Corpus source rules
The corpus is the moat. It needs evidence, confidence tiers, and correction loops before it can safely power consumer recommendations, retailer widgets, and enterprise data products.
Source tiers
| Tier | Meaning | Usage |
|---|---|---|
| Curated | Official brand, distributor, licensed dataset, or reviewed import | Can power public and B2B results |
| Community | User or contributor supplied with review trail | Can assist, but show lower confidence |
| Limited | Sparse, ambiguous, or weak source | Can appear with caveat; should not be definitive |
Source rules
- Store source URL, source type, retrieval date, and parser/import path.
- Preserve original text where licensing allows; otherwise store structured facts and attribution.
- Never scrape blocked or explicitly disallowed sources.
- Use official brand and retailer pages where possible.
- Corrections should create audit history, not overwrite silently.
- Country packs need importer templates and canonical brand identity rules.
Country showcase plan
Australia is the reference pack. Future country packs should repeat the same shape:
- country overview
- approved houses
- official maker links
- available fragrance rows
- guided lenses
- MCP country read tools
- brand claim and correction intake
- affiliate-safe click tracking