The house words.
Twenty-eight words we use often and understand strictly. Each one has, behind it, a table, a file, or a price — not a pretty definition.
In an ordinary agency, words widen along the way. "Strategy" ends up meaning any long document; "system" ends up meaning anything done more than once. The widening isn't noticed in time, because nobody loses anything on the day it happens — it's lost three months later, when two people think they talked about the same thing.
Here, words can't widen, for a mechanical reason: almost every one of them is also a thing in the machine. Brand isn't an opinion about you, it's a row in a table. Ban isn't a preference, it's a section that disappears from what the model reads.
Every entry has the same shape: what it is, what it gets confused with — the part that costs money when you mix it up — and where it lives. If a word has no third part, it isn't a house word: it's a figure of speech, and it didn't make the list.
Six words that get confused with each other and cost money when they do.
The most expensive confusions on the list are here: who presses the buttons, and how long what you bought stays valid.
Four one-off services, done once, in order: positioning (1.900 €) → visual identity (1.400 €) → web (2.400 €) → content production (1.800 €). It's not a subscription and not an application tier.
Confused withthe "starter package." It isn't: the order isn't negotiable, because each service consumes the output of the one before it — you can't do visual identity for a positioning that doesn't exist.
Where it livesthe four-row map in sop/README.md
1.490 €/month. The engines do the production, one of our people checks every piece before it goes out. You buy the result, not the tool.
Confused withthe Application. The difference isn't "more pieces" — it's who presses the buttons.
Where it livesthe same procedure as the one-off production: SOP-PRODUCTIE-CONTINUT.md
249 €/month, the first tier. The same engines, but the buttons are yours. Nobody on our side looks at your pieces before you publish them.
Confused withthe System, in both directions. Whoever buys the Application expecting a person is disappointed; whoever buys the System thinking they get access to the app pays six times more than they need to.
Where it liveswhat each tier gives, written out row by row
The subscription level for the Application: 249 € / 590 € / 990 €. What changes between them is the number of brands (1 / 3 / 6) and the monthly quantity of pieces that get rendered.
Confused with"plan." The word "plan" means something else for us — see the row at 05.
Where it livesthe clients.plan column in the database, plus the counters that read it
One single written piece, once, no card. It's not "a free month": it contains no explainer and no carousel, because those cost real rendering money from the first click.
Confused withthe usual 14-day trial. Ours is a piece, not a window of time.
Where it livesthe trial counter, checked before the first order
Extra quantity bought on top of the tier, for the current month: +4 explainers (149 €), +8 carousels (79 €), +1 brand for one month (99 €). They expire at the end of the month — they don't roll over.
Confused witha tier upgrade. The add-on is a single month; the tier is permanent until you change it.
Where it livesthe same counters, with an expiry date
Four words for four different things people usually call "client."
This is where the confusion we made ourselves lives, for months: a brand treated as a channel.
The account that pays. A client has a subscription, a tier, and an interface language.
Confused withthe brand. A client can have several brands; the brand never pays.
Where it livesthe clients table
The identity the content goes out under: voice, colors, audience, bans. It has its own Brain and its own profile.
Confused withthe channel. The most expensive confusion on this list. We made it ourselves: @vivid.alex was treated for months as a channel of Boosted Mode, even though it speaks to a different audience, in a different language, with a different voice. It's a brand.
Where it livesthe brands table; the content data, in the brand's vault
The place where it gets published: an Instagram account, a page, a newsletter. A brand can have several channels.
Confused withthe brand. The mechanical rule we set for ourselves: a channel can only narrow, never widen. If the brand shut off a pattern, the channel can't turn it back on — trying to stops the engine with an error. A ban that the banned party can lift on their own isn't a ban.
Where it livesthe channels table; the rule, in the matching layer
Who a piece talks to. We have two: the Delegator — the person who doesn't want to learn one more tool and would rather hand the work off; the Operator — the person who's already working with the tools and wants a better one.
Confused withthe income bracket. It's not about how much the person makes, it's about their relationship with the work. Two people with the same number can be different personas.
Where it livesportrete: from the brand's content file
Seven words for the material every piece is made of.
None of them mean "trained on your data." Everything here is text that enters the prompt every time.
Everything the system knows about a brand. Two layers: the profile (who you are, what you sell, how you speak) and the entries (what happened to you). It's read on every content order.
Confused witha chatbot trained on your data. It isn't training — it's text that enters the prompt every time.
Where it livesthe profile answers, plus the brain_intrari table
Twenty-three questions, eight mandatory, four groups. The answers are the source; the vault copy is generated from them, not the other way around.
Confused withan onboarding form you fill in once and forget. It's reread for every piece.
Where it livesthe list INTREBARI — written out in full here
A thing that happened to you, written in plain words: a client who said something, a number that came out, a mistake. The engine takes five per order, oldest first, and only marks them used after the piece exists.
Confused with"post ideas." An idea is already a half-written piece; an entry is raw material. Ten pieces come out of raw material; one comes out of an idea.
Where it livesthe brain_intrari table, the folosit_la column
The universal pattern library, written once for every client: 19 hooks and 5 skeletons, in Romanian and in English.
Confused withthe SOP. The playbook is what the model reads; the SOP is what the person reads.
Where it livesone folder per language in the vault
The procedure a person follows to deliver a service, step by step. Four services, five files — research has its own file, but it isn't sold separately: it's phase 3 of positioning. Separate from those, four category briefs (expert, shop, product, local services), which enter the prompt of every generation.
Confused withthe playbook, and "documentation." A SOP doesn't describe what we do; you follow it and the result comes out.
Where it livesthe sop/ folder, plus the category briefs in the vault
The layer that picks, from the universal library, what applies to this particular brand. A brand doesn't get a copy of the library — it gets a list of IDs.
Confused withcustomization. Customization would mean a copy per client, one that silently drifts from the original and nobody notices, because the piece still comes out just as plausible.
Where it livesthe matching layer, read together with the brand's content file
What isn't allowed to appear in a piece for this brand.
Confused witha negative instruction in the prompt ("don't use X"). It isn't: what's banned disappears from the text before it reaches the model. A model that still sees the pattern uses it every once in a while, and the ban turns into a suggestion.
Where it livesinterdictii: from the content file; the cut, in the function that builds the prompt
Eight words for the path from a click to an object you can publish.
This is also the one place where we count differently than the folders do: a folder isn't an engine someone can request.
One click: brand + channel + engine + language. One or more pieces are born from it.
Confused withthe piece. The order is the request; the piece is the object.
Where it livesthe orders table
A content object: an explainer, a carousel, a post, a graphic. It has a title, a type, a status, and a version history.
Where it livesthe content_items table
Every revision of a piece, with the reason written down. Nothing gets overwritten.
Confused witha "saved draft." All the versions stay, with the reason for the revision next to them.
Where it livesthe content_versions table
The program that produces a particular kind of piece. Four produce something: post, carousel, explainer, graphic. The order screen only shows three — the graphic isn't requested separately, it comes automatically alongside the post, and if it fails, the text still goes out. The fifth folder is a test engine that does nothing useful: it exists to prove the manifest → validate → run chain, and gets deleted when it's no longer needed.
Confused with"the AI." The model is one piece of the engine, not the engine. An engine also has a planner, a renderer, contrast checks, and an approval gate.
Where it livesthe engines/ folder, one folder per engine, each with its own manifest
The pattern of the first seconds — why someone stops scrolling. Nineteen, numbered H-01 … H-19.
Confused withthe headline. The headline is text; the hook is a shape, which gets filled with your text.
Where it livesthe hooks file in the playbook, per language
The shape of the whole piece, start to finish. Five, S-01 … S-05, each measured on real pieces, not invented.
Confused withthe template. A template has text in it; a skeleton has only the order and the duration of the parts.
Where it livesthe content production file in the playbook
The visual language of the explainers: a millimetric surface on which plates, numbers, and pictograms appear and disappear. All the text sits on the canvas — there's no text that floats.
Confused witha "style." Right now it isn't a style, it's the only language; the style switcher is on the list, not built.
Where it livesthe explainer engine, the render surface
A piece the system suggests, not you — born from a Brain entry. It passes through a mechanical gate: if the reason for the proposal doesn't share at least two word roots with the entry it claims to come from, it gets discarded and counted, not silently saved.
Confused with"AI-generated ideas." The difference is the anchor: a proposal with no verifiable source doesn't appear at all.
Where it livesthe proposal generator
Three words for where the machine stops and waits for a human.
The last of them exists only to avoid colliding with "tier" — and it's worth writing down, because the collision already cost a page rewrite.
A point where something stops and waits for a human. It has three states: pending, approved, rejected — and it records who decided and when.
Confused witha status. A status describes; a gate blocks.
Where it livesthe gates table
Where a piece has gotten to. Nine values, from draft to published, archived, or rejected. They aren't free-form labels: the database refuses any other value.
Where it livesthe item_status type in the database
For us it means the editorial plan — which themes get written, in what order. The subscription is never called "plan," so the two don't collide.
Confused withthe "subscription plan." That's why we say tier.
Where it livesthe editorial planner, not the subscription column
Two things this glossary doesn't cover, so it doesn't look more complete than it is.
Words change. "Done" was until recently published, and archived only showed up once we needed a status for what's no longer active. The date in the header is the date the list was checked against the code, not the date it was written once and for all.
It isn't technical documentation. Table names are there so you can verify, not so you can work with them. What the system can technically do is a different question and a different document.
These words look better at work than defined.
The list above is utilitarian: it's linked from the other texts, at the first appearance of a word. If you got here from an article, go back to it — the glossary has nothing to sell, only to clarify.