Structured,
not generated.
Brandrail is a social media platform where every post is built inside your brand's own design system, measured by gates that fail closed, and stamped by you before anything ships.
The AI does not author your brand. It operates the platform, under the same rules as everyone else.
Drag it. Nothing about the post changes, only whether the system holds its four plates in register. That difference is the whole product.
Northfield Bouldering
New set drops Thursday
40 problems · every grade · doors 6pm
@northfieldboulder
- Job
- NF-0431
- Layout
- event-poster
Synthetic Northfield Bouldering is an authored demonstration brand. No customer content appears on this page.
Every AI social tool ships the same pipeline.
Theirs
Probabilistic all the way down. A model constrained by a prompt regresses to the mean of its training data, and content at the mean is exactly what people mean by slop. No amount of prompt engineering fixes that, because the mean is where the model lives.
Ours
The platform is the creator. Layouts are authored specs in a registry, so they are not generated. Colour is a semantic token the engine resolves, so it cannot be invented. Publishing is a state transition gated on your approval, so it cannot be taken. The difference between a guardrail and a structure is the difference between please don't and there is no such operation.
“Some posts just come again and again — the system dont learn what i dont like.”
“Output requires editing in almost every case… lacks the specificity and voice.”
Eight gates. They fail closed.
A pressman does not judge a sheet by eye. A densitometer reads the colour bar and returns a number, and out of tolerance means the run does not start. Our gates work the same way: they are code, not a model's opinion, and a post that fails one does not degrade to published but mediocre. It degrades to not published.
Thresholded similarity against everything you have already published. This is the direct answer to the same post keeps coming back, the most common complaint about every generator on the market. Over the threshold, the draft never reaches your desk.
Every localized field required by your content schema must be present before a post is loadable at all. A missing translation is not a warning: it fails the load.
Your calendar carries a typed coverage policy: how far ahead it must be filled, how often each channel may fire, which slots have owners. The AI cannot flood it, because cadence is a constraint the calendar enforces rather than a habit it is asked to keep.
Assets resolved, references intact, channel contract satisfied, execution slot owned and dated. Readiness is checked before a draft is queued, not discovered at the moment of publishing.
Three scored dimensions (anti-slop, source integrity, editorial quality), each of which must pass on its own. A high average does not rescue a failed dimension. This is the one gate with judgement in it, and it is scored against a written rubric rather than a vibe.
Unknown references, duplicate IDs, orphaned assets, offers pointing at nothing. Content is typed, so a broken reference is a compile error. Post proposals have to compile before anyone reads them.
Your brand's colour ramp is derived, not hand-picked, and every text-on-field pairing the engine can produce is contrast-checked when the ramp is built. A layout cannot render type your own palette would fail.
Format, aspect, length and cadence checked against the channel the slot actually targets. A carousel does not go out as a short, and a slot with no owner does not go out at all.
Readings shown are illustrative values for the demonstration post above.
What the AI is allowed to touch.
Not a policy document. This is the access-control list the platform enforces, and it is the same list whether the operator is a model or a person. Think of it as a job description: a capable junior at a disciplined studio, structurally incapable of shipping without an editor's sign-off.
| May | Through | Gated by |
|---|---|---|
| Propose copy into typed fields | Content schema | Grammar, slop rubric, locale, near-duplicate |
| Select a layout | Template registry | Registry membership, strict spec validation |
| Choose imagery | The library | Provenance and rights metadata; no external fetch |
| Propose schedule slots | Calendar policy | Horizon rules, cadence caps, execution contracts |
| Draft variants and locales | Same schema as originals | The same gates as originals |
| Read performance context | Analytics API | Read-only; no customer-level data |
| May not | Because |
|---|---|
| Render pixels | Rendering belongs to the engine |
| Define colour or type | Both are derived from your brand |
| Create layouts | Layouts are authored and curated |
| Add to the library | Assets carry rights you own |
| Move a post past review | Approval is yours |
| Publish or spend | Autonomous publishing is refused in code |
You keep every one of these. Authoring is always open to you: new layouts, new assets, edits anywhere. The escape valve from a constraint is a human authoring more system, never a switch that turns the system off.
The same nine posts.
A magic demo wins the first minute. We are asking about the thirtieth day, when the grid is the thing a customer actually sees, and every choice made without a system has had a month to accumulate.
Generated · nine prompts, nine outcomes
Off-houseNew set drops Thursday
Northfield
Beginner session Sunday 10am
Northfield
Meet the setters
Northfield
Comp night entries open
Northfield
How to fall properly
Northfield
Member of the month
Northfield
Chalk up, warm up, send
Northfield
Youth club autumn term
Northfield
Open till 11 on Fridays
Northfield
Structured · nine slots, one system
In registerNew set drops Thursday
Northfield
Beginner session Sunday 10am
Northfield
Meet the setters
Northfield
Comp night entries open
Northfield
How to fall properly
Northfield
Member of the month
Northfield
Chalk up, warm up, send
Northfield
Youth club autumn term
Northfield
Open till 11 on Fridays
Northfield
Synthetic Both grids are authored for this page from one set of nine headlines. The left grid is what happens when each post is decided on its own; the right grid is the same nine posts resolved through one token system and one layout family. Neither is a screenshot of any product, ours or anyone else's.
Nothing runs until you stamp it.
Approval is not a tier, a toggle, or a plan feature. It is a state transition, and the transition past review has no automated caller. Ten minutes on a Monday, on your phone, with the gates' findings already attached to each sheet.
| Time | Sheet | Layout | Gates | State | Stamped |
|---|---|---|---|---|---|
| Mon 08:12 | New set drops Thursday | event-poster | 8 / 8 | OK to run | DM |
| Mon 08:14 | Beginner session | offer-card | 8 / 8 | OK to run | DM |
| Mon 08:15 | Chalk up, warm up, send | quote-plate | 7 / 8 | Blocked · near-duplicate | — |
| Mon 08:19 | Meet the setters | portrait-strip | 8 / 8 | At the desk | — |
| Mon 08:21 | Comp night entries open | event-poster | 8 / 8 | Running | DM |
Every sheet stays reopenable.
A post is structured data plus a layout reference, never a flattened export. Three months later you can open it, fix the copy, swap the image, correct a locale, and it re-renders identically inside your brand. Human edits and AI edits go through the same door, so there is no AI version drifting away from a manual version.
That is also why the library compounds. Month six is an archive that still renders, not a camera roll of dead JPEGs.
Revision 3 · in register
Northfield Bouldering
New set drops Thursday
40 problems · every grade · doors 6pm
@northfieldboulder
Point us at your website.
You are not going to build a design system. We read your live site, resolve its real colours, fonts and spacing back into brand tokens with provenance and a confidence score for each one, and show you what we found before anything is written.
Then you get a week of drafts, on your grid, in draft, waiting at the desk. The demo ends at the review queue, because the review queue is the product.
This form is a design demonstration and is not wired to a backend.
Still open
Pricing, availability and launch timing are not decided and are deliberately absent from this page rather than invented for it. Replace this panel with real commercial terms before any public use.
What we never say
Thirty days of content in five minutes. Supercharge. Unleash. Magic. 10X. Fast to polished is a claim we can keep; fast to plenty is the promise that made this category churn.