# Problem And Approach

## The problem

A convincing image is not necessarily the image a commercial brief calls for. The production problem is to carry the brand's creative intent, product truth, reference roles, and permitted variation into the work, then judge the result against those same standards. When that structure stays implicit, "creative direction" is made to carry too much at once.

"Creative direction" routinely fuses objects that do different jobs:

- **intent** — the aim the asset must express
- **constraints** — what must not be violated
- **references** — the visual anchors that override a model's defaults
- **creative discretion** — the bounded judgment exercised within the permitted-variance aperture (granted by the brief) and at selection to close it
- **selection** — the adjudicative transition that chooses one candidate and makes it true to the intent
- **artifact governance** — the mechanism that binds the accepted result to its claim, use, and answerability
- **the governance record** — the durable trace of that mechanism, not the mechanism itself

Collapse these into one "creative direction" field and the system can still produce convincing images while losing the standard that makes them right for the job. Model defaults can fill gaps the brief should resolve, and selection can favor an attractive result without establishing that it fulfills the creative intent. The definition layer makes the production ask explicit so generation and selection can be directed and judged against it. Generation produces candidates; it does not establish their fitness. A generated image is not yet a governed asset. Selection judges the realized alternatives; artifact governance binds the accepted result to its claim, permitted use, and answerability. The record preserves that basis. The purpose is to carry intent through production, not merely to explain an output after the fact. These distinctions are the difference between a prompt and an architecture.

## The approach

- Name those objects separately, and make the layered structure that carries them legible — where each lives, and how it inherits, overrides, and combines.
- Treat the bounded creative-discretion aperture as a first-class carrier, distinct from the constraints that bound it and the references that anchor it.
- Keep selection (the adjudicative transition) distinct from governance (the mechanism) and from the governance record (its durable trace).
- Carry all of this as a **portable asset-definition package** — the definition layer, designed to be substrate-independent, that a production system needs *before model choice matters*.

## Start with the concrete object

The fastest way to see the distinction is the worked package, not the abstract ontology:

- [`asset-definition-package-map-2c-marketing-hero-v1.md`](asset-definition-package-map-2c-marketing-hero-v1.md) — one package carried end to end — its bounded-generativity topology read as a function-order projection (intent → discretion → variance → selection → governance).
- [`package-lifecycle-partition-v1.md`](package-lifecycle-partition-v1.md) — the three lifecycle states a package spans (prospective definition / execution-run state / governed record), and how to read its statuses without contradiction.

## How the model is exercised

The conceptual structure (ontology) and its use in templates and worked examples (application) are pressure-tested across four worked-example workflow modes — built as the smallest honest structure first, kept provisional rather than claiming premature completeness. The ontology / application split and the discovery → diagnosis → redesign sequence are described in [`method.md`](method.md); the current architecture front door is [`architecture.md`](architecture.md).

The approach is still being refined in public.