Skip to content

Methodology and Sources

This page exists for a simple reason: anyone reading a regulatory analysis needs to be able to retrace the path. Here is the path.

Where the material comes from

Everything that supports a text here is a public document. Nothing comes from behind-the-scenes conversations, privileged access or sources that cannot be cited.

The sources, in order of preference:

  1. The agency’s own documents: normative resolutions, tariff-setting resolutions, orders (despachos), technical notes, board votes, RIA reports and minutes of public board meetings. This is what the agency actually decided and the reasons it gave, not someone’s reading of it.
  2. The proceeding itself: public consultations and hearings, with the draft rule, the comments received and the report on those comments. This is where you can see who asked for what and what the agency accepted or rejected.
  3. Open data: public databases from ANEEL; from CCEE, the electricity trading chamber; from ONS, the national grid operator; from IBGE, the national statistics institute; and from federal open data portals. When a number appears in a text, the database and the extraction date appear with it.
  4. Laws and case law: laws, decrees, and court or Federal Court of Accounts (TCU) decisions when the point is contested.
  5. Academic literature: for conceptual grounding and international comparison.
  6. The press: useful for knowing that something happened and when. It is not proof of what happened: for that, we go back to the document.

What is done with the material

Every factual statement carries a link to its source where it appears. Not in a bibliography at the bottom that no one checks, but in the middle of the sentence, where the reader is.

Numbers come with a date and a source. “The tariff rose 12%” says nothing. “It rose 12% between the 2024 and 2025 adjustments, according to Resolution No. X” does.

Cited documents are read in full, not through the executive summary or a news story about them.

When sources disagree, the disagreement is shown, not silently resolved in favor of the one that supports the argument.

What is not known is stated. Where public data do not exist, or exist only in poor quality, the text says so instead of filling the gap with an estimate disguised as fact.

Separating description from opinion

A text here usually contains both, and they are kept apart:

  • Description: what the rule says, what the agency decided, what the data show. Verifiable. If it is wrong, it is an error and it is corrected.
  • Analysis: what this implies, what was left out, what incentive it creates. Arguable. People can disagree without anyone being wrong.
  • Position: what the author thinks should be done. It is marked as such and never blended into the description.

Use of artificial intelligence

AI tools are used in producing this site, and it is worth saying exactly what for.

They are used to: find documents in large databases, organize raw material, proofread, check internal consistency, draft English versions for the author’s review and build the site itself.

They are not used to: produce analysis, form positions, generate citations or summarize documents the author has not read.

No text is published until every statement has been checked against the original source by a person. Language models make mistakes with confidence, and they invent references as fluently as they cite real ones. That is why checking is always done against the document, never against a summary.

Responsibility for what is written lies entirely with the author who signs it.

How to cite a text from this site

Each article includes, at the end, a ready-made reference in ABNT format, the Brazilian citation standard, with the publication date and the permanent link.

When a text changes

Corrections are made in the text itself, with a dated note at the end saying what changed. Updates prompted by new facts are handled the same way. The original text is not deleted or silently rewritten.

Found an error?

contato@regulador.org, with the link to the text and, if possible, the document that shows the error. Proven corrections are made quickly, with credit to whoever reported them, if they wish.

In effect since September 18, 2026. When the method changes, this date will change too.