Method · the protocol, written down

Several models argue. One person signs.

Every case on the main page follows the same short protocol. It exists so that a decision can be checked afterwards by someone who was not in the room: what was asked, what disagreed, what was verified, and who signed.

The rounds

  1. The brief is written before any model sees it.

    One question, the decision it serves, the deadline, and the evidence available. If the question cannot be written in three lines, the problem is upstream and the rounds wait.

  2. The same brief goes to several models.

    Four at the moment, from different makers. None of them is treated as a source of truth. What I want from them is the spread of answers, not the average.

  3. The disagreement is the material.

    Where the answers agree, I note it and move on. Where they disagree, I send the disagreement back with the evidence I hold, and ask each model to defend or drop its position. This is round two, and sometimes three.

  4. What survives is verified against the source.

    A number is kept only if it can be traced to a dataset, a document or a first-hand record. What I cannot verify is kept as a stated limit, visibly, never silently dropped.

  5. I sign the decision, with the trail attached.

    The signed decision carries the brief, the rounds, the limits and the date. The models advised. The decision is mine, and so is the accountability.

How evidence is labelled

Every figure on the main page carries one of four labels. The label is part of the claim.

Professional experience · first-hand

I did the work. Supporting material exists and can be discussed where confidentiality permits.

Public source

Open data or a published document, cited inside the engine or next to the figure.

Synthetic demonstration data

Invented for the purpose of showing a method. Never presented as a result.

Estimate

A model's or a period's estimate, labelled as such, with its basis stated on request. Not an audited total.

What I will not do

Present a model's answer as a finding. A model's answer is an input to a round, nothing more. Round a figure up for effect. The main page says "estimate" where it is one, and names the year of the model that produced it. Name a client or an institution that has not agreed to be named. Institutions on the main page are named by type; the full record is on LinkedIn and in the CV. Let a chart go out without its source and its unit.

"Not one model. A council I direct, and a decision I sign."

Oriol Cervantes · Decision Intelligence · this page is the citable version of the method

To cite: Cervantes, O. (2026). Method: a council of models, one signed decision. oriolcervantes.com/method. Version of 13 October 2026.