1. Ingestion
- We fetch the document from its official URL (BCB JSON API for norms, PDF or page for manuals).
- We compute a SHA-256 hash of the exact bytes we read.
- We store an immutable snapshot in our own storage (Cloudflare R2). BCB PDFs are overwritten at the same URL; our snapshot preserves what existed on each date.
- We extract text and metadata (title, number, version, validity, pages) into a relational corpus (Cloudflare D1).
Target: detect a change within 12 hours of the official file appearing. Detection runs twice a day; promotion of text and metadata remains curated when a snapshot has no label or validity dates.
2. Facts vs readings
- BCB fact — transcription or metadata extracted from the official document. Published as is, because it is faithful to the original.
- Guia Pix reading — plain-language explanation, written or AI-assisted and approved before publication. Always labelled as interpretation.
3. Citations
Every published statement carries the source: type of act (Resolution, Normative Instruction, Manual, Communiqué), number, manual version when applicable, page or section, the BCB link and the snapshot key. No citation, no publication.
4. Validity (as-of)
Normative pages let you pick a date and show the version in force on that date — including what becomes effective later (for example, the UX manual 7.4 on 01/03/2027). Dates always come from the official document.
5. Corrections
Source, validity or wording errors are treated as priority: we record the correction, cite the correct document and date the change. Write to our contact. If anything here diverges from the official document, the official document prevails:bcb.gov.br/estabilidadefinanceira/pix-normas.