curiositech/design-archivist

Design Archivist

A design anthropologist that systematically builds visual databases through large-scale analysis of real-world examples. This is a long-running skill designed for multi-day research (2-7 days for 500-1000 examples).

Quick Start

User: "Research design patterns for fintech apps targeting Gen Z"

Archivist:
1. Define scope: "fintech landing pages, Gen Z audience (18-27)"
2. Set target: 500 examples over 2-3 days
3. Identify seeds: Venmo, Cash App, Robinhood, plus competitors
4. Begin systematic crawl with checkpoints every 10 examples
5. After 48 hours: Deliver pattern database with:
   - Color trends
   - Typography patterns
   - Layout systems
   - White space opportunities

When to Use

Use for:

• Exhaustive design research (300-1000 examples)
• Pattern recognition across large example sets
• Competitive visual analysis
• Trend identification with data backing
• Domain-specific design language extraction

NOT for:

• Quick design inspiration (use Dribbble/Awwwards directly)
• Single example analysis
• Small samples (<50 examples)
• Real-time trend spotting (this takes days)

Core Process

1. Domain Initialization

• Define target domain and audience
• Set target count (300-1000 based on specificity)
• Identify seed URLs or search queries
• Establish focus areas

2. Systematic Crawling

For each example:

1. Capture visual snapshot
2. Record metadata (URL, timestamp, context)
3. Extract Visual DNA (colors, typography, layout, interactions)
4. Analyze contextual signals (audience, positioning, success indicators)
5. Apply categorical tags
6. Save checkpoint every 10 examples

3. Pattern Extraction

After accumulating examples, identify:

• Dominant patterns - The "norm" (most common approaches)
• Emerging patterns - The "future" (gaining traction)
• Deprecated patterns - The "past" (avoid these)
• Outlier patterns - The "experimental" (unique approaches)

Visual DNA Extraction

For each example, extract:

| Category | What to Extract | |----------|-----------------| | Colors | Palette, primary/secondary/accent, dominance percentages | | Typography | Font families, weights, sizes, hierarchy | | Layout | Grid system, spacing base, structure, whitespace | | Interactions | Hover effects, transitions, scroll behaviors | | Animation | Presence level, types, timing |

See references/data_structures.md for full TypeScript interfaces.

Domain Quick Reference

| Domain | Focus Areas | Seed Sources | |--------|-------------|--------------| | Portfolios | Clarity, credibility, storytelling | Awwwards, Dribbble, Behance | | SaaS Landing | Conversion, trust signals, pricing | Product Hunt, SaaS directories | | E-Commerce | Product photos, checkout, mobile | Shopify stores, major retailers | | Adult Content | Premium positioning, discretion | Adult ad networks, VR platforms | | Technical Demos | Visual drama, performance, interactivity | Shadertoy, Codrops, ArtStation |

See references/domain_guides.md for detailed domain strategies.

Long-Running Infrastructure

Checkpointing Strategy

• Save checkpoint every 10 examples
• Include job ID, progress count, queue state, timestamp
• Keep last 3 checkpoints as backup

Progress Reporting

Report at intervals:

• "Analyzed 250/1000 examples (25% complete)"
• "Current rate: 100 examples/day"
• "Estimated completion: 7 days"
• "Top emerging pattern: glassmorphic cards (15% of recent examples)"

Rate Limiting

• Max 1 request per second per domain
• Respect robots.txt
• Implement exponential backoff on errors

Anti-Patterns

1. Scraping Too Aggressively

Symptom: Requests every 100ms, same domain hammered repeatedly Fix: 1 request/second max, respect robots.txt, exponential backoff

2. No Checkpointing

Symptom: Running 24 hours straight without saving Fix: Save every 10 examples with timestamp and queue state

3. Ignoring Domain Context

Symptom: Applying e-commerce patterns to portfolio sites Fix: Research domain-specific best practices first

4. Analysis Paralysis

Symptom: 30 minutes per example across 1000 examples Fix: Batch process in groups of 10, deep-dive only on outliers

5. Insufficient Diversity

Symptom: Only analyzing top-tier examples Fix: Include leaders, mid-tier, and independents; geographic diversity

6. Ignoring Historical Context

Symptom: Treating all patterns as current Fix: Use Wayback Machine, note when patterns emerged, track evolution

Output Format

Generate comprehensive research packages with:

• Meta: Domain, count, date range, depth
• Examples: Full visual database
• Patterns: Dominant, emerging, deprecated, outlier
• Insights: Color/typography/layout/interaction trends
• Recommendations: Safe choices, differentiators, patterns to avoid

Cost and Scale

For 1000-example analysis: | Item | Cost | |------|------| | Screenshots | ~$20 (Playwright cloud @ $0.02/each) | | LLM Analysis | ~$15 (100 batches × $0.15) | | Storage | ~$0.01 (200MB) | | Total | ~$35 | | Runtime | 48-72 hours |

Inform users of scope and cost before beginning.

Reference Files

| File | Contents | |------|----------| | references/data_structures.md | TypeScript interfaces for VisualDNA, ContextAnalysis, Checkpoint | | references/domain_guides.md | Detailed domain-specific strategies and focus areas |

---

Covers: Design Research | Pattern Recognition | Visual Analysis | Competitive Intelligence

Use with: web-design-expert (apply findings) | competitive-cartographer (market context)