Protected capability build

Advertising with a spine and a soul.

The Advertising Performance Engine converts either Decision Engine output or a direct user intake into one campaign brief, then applies reusable brand assets, composition specs, hypothesis, evidence, visual idea, measurement, and review gates.

2input paths
1canonical brief
5export profiles
Open intake Open workbench Open assets
Advertising Performance Engine architecture map

Two ways in. Same discipline out.

The system accepts strategy-led advertising from the Decision Engine or a guided direct intake. Both routes are normalized before creative work begins, so no campaign bypasses measurement, compliance, or review.

Decision Engine handoff

  • Imports strategy output, influence population, delivery actions, source orientation, and legal limits.
  • Preserves the Decision Engine's distinction between accepted context, candidate context, and blocked material.
  • Converts strategy into a campaign hypothesis without pretending execution has already been validated.

Interactive intake

  • Asks for objective, audience, offer, channel, budget, timeline, landing destination, conversion event, brand voice, and legal constraints.
  • Forces missing parameters into the open before creative generation starts.
  • Produces the same CampaignBrief object as the Decision Engine path.

The brief is the control surface.

Creative output must be traceable to the brief and the reusable brand-asset layer. The brief names the audience truth, strategic tension, evidence basis, channel rationale, and the measurement rule that decides whether the idea lives.

CampaignBriefObjective, audience, offer, message contract, channels, constraints, measurement, and review.
BrandKit + AssetRegistryReusable logos, colours, typography, imagery, generated assets, evidence objects, rights, and approved contexts.
CompositionSpecLayered final composition for the workbench, Figma, Photoshop finishing, platform exports, and approval history.

AI image generation is restricted to raw source visuals and palette mood only. Final text, logo, typography, branding, legal lines, side panels, copy bays, crops, and layout stay editable in the CompositionSpec and are chosen by brand rules, colour theory, accessibility, and platform legibility. Visual prompts must name the precise demographic and use subtle recognised symbols: objects, routines, places, documents, tools, queues, vehicles, rooms, weather, or time-of-day cues that the intended group reads instantly without crude stereotyping. When people appear, they are evidence-led messengers: ethnicity, age, household role, occupation cues, and suburb/seat context must fit the persuadable audience using AEC, ABS, polling, booth, seat, and segment data. Creative direction now requires a concept bridge before execution: empathy, relatedness, indirect symbolic connection, humour or twist, medium choice, then typography and layout.

Creative soul is a gate, not a mood.

Output is rejected if it looks like a generic services advert or sounds like strategy talk. Strong work must make people feel the unfair cost before they think about the evidence.

Creative direction test
Albo promises. You pay.
A concept must make the cost feel personal before it explains.

Ten scoring dimensions

  • Audience truth
  • Strategic tension
  • Offer clarity
  • Proof
  • Visual idea
  • Verbal idea
  • Human language
  • Testability
  • Symbolic bridge
  • Claim safety

Production spine

The first production slice is now explicit: one end-to-end local runner, durable local run registry, 24-case deterministic canary, CampaignBrief-to-CreativeActionBrief generation, creative territory and image-prompt specification, governed Google image generation, CreativeActionBrief-to-CompositionSpec generation, creative scoring, brand-asset validation, composition rendering, design-handoff packaging, GIMP adapter packaging, Figma plugin packaging, and platform export profiles. A draft can pass, revise, or block before anything moves toward campaign deployment.

End-to-End RunnerRuns either interactive intake or Decision Engine handoff through CampaignBrief, CreativeActionBrief, score, CompositionSpec, asset validation, and rendered preview. Both required fixture paths pass.
24-Case CanaryCovers both input modes, paid channels, objective variants, immigration-specific civic framing, positive Liberal opportunity framing, legal limits, malformed inputs, adversarial visible-copy leak cases, 50-point creative scoring, pass/fail language-style vocabulary, validation, and rendered-preview artifact checks. Latest deterministic run: 24/24 pass.
Local PersistenceSaves run records, generated artifacts, score, validation state, review boundary, and artifact paths under a durable local registry. Current registry has both required paths passing.
Creative GeneratorTurns a governed CampaignBrief into a CreativeActionBrief with campaign line, key visual, audience insight, evidence asset when needed, variants, platform specs, measurement, risk review, concept type, persuasion stage, frame contract, legal/publication classification, second-frame audit, messenger evidence card, medium-fit note, satire legal-risk note, and opponent clip test. It now routes immigration permission-frustration before positive Liberal offer language, so immigration-pressure briefs lead with “Why can’t we talk about it, Albo?” instead of generic relief copy. The same fixture then tests “More people. Fewer homes. Same excuses.” and “They told you not to talk. You paid the rent.” Positive Liberal offer output remains separate, testing “Let families breathe again.”, “Make work worth it again.”, and “A stronger country can help more people.” Current generated immigration fixture scores 47/50 with verdict pass.
Composition GeneratorTurns CreativeActionBrief into layered CompositionSpec JSON for the visual workbench. It generates background, evidence, headline, body, call-to-action, legal, export, and review layers; the generated fixture passes asset/composition validation.
Design Handoff PackageTurns a CompositionSpec into a Figma/Photoshop-ready local package: layer manifest, linked asset manifest, editable copy sheet, platform export brief, adapter notes, and review boundary.
GIMP Runtime ExporterTurns a CompositionSpec into a GIMP-ready package and now runs GIMP headlessly on Albert to place registered raster assets, produce real layered XCF files, and export full-canvas PNG drafts.
Figma Adapter PackageValidates Figma auth without embedding the token, then turns a CompositionSpec into a local development plugin package with manifest, code, UI, composition data, layer mapping, embedded image-fill payloads, run notes, and review boundary.
Creative Territory PackTurns CreativeActionBrief output into five art-direction territories with named first and second frames, symbolic bridge scores, messenger profiles, evidence class, medium recommendation, backfire risk, image prompt specs, negative constraints, platform crop notes, art-director review questions, and a prompt-review boundary before any image generation.
Google Image CandidateGenerates a governed Google Gemini image candidate from an approved territory prompt, registers it as a review-required generated asset with provenance and SHA-256, then places the raster into GIMP and Figma draft packages with publication still blocked.
Scoring HarnessScores audience truth, strategic tension, alignment to strategy, offer clarity, visual idea, verbal idea, emotional impact out of 10, testability, and symbolic bridge. Language style is pass/fail on approved voter-facing vocabulary, not a numeric score. Legal/publication classification is a separate law gate, not a creative score. The harness penalises jargon, long sentences, abstract phrases, missing concept type, missing persuasion stage, missing frame contract, positive-offer drift inside immigration permission frames, absent messenger evidence, absent medium fit, legally unsafe factual assertions, and variants that do not test one primary hypothesis. Creative-quality revisions are separated from law gates; political advertising limits are defined by law only, and inferences are not factual claims merely because they are sharp. Current deterministic gate is a floor check, not a frontier creative judge; independent review found the next upgrade must make symbolic bridge and verbal quality earned rather than template-filled.
Asset ValidatorChecks brand kit, rights, provenance, approved contexts, forbidden contexts, review status, and composition layer asset references.
Asset LibraryShows reusable brand tokens and asset registry items with rights, review state, approved contexts, forbidden contexts, and production readiness. Open assets.
Intake SurfaceAccepts Decision Engine handoff JSON or direct user intake and normalises both paths into a CampaignBrief preview. Open intake.
Review WorkbenchTurns CompositionSpec JSON into a browser-renderable draft with layer selection, local review persistence, editable comments, approve/revise/block state, and exportable review JSON before Figma, Photoshop, or deployment adapters. Open workbench.

Build plan

The project is broken into dependency-gated tasks. Current deployment covers the protected capability page, search entry, schemas, Gantt, task register, monitoring plan, creative standard, two-input intake, end-to-end runner, local run registry, 24-case canary, local asset importer, creative generator with concept type, persuasion stage, legal/publication classification, second-frame audit, messenger evidence, medium-fit, satire legal-risk, and opponent-clip gates, creative territory pack generator, governed Google image candidate generation, composition generator, design-handoff package generator, GIMP adapter package generator, Figma adapter package generator, scoring harness, asset validator, composition renderer, workbench, asset library, and export profiles. Live ad execution remains blocked pending explicit approval.

Workstream25 May26 May27 May28 May29 May30 May31 May1 Jun
Schemas and normalisationscopeschemasnormalise
Creative standard and templatesstandardconceptstemplatestemplates
Evaluation and measurementcanaryscorermeasurement
Adapters and reviewDE adapterintakereview gate
Deploymentvisualssearchreport