# CE Brand Pipeline — Living Spec
**Last updated: 2026-03-24 | Status: Definition phase**
**Discord channel:** #brand-pipeline-spec (1485891369024094283)

---

## WHAT THIS IS
A **human-in-the-loop brand production suite**. Reduces the human time required to produce a complete, CE-quality brand from weeks of work to **6 structured decision moments totalling under 2 hours**. The human is the taste layer, positioned at the moments where taste genuinely belongs. The system handles everything else.

## WHAT THIS IS NOT
- Not an autonomous brand generator
- Not a replacement for CE's creative judgment — it compresses the time that judgment requires
- Not a product someone else can run without Assaf yet (Phase 4, conditional on calibration data)

## THE HONEST CEILING
> "This doesn't give AI taste. It gives AI a model of CE's taste, trained on real decisions over time."

---

## THE 8 TOOLS + 6 HUMAN TOUCHPOINTS

```
TOOL                   RUNS AUTONOMOUSLY              HUMAN TOUCHPOINT
────────────────────────────────────────────────────────────────────────
01. Brief Enricher  →  research + gap-fill         →  T1: Review enriched brief
02. Research Engine →  market / comp / audience    →  (feeds T1, no extra gate)
03. Strategy Maker  →  3 positioning options       →  T2: Pick one + say why
04. Territory Finder→  3 visual directions         →  T3: Pick one + say why
05. Style Forge     →  style guide + JSON          →  T4: Approve or adjust
06. Asset Generator →  image batches               →  T5: Select the cuts
07. Deck Builder    →  Google Slides deck          →  T6: Review narrative flow
08. Page + Social   →  web page + posts            →  T6: Review before publish
```

**Time at each touchpoint:** 10-15 min max
**Total human time per brand:** under 2 hours
**Total elapsed time:** 3-4 days

---

## TOOL DEFINITIONS

### Tool 01 — Brief Enricher
- **Input:** anything — even one sentence
- **Does:** fills gaps through research. Sector context, category norms, competitors, audience signals
- **T1:** Read enrichment. Correct what's wrong. Add what research couldn't know. 10 min.
- **Output:** validated brief

### Tool 02 — Research Engine *(runs inside Tool 01)*
- Julia runs market/competitive/audience research. Output populates brief automatically.

### Tool 03 — Strategy Maker
- **Input:** enriched brief
- **Does:** 3 positioning directions. Each: named angle + 3 reasons viable + what it rules out
- **T2:** Pick one. Write one sentence why. Logged.
- **Output:** positioning.json

### Tool 04 — Territory Finder
- **Input:** positioning.json + taste board query
- **Does:** queries signal library for what's emerging in this category. 3 named visual territories — each with: typography pair, color temperature, photography style, 3 real brand references, rationale
- **T3:** Sees 3 territories as visual cards. Picks one. Can flag elements from others. Writes one sentence.
- **⚠️ THE TASTE GATE.** Everything downstream follows from this pick.
- **Output:** visual-direction.json

### Tool 05 — Style Forge
- **Input:** visual-direction.json
- **Does:** produces VISUAL-STYLE.md + visual-style.json + Gemini prompt framework. Scores itself against 3 taste board references.
- **T4:** Review against references. One correction maximum.
- **Output:** visual-style.json (locked)

### Tool 06 — Asset Generator
- **Input:** visual-style.json
- **Does:** generates image batches. Vision model criteria-check before human sees any image.
- **T5:** Selection pass from pre-filtered batch. Picks 3-5 per use case.
- **Output:** asset-library/

### Tool 07 — Deck Builder
- **Input:** positioning + style guide + asset library
- **Does:** Google Slides deck. CE design system applied.
- **T6 (combined):** Review deck + page + social in one pass.

### Tool 08 — Page + Social
- **Input:** visual-style.json + research signals
- **Does:** web page + social posts

---

## SHARED DATA MODEL

```
brand-brief/
  ├── intake/           ← what the client provided
  ├── enriched/         ← what research added
  ├── positioning/      ← chosen strategic direction + rationale
  ├── visual-direction/ ← chosen design territory + rationale
  ├── style-guide/      ← visual-style.json (locked)
  ├── asset-library/    ← curated image set
  ├── deliverables/     ← deck, page, social
  └── decisions/        ← log of every human choice + context [CALIBRATION CORPUS]
```

---

## TASTE LAYER

| What taste does | Where | Tool |
|----------------|-------|------|
| "This positioning is more CE" | T2 | Strategy Maker — 3 options side by side |
| "This visual world is right" | T3 | Territory Finder — visual cards, not text |
| "These criteria need adjusting" | T4 | Style Forge — one correction |
| "This image is CE / this isn't" | T5 | Asset Generator — selection interface |

---

## TASTE BOARD AS INFRASTRUCTURE

Each entry tagged:
```json
{
  "id": "tb-0041",
  "image_url": "...",
  "added_by": "assaf",
  "added_date": "2026-03-15",
  "category": ["food", "premium-consumer"],
  "timing": "emerging",
  "design_signals": {
    "color_temperature": "warm-gold",
    "typography_weight": "heavy",
    "photography_style": "macro-texture",
    "mood": "editorial-luxury"
  },
  "ce_resonance": "high",
  "notes": "Aesop meets artisan dairy"
}
```

Tool 04 queries: `category: [X] AND timing: emerging AND ce_resonance: high`
CE resonance field stays human — Assaf curates the board. That curation IS the taste input.

---

## VISION QA (Eyes Workaround)

At Stage 5, vision model evaluates images against style guide criteria before human sees them.
- Not "does this look good?" → "does this exhibit [specific characteristics]?"
- Score = % criteria met. Below threshold → loop back with specific critique.
- This is a floor, not a ceiling. Removes garbage. Taste lives at the gate.

---

## BUILD PLAN

### PHASE 0 — Design the foundations *(~1 week, no code)*
- [ ] Design visual-style.json schema (validate against Phat)
- [ ] Design taste board tagging schema
- [ ] Map Phat signal-to-decision chain forensically
- [ ] Write "good brief" definition

**Gate: Don't start Sprint 1 until all four exist.**

### SPRINT 1 — Harden connective tissue *(2-3 weeks)*
- [ ] Build visual-style.json into pipeline
- [ ] Write visual-territories SKILL.md
- [ ] Tag 50+ taste board entries
- [ ] Build Stage 0 brief intake
- [ ] Test: semi-cold brief (not Mission)

### SPRINT 2 — Quality layer *(3 weeks)*
- [ ] Signal-to-decision framework as skill
- [ ] Vision model QA at Stage 5
- [ ] Decision capture protocol
- [ ] Stage 4 style guide gate
- [ ] Test: second cold brief. Assaf + Anton both score. Calibration.

### SPRINT 3 — Full pipeline hardened *(2 weeks)*
- [ ] Deck Builder hardened
- [ ] Page + Social connected through visual-style.json
- [ ] Taste board live (automated signal ingestion)
- [ ] Full end-to-end cold brief test
- [ ] Calibration review

### PHASE 4 — Productize *(conditional on Sprint 3 stable)*
- [ ] Brief intake UI
- [ ] Progress dashboard
- [ ] Deliverable packaging
- [ ] Pricing/packaging model
- **Gate:** "Quality without Assaf" only after 10+ runs with calibration validation

---

## OPEN DECISIONS

| # | Decision | Status |
|---|----------|--------|
| 1 | What makes a brief good vs. complete? | ⬜ OPEN — Phase 0 |
| 2 | visual-style.json schema fields | ⬜ OPEN — Phase 0 |
| 3 | Taste board tagging vocabulary | ⬜ OPEN — Phase 0 |
| 4 | Which semi-cold brief for Sprint 1 test? | ⬜ OPEN — Assaf |
| 5 | Phase 4 product price/packaging | ⬜ OPEN — Later |
| 6 | Who operates this besides Assaf? | ⬜ OPEN — Later |

---

## CONTEXT LOG
*(Append decisions and key moments as we build)*

**2026-03-24** — Product definition session in #product-definition. Three rounds of Anton critique on pipeline thesis, development plan, and revised plan. Decisions locked:
- Human-in-the-loop product, not autonomous
- Minimal intervention = time saved + decision optimization
- Quality ownership = Assaf for now, calibration path toward Phase 4
- Taste board = reference signal infrastructure, not mood board
- Phase 4 conditional on calibration data — not promised to clients