Editorial policy

We publish at most one to three articles a day, each with its own angle. We do not build pages for keyword variations and we do not rewrite press releases.

This limit is written into the panel where articles are entered, in plain sight of whoever publishes. It is a rule of ours, not a lock in the system: the real locks are the ones below.

No AI-assisted content is published without an identified human reviewer, and that does not depend on anyone remembering. When an article is saved as published, the system looks up the named reviewer, checks that it is a human account with editing rights, and writes that name inside the article.

If the field is empty, or points at the automated account, the write is rejected and the article does not become published.

An article declared as written by a person does not go through that gate: there, the one answering for it is the author, named and dated above the text.

The content agent connects with an API key and has no access to the panel. It writes drafts only: if it tries to publish, it gets an error; if it saves without marking a draft, the system pins the write to draft anyway.

With that same key it only reads what is already published — no drafts, no version history, no unpublished client cases — and it cannot write the name of its own reviewer.

Each of those paths has an automated check that runs on every change proposed to the site. If one of them opened, the check fails and the change is not published.

Every article states how it was produced, including when a person wrote it: written by a human, drafted with AI or translated with AI. The notice sits inside the article, at the foot of the text, not on a separate page.

Where AI assisted, that same notice carries the signature of the reviewer responsible and the model used.

The panel stores other details of the same production — the prompt version, how much the machine's draft was edited — which are an internal record and are not published.

Cryptoassets, regulation and taxes carry a mark of their own in the system, and that mark exists for one purpose: to keep them out of any automated publishing pipeline.

There is no such pipeline today: everything published is published by a person, from the panel. The mark has been in place since the first data load, before the pipeline exists — writing it afterwards, with the pipeline running, would be too late.

Provenance is declared once per article, not once per language. Today we cannot state, within the same piece, that a text was written by a person in Spanish and translated into Portuguese with AI.

While the field works this way, we do not publish mixed articles. If that ever becomes necessary, we change the field before publishing — not the wording of this page.

When there is an error, the correction is written into the article itself, with a date. There is no field that does it on its own: a person writes it.

The panel keeps a version history, but that history is internal; the public record is the corrected article. If you find an error, write to us from the contact page.