library keeps a name you never chose, and the two things that really are born social have nowhere to be named, so a piece made with a creator and a piece built on a meme format both keep arriving as social posts — the label we have already agreed does not say what a piece is.

Left off this page on purpose: “Generated by Sora, without modification” — the old card shortened this clip's caption; the page now quotes it in full (“This video was generated by our text-to-video model, Sora, without modification.”), so nothing is lost — the checker keys on the exact shortened string

Where this was worked out: proposal 3 on the reconciliation page, with every count on this decision read from the database on the day the page was built

P-04Which five worlds you look through · full

In July you typed one set of five worlds; in August you dictated a different set. Which one is the set the product looks through?

A world here is a saved view. Open the Campaigns world and you get the pieces that belong to a campaign; open Content and you get the ones that belong to no campaign at all. Nothing about a world is ever written on a piece — the product works it out from facts it already holds. That is what makes this pick cheap: you are choosing how to look, not what to store.

You typed the five on 9 July, describing them as the highest level of creative services: Brand Identity & Systems, Campaigns, Content, Digital Products & Experiences, Physical Products & Experiences. At the walk in August you said them again and they came out differently — “brand and identity, campaigns and content, events and experience, digital…, physical…”. Two real changes hide in that sentence: Campaigns and Content became one world, and events became a world of their own.

Here is why I would keep the July five. Content is your always-on world precisely because those pieces belong to no campaign. Take the Nike film “Get Ada Her Way…”, the first of the three pieces below: it belongs to no campaign we hold, and it is not alone — 13,626 pieces in the library belong to none, against 312 that do. The Ramp case study next to it is the other side of that same cut: it belongs to the Customer Stories pile, so the Campaigns view picks it up and the Content view leaves it alone. Merge the two worlds and that always-on body of work folds into a word that means the opposite of always-on, and the one cut those two pieces sit either side of is gone.

Events are the other half. The embroidered DevDay patches, the third piece below, belong to a pile named for that event, and the pile behaves the way a campaign behaves: it has a name, a moment in time, and pieces that belong to it. Events already work as a container that other things attach to. Making events a sixth world would ask every piece to answer a question its container has already answered, and the two answers would then have to agree with each other forever.

That said, this is your ground and you are entitled to move it. If your instinct says events deserve a world of their own, say so plainly and it becomes the ruling — and because these are views rather than stored facts, nothing has to be rebuilt to make it true.

My take: Keep the July five, and leave events as a container. Content earns its place as a separate world only because it is defined by the absence of a campaign; merge the two and the definition eats itself.

At Ramp, engineers are using Codex with GPT‑5.5 to accelerate code review and develop internal agentic tooling, helping teams get substantive pull request feedback in minutes instead of hours. Thanks to its reasoning capabilities, Codex with GPT‑5.5 is uniquely able to reduce the amount of manual, hands-on work they’d otherwise have to do.
text only — no image on this piece

What this call does to everything else written on these pieces

This call touches nothing that is written on a piece, because no world ever is. What it does touch is the one rule you asked for: a piece's world reading and the containers it belongs to must never contradict each other. Keep the July five and that rule has one clear job — check that a piece the Campaigns view picks up really does belong to a campaign, and that a piece in Content really belongs to none. Merge Campaigns and Content and the rule loses its meaning, because the merged world would take both the pieces that belong to a campaign and the ones that do not, which is every piece there is.

The alternative, and its cost: Take the August set: Campaigns and Content as one world, events as a sixth. The cost is two-sided. You lose the always-on cut — the 13,626 pieces that belong to no campaign stop being a group you can open on its own. And every piece attached to an event has to answer the same question twice, once through its container and once through its world, with no rule for which answer wins when they drift apart.

Where this was worked out: proposal 4 on the reconciliation page, where your July typing and your August dictation are quoted side by side

P-05The two names you parked · full

Two of the ten family names had a naming question parked against them. Does confirming the mechanism close it, or do you still want them named?

This one is on the page so that nothing disappears quietly. On 10 August you went through the ten family names and set two of them aside rather than ruling on them: “Visual & identity assets”, which you said sat right up against brand identity, and “Event, retail & interactive experiences”, which you said spanned physical and digital at the same time. Both were parked, and parked things go missing.

Under the mechanism you leaned toward, that question does not get answered — it dissolves. The family layer stops being a place where anything is written down and becomes browsing furniture: shelves. A shelf is just a group you browse the list of names by. The members of those two families then regroup under four names you chose yourself: Brand & Identity, Events & Experiential, Retail & Commerce, and Interactive & Digital Experience.

The two pieces below are the plain version of that. The embroidered character patches from DevDay and the photograph of the venue atrium were stuck inside the same family before, because both came out of the same event. Under your names they separate the way a person would separate them: the patches are a piece of the brand's identity, the atrium photograph is a record of the event. 13 pieces carry the first label today and 5 carry the second — small numbers, and the split is still obvious to the eye.

So there is nothing separate to build and nothing separate to answer here. If you would rather keep the two family names and finish naming them properly, that is a genuine choice and it is set out below — but it is a choice to keep a layer that the rest of this page retires.

My take: Nothing here needs its own answer. Confirm the mechanism and this closes with it — it is listed only so you can see it close.

What this call does to everything else written on these pieces

Nothing else written on these pieces is touched. The brand each belongs to, the crafts read out of it, the pile it came from — all unchanged, because none of them ever pointed at a family name in the first place. The only thing that moves is where a name appears when you are browsing the list of names, which is a page, not a fact about a piece.

The alternative, and its cost: Keep the two names and finish naming them. The cost is that you would be settling wording for two groups whose members you have already split across four of your own names, so the two new names would describe nothing anyone actually browses — and the parked question comes back every time someone reads the list.

Where this was worked out: proposal 5 on the reconciliation page

P-06Piles that never end, like Customer Stories · full

Four customer-story piles are stored as campaigns and one of them belongs to a different brand entirely. Is a pile that never ends an always-on campaign, or a new kind of container?

A campaign in the model is a container: it has a name, a stretch of time, and pieces that belong to it. Most campaigns end. The awkward ones are the piles that do not — a customer-stories programme just keeps taking new stories, year after year, and nobody ever declares it over.

This stopped being hypothetical while we were walking. Four of them are live right now, each stored as a campaign whose kind is “customer story campaign”: Customer Stories with 30 pieces, Customer Proof: AI in Action with 9, Humans of HOKA with 8, and Customer Stories: AI in Action with 4.

The Ramp story below is what one of them looks like from the inside. “How Ramp engineers accelerate code review with Codex” is a case study — that is what is written on it — and belonging to the Customer Stories pile adds nothing to the piece except membership. The hero image captured from the same page is its own piece, in the same pile, with its own label.

The one that should make you sit up is Humans of HOKA. That is a second brand, with 8 pieces, doing the identical thing: an ongoing pile of human stories with no end date. Nothing in our own machinery asked for that. It turned up because the pattern is real out in the world, not because we modelled it.

So what is missing is a word, not a structure. An outside review argued these deserve a new kind of container — a programme, or a series. Your reaction was that “ongoing advocacy programme” is not how this industry talks, and that these are still campaigns, just ones that never switch off. Adding always-on to the list of campaign kinds costs one word and changes nothing about where the four piles live.

My take: Take the word, not the new container. Your own vocabulary and the industry's agree here, the storage is already right, and if a pile ever genuinely refuses to behave like a campaign you can mint a container then, with a real example in front of you instead of a hypothetical one.

At Ramp, engineers are using Codex with GPT‑5.5 to accelerate code review and develop internal agentic tooling, helping teams get substantive pull request feedback in minutes instead of hours. Thanks to its reasoning capabilities, Codex with GPT‑5.5 is uniquely able to reduce the amount of manual, hands-on work they’d otherwise have to do.
text only — no image on this piece

What this call does to everything else written on these pieces

Nothing written on a piece changes and no piece moves between piles. The link that says “this case study belongs to Customer Stories” is the same link before and after; it simply gains a truthful word for the kind of thing it points at. The one place you will feel it is in questions you ask later: today, asking for the campaigns that ran last quarter quietly sweeps up four piles that never started and never stopped, and after this you can ask for one without dragging in the other.

The alternative, and its cost: Mint a new kind of container — a programme, or a series — alongside campaign, brand-identity programme, project and event. The cost is real: the four piles have to be re-filed out of campaigns and into it, and every question already written against campaigns has to learn about a second container or quietly stop returning these pieces. There is a middle road too — take the word now and revisit the container only if a real pile refuses to fit — which costs nothing and keeps the option open.

Where this was worked out: proposal 6 on the reconciliation page, and the third of your three use cases on the recommendation page

P-07A name for the middle level · full

Between one creative — the film or image as it was made, before any cut or upload — and each place it appeared, is there a middle level worth giving a name to?

Start with the ground you already ruled, because one word on it went wrong. The smallest thing we store is a single item: an Instagram post, a video, the hero image on a page. Each one is tagged on its own, and that is the product. The word “work” was tried for it and you rejected it — you never call anything a work — so it is gone and nothing has replaced it.

Above an item there is a link, not a container. When the sixteen-by-nine upload and the nine-by-sixteen cut are the same film, a link between those two items is what lets you ask to see that film everywhere it ran. You have already ruled that this link exists and that aspect-ratio cuts are covered by it.

The pick is only what to call the middle of that chain. Take a short vertical edit of a film, made for French-speaking markets — a French cut, in the trade's shorthand: it means the French-language edit, not a style of editing. That edit is a version of the creative; the upload of it is an appearance of that version; the capture we hold is the evidence of the appearance. The two pieces below are the everyday shape of this: the Ramp case study page and the hero image captured from that same page are two separate items, each fully tagged, each pointing back at one thing.

Naming the middle earns its keep later rather than now. It gives you a word for “these four are the same film” when you are spotting duplicates, and it is the level that real numbers and rights eventually hang off — the French-language edit is licensed differently from the English one. Without the word, every one of those conversations has to be had longhand.

And it stores nothing. There is no new place to put anything and nothing to fill in; all 13,938 pieces in the library stay exactly as tagged as they are today.

My take: Name it. It costs a sentence of vocabulary now and saves you the longhand every time the same film turns up in four shapes across three places.

At Ramp, engineers are using Codex with GPT‑5.5 to accelerate code review and develop internal agentic tooling, helping teams get substantive pull request feedback in minutes instead of hours. Thanks to its reasoning capabilities, Codex with GPT‑5.5 is uniquely able to reduce the amount of manual, hands-on work they’d otherwise have to do.
text only — no image on this piece

What this call does to everything else written on these pieces

This touches nothing that is written on a piece. Labels, brands, crafts, products, the pile a piece belongs to — all unchanged, because the version idea sits above items rather than on them. The one relationship it does touch is the link between items that are the same creative: today that link is a flat set of appearances, and naming the middle gives that set an internal shape, so “the French cut” and “the French cut on YouTube” stop being the same sentence. Nothing has to be built for that to be true — you are agreeing what the word means before anyone needs it.

The alternative, and its cost: Keep the two levels exactly as ruled — a creative and its appearances, with nothing named in between. The cost is steady rather than dramatic: when one film exists as four cuts across three places there is no word for the four, so the duplicate-spotting work later has nothing to hang on, and every conversation about rights or numbers has to describe the group from scratch.

Where this was worked out: proposal 7 on the reconciliation page

P-08One label per piece, or several · full

May a piece carry more than one whole-piece label, with one of them marked as the one that leads?

Your own example already states the rule, so most of this is confirmation rather than argument. A screen recording of a feature being used is a whole piece, and what is written on it is product demo. A product demonstration inside a hero commercial is a moment inside that piece — never the commercial's label.

The first two pieces below are that sentence in the library. “A research preview of Operator…” is a demonstration from beginning to end; product demo is written on it, and 751 pieces carry that label today. The Nike film “Get Ada Her Way…” stays a commercial — one of 55 — whatever happens inside it: if a few seconds of it showed the product in use, those seconds would be a moment cut from the film, never the film's own label (no moment has been cut from this one yet, so it carries the label alone). That is the rule: the label says what the whole piece is, and anything shorter inside it is moments inside the film rather than what the film is.

The one thing genuinely open is the case where two labels are both true of the whole piece. The third piece below — the hero image from the Ramp story — is that kind of piece: end to end it is a customer's story and a demonstration of the product in the same frame, with no honest way to call either one a moment inside the other. What is written on it today is blog post, which is not wrong and is not the whole truth either. Your August ruling says a piece may carry both, with one marked primary.

You flagged “secondary type” as sounding wrong, and you were right. It makes the second label sound like a lesser fact about the piece, which it is not. Primary is the word that survived: it says which label leads when the piece has to be shown in one place, and says nothing at all about which of the two is more true.

This shape is already running elsewhere in the product. Moments sit on several shelves at once with one main shelf, and nobody finds it confusing. Whichever way you mark this, the rule about moments is untouched.

My take: Confirm several with one leading. The bar — a second label must describe the whole piece — is what stops this becoming a free-for-all, and that bar is your own sentence, not mine.

What this call does to everything else written on these pieces

The only thing this reaches is the boundary between a whole piece and a moment inside it, and it reaches it by holding it exactly where it already is. Moments keep their own words: a word that describes one moment inside a piece is never promoted up to stand for the whole piece, and the label on the whole piece is never pushed down onto a moment inside it. Where a second label does exist, the one marked as leading is what a browse list shows and what a count counts, so the numbers you have been reading on these pages stay comparable with the ones you read after.

The alternative, and its cost: Amend to one label only, and add a link for a second label later if it is ever needed. The cost is that a piece that is honestly two things has to be filed as one, and that choice gets made silently at the moment the piece is read in, where nobody sees it and nobody argues with it.

Where this was worked out: proposal 8 on the reconciliation page

P-09Content scraped from inside a product · full

Content we scrape from inside a product — the ChatGPT app, the web app — does it need a family of its own, or is coming from a product just a link?

Your question, in your words: “should we probably put Digital Product somewhere… like where do we put that ChatGPT like ios and or web app etc — re content we scrape from it.” It is a fair question, and the answer is smaller than it looks.

Coming from a product does not change what kind of piece something is. A screen captured from the ChatGPT app is a screenshot of an interface; a help page scraped out of the web app is a help page. Which product it came from is a connection we already store between the piece and the product — 139 products are on file, each tied to the company that makes it, with 394 more sitting as product lines inside their companies. Where in the product it appeared is answered by the where-it-ran fields you already settled.

The first of the two pieces below shows that already working. “A research preview of Operator…” is linked to Operator and to ChatGPT, and nothing along the lines of “this came from a product” had to be written on it for that to be true. The Microsoft page beside it is the other half of the point: we hold it as a full-page capture, an image of a web page — and it is still a blog post. Capturing a thing did not change what the thing is.

One gap is worth saying out loud rather than glossing: we do not hold a capture from inside the ChatGPT app today, so the exact case you asked about is one the library has not met yet. That is precisely why I would not build a family for it now. If raw interface captures do start arriving and they genuinely feel like a kind of their own, one new label — product interface screen — covers it, and it sits with the other interactive and digital-experience names.

The alternative writes the same fact down twice, which is the one habit this whole review has been trying to break.

My take: Product-ness is a link. The only part I would leave open is the new label — I would not mint it until a raw capture actually arrives and needs it.

What this call does to everything else written on these pieces

Two things already carry this answer and both stay exactly as they are: the link from a piece to the product it shows, and the where-it-ran fields that already tell an owned app screen apart from a public web page. A stored family would sit on top of both and repeat them, which means every product-scraped piece would carry a value that has to agree with a link forever — and when the two disagree, nothing in the model says which one is right. Everything else written on such a piece — its brand, its crafts, the pile it belongs to — is untouched either way.

The alternative, and its cost: Store a Digital Product family on those pieces. The cost is exact, and it is the same cost measured on every other stored-family option: a second copy of facts the product link and the where-it-ran fields already carry, kept in step by hand forever, with no rule for which copy wins when they drift apart.

Where this was worked out: proposal 9 on the reconciliation page, and the section of the recommendation page headed “Where digital-product content lives”

P-10The Groupings name — your glance · full

The part of the model that covers campaigns and every other pile a piece can belong to is already called Groupings, on your own words. Does the name stand?

Your banked words, in order: “maybe Groupings or Grouping Content idk… i still dont like works bc i never call anything really ‘works’… campaign & groupings isnt that bad.” Three names, and you tested them yourself as you said them.

Groupings was taken because it is the only one of the three that does not tilt. Grouping Content you were lukewarm about in the same breath as saying it. Campaign & Groupings names one kind of pile in the title of a section that holds five kinds, so a reader arriving at a customer-story pile or an event would have to be told the title does not quite mean what it says.

The pieces below are what the word covers. The embroidered patches and the photograph of the venue atrium both belong to the same pile, named for that event; the Ramp case study belongs to the Customer Stories pile. 312 pieces in the library belong to at least one pile like that and 13,626 belong to none — and nothing extra is written on any of them because of what the section describing piles is called.

Two edge cases you raised are answered on the page rather than left hanging. An event's pile is usually campaign-shaped, because campaigns attach to events, so it is a pile and Groupings covers it. App and web product content is a link to a product rather than a pile, so it is not in here at all.

The ruling underneath the name is unchanged and stays in your words: one creative, many appearances. This is a glance rather than a decision — if you say nothing, I will read it as confirmed.

My take: Confirm. Of the three names you floated, this is the only one that does not put one kind of pile above the others.

At Ramp, engineers are using Codex with GPT‑5.5 to accelerate code review and develop internal agentic tooling, helping teams get substantive pull request feedback in minutes instead of hours. Thanks to its reasoning capabilities, Codex with GPT‑5.5 is uniquely able to reduce the amount of manual, hands-on work they’d otherwise have to do.
text only — no image on this piece

What this call does to everything else written on these pieces

A name for a section of the model touches no piece and no link between pieces. What it does touch is everything that refers to it — pages, cross-references, the address in a browser — and that has already been done: the old address is gone rather than quietly forwarding, so nothing can keep using the old word by accident. If you amend the name, the same pass runs again and nothing stored moves.

The alternative, and its cost: “Campaign & Groupings”, the one you said isn't that bad. The cost is the tilt: it names one kind of pile in the title of a section that holds five, so a customer-story pile, an event and a brand-identity programme all arrive under a heading that names something else. One pass over the settings and the pages either way.

Where this was worked out: the pillar renamed on 17 August on your words, and the display wording ruled with it