Database accuracy
Database accuracy is the difference between a fun app and a sellable fragrance intelligence layer.
Accuracy domains
- Brand canonicalisation.
- Fragrance canonicalisation.
- Concentration and variant modelling.
- Flanker relationships.
- Reformulation and year handling.
- Perfumer joins.
- Note joins.
- Accord and family taxonomy.
- Country and distributor aliases.
- Image and thumbnail metadata.
- Confidence tier and evidence history.
Existing foundation
The repo already has a note-image and fragrance-note migration prepared under
supabase/migrations/20260525160000_note_images_and_fragrance_notes.sql.
It defines note thumbnail metadata fields and a structured
public.fragrance_notes join table.
That is a foundation for the visual note atlas and note-aware matching. Any live migration or RLS change still needs explicit approval before application.
QA queue
Every data correction should track:
- entity type
- old value
- proposed value
- evidence
- proposer
- reviewer
- confidence change
- downstream affected rows
- timestamp
This creates a corpus that brands and retailers can trust.