Legal-entity dossier — upload and CNPJ anchoring
Send a company's registration dossier in a single batch and DATTA anchors every document to the right company by CNPJ, already assembling the ownership structure. The legal-entity (KYC/KYB) dossier enters through the same batch upload as every other context — there is no new flow to learn. The difference is the destination: instead of sitting loose, articles of association, powers of attorney and certificates all point to the same company, and each extracted ownership link carries the participation percentage on the graph edge itself.
Sending a dossier
- Create (or select) a context of type Dossiê Cadastral (KYC/KYB) in — the context identifier is the company's CNPJ.
- Open the upload screen already carrying the context and the target CNPJ:
screens/upload.html?domain=<contexto>&cnpjAlvo=<CNPJ>. - Send the dossier documents in a single batch: articles of association and amendments, CNPJ card, ultimate-beneficial-owner declaration, powers of attorney, certificates and balance sheets.
The cnpjAlvo is validated before the batch is accepted (14 digits + check digits) — a wrong CNPJ would anchor the whole dossier to the wrong company.
| Situation | How the document is anchored |
|---|---|
| With a target CNPJ provided | Every document in the batch is linked to that CNPJ's company. |
| Without a target CNPJ | The platform extracts the CNPJ from the document text (in formatted form, or 14 digits preceded by "CNPJ"), tie-breaking by the most frequent number. |
| With no identifiable CNPJ | The document is not linked to any company — and the absence is recorded, to become an "insufficient" screening criterion. It is never silent. |
What happens after upload
- Corporate extraction. The context prompt extracts companies, partners, declared beneficial owners, powers of attorney, certificates, addresses and accountants, and writes the corporate graph. The participation percentage is only stored when it is explicit in the document — the platform never estimates. A missing percentage shows as "unknown" in the analysis, not as zero.
- Identity reconciliation. At the end of the batch, companies with the same CNPJ and people with the same CPF are unified into a single node, so that spelling variations of the corporate name stop fragmenting the ownership chain. People without a CPF are never unified by name — homonyms stay separate on purpose, to avoid fabricating a false ultimate beneficial owner.
- Search and screening. The dossier's chunks become searchable filtered by CNPJ — the same mechanism the case number plays in legal contexts.
Downloading the full dossier
On the screens that list the dossier there is a button to download every PDF linked to the company in a single ZIP file. The package is built on demand and always reflects the current state: each new document sent becomes part of the next download. The same download is available to automations — see the API reference.
Known limitations
- A document with no identifiable CNPJ (neither the batch target nor extractable from the text) stays unanchored. It shows up in screening as a pending item and is never linked by guesswork.
- Powers of attorney citing more than one CNPJ use the batch's target CNPJ as the tie-breaker. With no target and no clear majority, they are not linked.
- A masked CPF (
***.123.456-**) never becomes a CPF property. Probabilistic matching with a score belongs to the KYB cross-linking, with a human decision.
See also
- KYB criteria — how the links extracted here are assessed.
- Legal-entity onboarding flow
- Upload and ingestion — the batch flow common to every context.