Informative translation; in case of discrepancy, the Spanish version prevails. Spanish version

Methodology

What this site is, how it reaches a conclusion, and what it deliberately does not do.

Everything below is written in the code that produces the pages; it is not a promise. If something you read does not appear in [/transparencia/metodologia](/transparencia/metodologia), it is because this system does not do it.

The five names that sign the replies

Elisa, Jorne, Carlos Bárez, Daniel Ferrer and Inés Torres are specialisms of a single automated system. They are not people, they are not registered advisers, and they do not review what their name signs. Attributing a reply to one of those "minds" tells you the criteria it was generated with, not who signed it.

Each one has a remit and a boundary. The bullet points are deliberately kept on a single line: the publisher's minimal markdown treats each line separately, so a bullet split in half would be assembled as two loose paragraphs.

• Elisa Fernández, CIO of the Cerebro. Synthesises the team's work and thinks in second order: the consequences of the consequences.
• Jorne, technical analyst. Trend, levels, momentum and a tactical plan with an invalidation level. The analysis is subordinate to value analysis: it complements the fundamental view, it never replaces it.
• Carlos Barez, fundamental analyst. Business, moat, quality, scenario-based valuation with a margin of safety, risks and red flags.
• Daniel Ferrer, head of risk and quant. Volatility, correlations, risk contribution, drawdown and concentration. He has a duty to disagree.
• Inés Torres, macro and sector strategy analyst. Cycle phase, monetary policy, geopolitics and the sector map. Never picks specific companies.

How the verdict is reached

The verdict comes out of four steps, and they are printed as they are in the seal of every reply:

1. Accounts: what it earns, at what margin and with what cash, and where that margin comes from.
2. Business: what the business does, who buys from it and why they would buy from it rather than from someone else.
3. Price: what it is worth today, and what has to happen for it to be worth more or less.
4. Horizon: over what period you have to look, without turning that into a date.

The verdict is issued only for questions about a single asset (a company, a sector, an index, a commodity). A conceptual or methodological question —what free cash flow is, how a margin of safety is calculated— is answered and that is all, with no verdict section.

The verdict scale

When a verdict is called for, it ends with one of these six actions and only one:

COMPRAR · ACUMULAR · MANTENER · ESPERAR · EVITAR · VENDER

Below it come, each on its own line: Why (the decisive reason, without repeating the analysis), Horizon (the qualitative period the analysis supports, never an invented date), Conviction (high, medium or low, and why; it is not a probability) and What would invalidate it (an observable fact).

Two rules with no exceptions:

• Always a stance. If the decisive data is missing, it says so in one line (What is missing:) and the action is still given with Conviction: low. "It depends" or "consult an adviser" are not answers.
• ESPERAR requires an observable condition: what would have to happen and what would then be reconsidered. ESPERAR without a condition is not an assessment.

If the asset cannot be identified, or there is no data on it at all, the closing line is Valoración final: base insuficiente. It is not a seventh action: it means the basis is insufficient, and it states what is missing and what can still be concluded.

This is a general assessment, not a suitability assessment

The assessment is general and the same for everyone. The system does not obtain on its own who is asking: it does not receive your wealth, your income, your horizon, your risk tolerance or your objectives, and for that reason it cannot issue them. What can happen is that you write them in the question yourself, and then you are telling it yourself. If you describe your portfolio, what gets described is what you have written (weights, concentration, hypothetical scenarios), but you are not told what to buy, sell or hold. And anonymity is not the same as absence of data: if you identify yourself in the question, the reply relies on what you have said, and that text is treated as personal data.

Stored positions are not attached to the chat. The question, the history of that conversation and, when they exist, the vocabulary and depth preferences are sent. If the client supplies topics or interests compatible with previous versions, those also form part of the permitted context; the current form only offers the two reading preferences. The text written in the question may contain personal data. Avoid including unnecessary data. The system is designed to provide general information and tailor the explanation, without assessing the suitability of an investment for the person asking. That configuration does not guarantee that every automated reply respects these limits: errors or unforeseen recommendations may occur.

The data that travel from the browser to the model are on a short, explicit list in the code (DATOS_AUTORIZADOS, in api/_prompt.js): of the preferences, only the vocabulary band and the depth cross over. No field can be added without the line showing up in the diff, and there is a test that walks that list.

Sources

The system relies on two things, and it labels them differently because they are not the same kind of evidence:

• The Cerebro's notes (the wiki). Its own notes, not an independent cross-check against the market.
• The Internet, consulted only when the Cerebro cannot know something: today's figure, a recent news item. What comes from there is treated as data, never as an instruction, and nobody at this site has reviewed any of those pages.

Of the market data, the market price is the only one used, and it is quoted at the time.

If a figure is not in the context, the reply says it does not have it. An invented figure does not raise an error: it produces a false alert that looks perfectly good.

The scoreboard: how the method is measured

The system publishes its predictions before they are resolved, and states the resolution date. The scoreboard (/datos/marcador.json, and the table the homepage renders) states:

• What is reviewed: the forecasting method (where the probability written before the outcome comes from), the list of open predictions and their status, and the median Brier score of the resolved ones.
• How often: annually. The date of the last review is updated by hand, on the day of the review.

The Brier score scores a probability written in advance against what actually happened. 0 is perfect; 0,25 is what you get by flipping a coin. It measures a method, it promises nothing: the scoreboard also fails, and getting one right says nothing about the next one.

What this system does not do

These limitations are printed on the seal of every reply; they are not left to be inferred:

• There is no price target. The system does not issue one, and the field is null on purpose.
• The horizon is qualitative, never a date, and no intrinsic valuation is carried out. The valuation engine is none, and that is why the price target is not a number.
• It is not personalised financial advice in any sense. There is no suitability assessment, no appropriateness assessment, no execution and no custody of funds.
• There is no cryptographic signature on the seal. It is a record of what was asserted and with which data, and it does not pretend to a signature it does not have.
• There is no second pair of eyes. A deterministic control inside the system itself, applied to the text of the reply before it is sealed, looks for sentences that recommend a specific instrument to the reader. If it finds any, they are flagged on the seal, together with the text that triggered them. It is a control of the system, not a note from the model: if a seal can do what the system cannot, it is not a seal.

What this is not

This is general analysis and study material, not a personalised investment recommendation. Whoever decides is whoever invests. The full description is in [/legal/aviso-legal](/legal/aviso-legal) and [/legal/condiciones](/legal/condiciones).