Architecture

For detailed data specifications and loading implementation, see DATAS.md in the root directory.

Overview

dure.one uses a progressive loading architecture with postcode-based organization for efficient data delivery.

Data Pipeline

3-Stage Generation

Stage 1: Static Classifications (Build Time)

• Business categories: KSIC, Hometax, Sanjae, Goyong (6 systems)
• Product categories: Google taxonomy
• Postcode data: UTM-K coordinates for 34,363 postcodes

Stage 2: Business Data Processing

• Google Sheets → Rust validation → JSONL output
• Organized by postcode and category
• Shard metadata embedded in records

Stage 3: Aggregation

• Copy all outputs to public/data/
• Ready for static deployment

Frontend Architecture

Progressive Loading

Design Principles:

• Postcode-centric: Organize by 5-digit Korean postcodes
• Distance-based: Load nearest postcodes first
• On-demand: Fetch JSONL only when needed
• Cached: Keep loaded data in memory

Key Components

dataLoader.js - Main orchestration

• Loads indices at startup
• Coordinates progressive loading

postcodeLoader.js - Postcode-based loading

• Fetches /data/postcodes/{postcode}.jsonl
• Caches loaded postcodes
• Deduplicates by bizno

categoryLoader.js - Category filtering

• Uses category_index for relevance
• Delegates to postcodeLoader

postcodeDistance.js - Distance calculations

• Haversine distance between postcodes
• Cached distance matrix

Embedded Shard Optimization

Each business record includes _shard metadata for direct detail view loading:

{
  "bizno": "1234567890",
  "name": "Example Business",
  "_shard": {
    "category": "HT11000",
    "shard_id": 18
  }
}

Benefits:

• Eliminated 2.9MB search_index.json
• Eliminated 859KB coordinate_index.json
• O(1) shard lookup from record
• Faster detail view rendering

Data Flow

User selects postcode
  ↓
postcodeLoader fetches /data/postcodes/{postcode}.jsonl
  ↓
Business list rendered with _shard metadata
  ↓
User clicks business
  ↓
Direct shard fetch: /data/shards/biz/{category}/shard_{id}.jsonl
  ↓
Detail view rendered

File Organization

public/data/
├── KR_KSIC11TH.json          # Classification metadata
├── KR_HTCODE26.json
├── KR_SJCODE.json
├── KR_GYCODE.json
├── KR_prdcat.json
├── postcode_index.json        # Dropdown data
├── index/
│   ├── postcode_index.json    # Postcode metadata
│   └── category_index.json    # Category distribution
├── postcodes/
│   └── {postcode}.jsonl       # Business records by postcode
└── shards/
    ├── biz/{category}/shard_*.jsonl
    └── prd/{category}/shard_*.jsonl

Performance

Bundle Size:

• JS: ~8KB (~3.4KB gzipped)
• CSS: ~10KB (~2.7KB gzipped)
• Total: ~6KB gzipped

Loading Strategy:

• Indices: ~8MB loaded at startup
• Business data: Loaded on-demand per postcode
• Detail data: Loaded on-click per business

Caching:

• Postcode data cached in memory
• Distance calculations cached
• No duplicate fetches