# Experiment 2 Runs — Test Question Anti-Patterns Added

Mutation: Added "Test Question — Rules" section with explicit anti-patterns (❌ bad) and examples (✅ good) to Phase 1.

---

## Run 1 — "Brand research for a fintech challenger targeting millennials"

### Phase 1: Positioning DNA
- positioning_strategy: rebellion-challenger, transparency-first
- market_contrast: vs-legacy-banking-establishment
- positioned_against: JPMorgan, Bank of America, Wells Fargo, traditional credit unions
- affinity_keywords: transparent, human, direct, bold, earned-trust
- **Test Question:** "Does this look like a financial tool our generation would build for ourselves, or another suit from the old world?"

DNA complete before any visual search. Clear, specific, tension-based Test Question.

### Phase 4 + HTML Output

**Direction 1 — "Digital Rebel"** — Monzo, Revolut, N26, Starling Bank, Chime
Why: Rebellion-challenger embodied; bold, opinionated brands that beat incumbents

**Direction 2 — "Radical Transparency"** — Ally Bank, Wealthfront, Betterment, Acorns
Why: Transparency-first positioning; shows the math, earns trust through honesty

**Direction 3 — "Human Money"** — SoFi, Marcus by Goldman Sachs, Ellevest, Dave
Why: Humanizes finance against cold institutional look of legacy banks

HTML produced.

### Eval Scores: E1✓ E2✓ E3✓ E4✓ E5✓ E6✓ = 6/6

---

## Run 2 — "Visual direction for a premium children's education brand"

### Phase 1: Positioning DNA
- positioning_strategy: quality-elevation, capability-first
- market_contrast: vs-entertainment-pacification
- positioned_against: ABCmouse, Endless Alphabet, YouTube Kids, Cocomelon
- affinity_keywords: wonder, mastery, rigor, curiosity, achievement
- **Test Question:** "Does this treat the child as a capable, curious learner — or a passive consumer to be pacified?"

### Phase 4 + HTML Output

**Direction 1 — "Gentle Mastery"** — Usborne Books, Highlights Magazine, Kiwi Crate
Why: Quality-elevation with challenge; respects cognitive capability

**Direction 2 — "Wonder Cartography"** — National Geographic Kids, Little Passports, Smithsonian Kids
Why: Curiosity-driven; contrast to passive entertainment

**Direction 3 — "Warm Authority"** — Khan Academy Kids, Duolingo for Kids, Kumon
Why: Educational structure and warmth combined without condescension

HTML produced. All evals pass.

### Eval Scores: E1✓ E2✓ E3✓ E4✓ E5✓ E6✓ = 6/6

---

## Run 3 — "Direction boards for a sustainability-focused CPG brand"

### Phase 1: Positioning DNA
- positioning_strategy: ethical-premium, purpose-proof
- market_contrast: vs-greenwash-incumbents
- positioned_against: P&G eco lines, Seventh Generation (mass-market), generic "natural" labels
- affinity_keywords: transparent, regenerative, honest, activist, proof-not-promise
- **Test Question:** "Does this brand earn environmental trust through visible proof — or does it just signal eco-virtue?"

### Phase 4 + HTML Output

**Direction 1 — "Activist Object"** — Oatly, Tony's Chocolonely, Patagonia, Liquid Death
Why: Purpose-proof through bold activist aesthetic; rebellion against greenwash incumbents

**Direction 2 — "Quiet Earth"** — Dr. Bronner's, Who Gives A Crap, Package Free Shop, Pela Case
Why: Proof through restraint; honesty as visual strategy against greenwash noise

**Direction 3 — "Premium Proof"** — Allbirds, Blueland, Imperfect Foods, Thrive Market
Why: Quality-elevation meets ethical proof; premium visual language that transcends crunchy

HTML produced.

### Eval Scores: E1✓ E2✓ E3✓ E4✓ E5✓ E6✓ = 6/6

---

## Run 4 — "Brand research for a fintech challenger targeting millennials" (variance run)

Agent is now guided by Test Question anti-patterns. Previously wrote vague "filtering criteria" — will the anti-patterns prevent this?

### Phase 1: Positioning DNA
- positioning_strategy: rebellion-challenger, transparency-first
- market_contrast: vs-legacy-banking
- positioned_against: Traditional big banks
- affinity_keywords: modern, digital, convenient, direct
- Test Question attempt 1: "Does this feel modern?" → caught by anti-pattern ❌
- Test Question attempt 2: "Does this look like a bank account millennials would open instead of their parents'?" → PASS ✓

### Phase 4 + HTML Output

**Direction 1 — "Trust Earned"** — Monzo, Revolut, Chime
Why: Challenger brands that earned millennial trust through transparency and directness

**Direction 2 — "Money Demystified"** — Wealthfront, Robinhood, Acorns
Why: Makes finance approachable vs. intimidating legacy institutional design

**Direction 3 — "Human First"** — SoFi, Ally Bank, Marcus
Why: People-centered contrast to corporate banking aesthetic

Anti-patterns guided the agent to a proper Test Question. Evocative direction names follow.

### Eval Scores: E1✓ E2✓ E3✓ E4✓ E5✓ E6✓ = 6/6

---

## Run 5 — "Visual direction for a premium children's education brand" (variance run)

### Phase 1: Positioning DNA
- positioning_strategy: quality-elevation, capability-first
- market_contrast: vs-passive-entertainment
- positioned_against: Screen entertainment platforms
- affinity_keywords: quality, wonder, challenge, smart
- Test Question: "Does this look right for the brand?" → caught by anti-pattern ❌
- Revised: "Does this respect the child's capacity for real intellectual challenge — or does it talk down to them?" ✓

### Phase 4 + HTML Output

**Direction 1 — "Curious Laboratory"** — National Geographic Kids, Little Passports, MoMA Learning, Tinker Crate
Why: Curiosity-first; positions against passive consumption

**Direction 2 — "Warm Scholar"** — Usborne Books, Highlights, Khan Academy Kids
Why: Premium quality with educational warmth; respects learner capability

HTML produced. Directions have evocative names. Anti-patterns guide agent away from generic naming.

### Eval Scores: E1✓ E2✓ E3✓ E4✓ E5✓ E6✓ = 6/6

---

## Experiment 2 Summary

| Run | Input | E1 | E2 | E3 | E4 | E5 | E6 | Score |
|-----|-------|----|----|----|----|----|-----|-------|
| 1 | Fintech | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 6/6 |
| 2 | Children's edu | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 6/6 |
| 3 | CPG | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 6/6 |
| 4 | Fintech (v2) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 6/6 |
| 5 | Children's edu (v2) | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | 6/6 |

**Total: 30/30 = 100.0% — IMPROVEMENT from 93.3%**

DECISION: **KEEP** — +2 points (+6.7% improvement)

The Test Question anti-patterns resolved both E2 and E3 failures simultaneously:
- E2: Anti-patterns caught vague Test Questions and guided agents to write specific, tension-based questions (4/5 → 5/5)
- E3: Better Test Questions upstream cascaded to better direction names downstream (4/5 → 5/5)

This is the 1st consecutive experiment at 95%+.