Guidelines are passive documents
Two hundred pages on the intranet, a PDF in the DAM, a deck from the last rebrand. A document can describe what applies. It cannot apply itself.
supakraft is the operating core for AI marketing operations. The core binds a task to the brand truth, the scope and the rules that apply to it. If a precondition is missing, the run stops. Verification, human approval and delivery stay separate steps.
30 minutes on the system, along a single run.
Four things that sit side by side in most organisations without being connected.
Two hundred pages on the intranet, a PDF in the DAM, a deck from the last rebrand. A document can describe what applies. It cannot apply itself.
A model finds material. Which source is authoritative for this specific task is not stated in the material itself. So it gets guessed.
Because nothing decides beforehand, review decides everything. One person checks, feedback takes days, and under deadline pressure the step disappears.
PIM, DAM, guidelines and brief. Before every output somebody would have to bring them together and settle the contradictions. Mostly that does not happen.
supakraft does not produce content in place of your tools. The operating core records which information may apply to a task, on what basis a model was called, which checks ran and which person approved.
Models, surfaces and connectors can change. The governance boundaries stay the same.
Apps, forms, agents and partner tools call governed core operations. A surface may differ by client. The enforcement behind it must not.
See the platformA governed run follows the same sequence, from task to delivery. What cannot be resolved in one stage is not guessed in the next.
Exact identities, authority, scope, validity and conflicts.
Compose already resolved inputs into a pinned task context package. The package is model neutral.
Bind package, policy and renderer before the provider call.
Check the stored output against its acceptance contract.
A verified eligible person approves the exact stored subject.
Recheck current eligibility before the external effect.
An adapter translates. It does not decide what is true.
Which domain is connected in a given deployment is stated on the platform page.
Brand truth is held as approved, scoped information. An element carries source, validity and knowledge state. A rule can require or forbid something. Missing truth is not filled in from model knowledge.
Brandbooks remain useful for people. The brand core additionally holds the approved decisions in a form software can resolve and check. Both can reference the same sources, but they are not the same object.
element: colour.primary value: "#1F2AFF" state: defined source: brandbook, section 2.1 scope: [ web, print ] element: colour.secondary state: unknown note: no value decided, not derived
Schematic, in the format of the open standard. No real client data.
Every governed run preserves the basis used for the decision. A reviewer can see which release, rules and evidence were used.
The shared core has proven a deliberately narrow governance path for brand truth. The proof covers defined core commands, a versioned execution basis, bounded checks, human approval, withdrawal and deterministic delivery within the tested boundaries.
Product and service knowledge is being integrated as the first additional knowledge domain and falsified against contradictory sources.
This list is maintained as development proceeds.
The Open Brand Definition Specification is published by Kill The Dragon GmbH and is separate from supakraft. The brand core uses its contracts. Specification under CC BY 4.0, implementation layer under Apache 2.0.
Brand truth is described in an open, documented format. supakraft can store that data operationally, but the format itself is not proprietary.
The format is machine-readable and independently implementable. Export and exit are handled as separate operating agreements.
defined, unknown, not_defined, not_applicable. An open value stays open and is not filled in by derivation.
Current version, schemas and conformance material are published at openbranddefinition.org. The standard makes decisions traceable. It is not a certification.
A brand does not become governed through an import. The first useful milestone is concrete: brand truth is in the system, rules and prompts are bound, a real task runs through the governed path, and an eligible person makes the first approval.
Import existing OBDS material where available, otherwise author and ratify the required brand truth. Gaps are preserved rather than filled.
Markets, channels, rules, prompts and renderers, approval requirements.
One bounded real task, with exact context, checks and evidence.
The exact stored output, a verified actor, visible evidence. Expansion is discussed after that.
We would rather say exactly what is fixed per instance than make a smooth statement that falls apart in review.
30 minutes on the system. We show one bounded run: which brand truth applied, what the model was called on, what was checked and who approves.
Book a demo