How the engine works

Same input, same output. Always reproducible. No black box.

01

Document intake

The engine accepts PDFs, images, document ZIPs and historical batches. Also via email, webhook, or form. Documents are associated with the account and enter the processing pipeline.
02

Structured extraction

Each document is analyzed and its fiscally relevant data is extracted in structured format: supplier, amount, VAT, date, invoice number. No free interpretation — only verifiable extraction.
03

Deterministic validation

The engine validates NIF/VAT with VIES, applies BCMath for tax calculations, and cross-references documents with configured rules. Each validation leaves a trace in the audit trail.
04

Chained calculation and rules

Parameterized depreciation, retirement provisions, accepted payroll and tax contributions are calculated with deterministic precision. Rules are applied in order and each step is recorded.
05

Status markers

Each document receives one of four statuses: accepted, accepted with rule, flagged or escalated. What's doubtful isn't hidden — it's separated and explained.
06

Export ready to work

The result: JSON with complete audit trail, CSV grootboek ready to import, UBL 2.1, ZIP with originals, and list of discrepancies and doubtful items. The final word is always the user's.

Four statuses, not two

The engine doesn't decide everything. It separates the verifiable from what requires judgment.

The engine verified and the document enters without objections.

A specific rule was applied. The result is valid but the reason is documented.

The engine detected something doubtful. The document remains visible and explained for review.

Requires explicit decision. The engine cannot resolve this alone — the user decides.

The sandbox shows the engine with example documents. No account, no real data.