Data source policy
Every public page should be traceable back to structured fields in the database. LLM text may organize the language, but numbers, dates, HS codes, and source labels must come from input data.
The planned production pipeline can use bill-of-lading datasets, fair exhibitor directories, registry signals, supplier websites, and manual verification notes. The current repository contains seed/mock records for build validation only.
Pages currently label the dataset as mock ETL output. Buyers should not treat mock shipment rows as live customs records until production ingestion and validation scripts replace the seed database.
Generated content must pass schema checks and numeric back-checks against the input data. If a field is missing or unverified, the site should omit that claim rather than invent a substitute.