InTake:back-office, without blind spots.

A container doesn't cross because it exists. It crosses because a file exists that says what's inside, where it came from, what it's worth, who sold it, who bought it, under which regime it enters and which rules it meets. The goods are the consequence; the file is the transaction.

No single person assembles that file. It's assembled by a supplier in another time zone, a carrier, a customs broker, a trade compliance team and, when something jams, a lawyer. Each sends their piece by email, in whatever format they happened to have, on whatever day they managed.

Mexico runs one of the largest foreign trade operations in the world, and it runs it document by document on an infrastructure whose last mile is still human: someone reads, someone types, someone cross-checks.

That last mile is the product. And it still hasn't been built.

Pedimentos, invoices, packing lists and certificates of origin land in different threads, at different times, from different senders. Someone has to recognize that those five loose files are a single operation, and put it back together by hand.

Then the same figures are keyed from the PDF into the system, and checked against the PDF again. It's the same work done twice, and the second pass exists only because nobody trusts the first. One wrong digit travels all the way to the pedimento with nothing to stop it.

Meanwhile LIGIE codes, NOMs, permits and countervailing duties change in the DOF. If nobody read it that week, the operation clears under a rule that no longer exists.

The work that protects you from a fine is exactly the work nobody gets to finish.

The first: the authority went digital. VUCEM, the electronic file and the transmission of the Manifestación de Valor over a web service moved the process from a folder to a receipt. What used to be filed away is now transmitted and recorded.

The second: models can finally read a bad document. A pedimento scanned crooked, an invoice in another language, a packing list annotated by hand. Until recently that required a template per supplier; it doesn't anymore.

The third: volume isn't going down. Supply chains relocating to Mexico add operations, and every operation adds files. The bottleneck stops being the border and becomes a team's capacity to review paperwork.

None of the three is enough on its own. Together they make possible something that wasn't five years ago.

A general model can read a pedimento. What it doesn't know is that one tariff code can carry several entries in Anexo 2.2.1 depending on the scenario, or that the DTA on a return isn't calculated like the DTA on a definitive import, or that a certificate of origin only changes what's owed if it meets the specific rule for that code.

That isn't improvised: it's catalogued. InTake was built on the real catalogues — typed, with their effective dates — not on an approximate description of them.

On top of those catalogues run three operators that hand work to each other over the same operation instead of living in three separate tools: SCAN extracts, AUDIT checks against the rules and quantifies fiscal exposure, SCOUT classifies the tariff code and puts its reasoning on the record.

And the checks aren't a model guessing: they're deterministic rules. When the answer has to be the same twice, it is.

A system that runs regulated filings doesn't earn trust by promising it never gets anything wrong. It earns it by showing what it leaned on, where it hesitates, and what it puts on the record.