We Digitised the Document. We Never Fixed the Mental Model.

Customs modernisation has a framing problem.

The industry spent a decade digitising. We turned paper into PDFs, then PDFs into structured electronic formats. eBOL. eCMR. eInvoice.

As Jonathan Koh noted, the format has changed. The mental model hasn't.

We are still thinking in documents. We are still assembling data at the point of filing and presenting it as if filing is where the process starts.

It isn't. And that assumption is precisely why customs AI keeps producing results that need a human to verify them.

The Canon's Technical Architecture starts from a different premise entirely.

Data is declared at the point of origin — when the order is placed, when the goods are packed, when the logistics event occurs.

That data lives in a knowledge graph.

Regulatory outputs are generated from it when an obligation requires them.

They are derived on demand. They are not the source. They are never the source.

This is not a subtle distinction.

It is the difference between a system that reasons across declared facts and one that reads outputs assembled by hand.

When data relationships are declared structurally — not inferred from a set of outputs filed at the border — a risk determination can be made before goods arrive.

A shortage can be flagged before the goods leave.

A classification can be entailed from product attributes by formal rules, not guessed from document text by a model.

The Technical Architecture Explainer documents how that system is built and why every design decision was made the way it was.