How the Atlas tells AI what to check, what not to assume and when to leave a question unresolved.
This page explains the rules built into the Atlas Prompts and reference sections, and why they can produce a more disciplined answer than a broad AI chat question.
The Atlas starts from the specific question your team needs answered. Its Prompts tell the AI which facts to examine, which assumptions to avoid and which unresolved issues must remain unresolved.
Choose the question you need answered, then copy the matching Prompt.
Upload the full Atlas PDF to an organisation-approved AI system, choose the question you need help with, copy the matching Prompt and add the player and deal information it asks for.
The Prompt tells the AI what to analyse, what information you need to provide, what it must not assume and how it should answer. When deeper checks are useful, it directs the AI to the relevant decision questions and rules against unsupported assumptions inside the Atlas.
AI access is not included. Provider selection, confidentiality controls, verification and final decisions remain with the customer.
Do not analyse the whole transfer when one question is holding it up.
A fee gap, seller condition, player terms, internal approval and an expiring alternative can all be open at the same time. Treating them as one large problem creates more work and makes it easier for the important issue to disappear inside the summary.
The Atlas contains seven ready-to-use Prompts for different questions: what to do next, where the deal stands, what the case currently supports, what could change the current view, what a decision-maker needs to know, what to verify before a conversation, and what changed after new information arrived.
The AI uses the deeper reference sections only when the selected Prompt needs them. It is instructed to stop once it has enough relevant information to answer responsibly, rather than searching the whole PDF for extra material.
The Atlas is designed for the possibility that AI can be wrong.
AI can produce fluent answers that go beyond the evidence. More context does not guarantee that every relevant fact receives equal attention. The methodology is built to constrain those failure modes rather than pretend they do not exist.
The Prompts tell the AI to keep missing, uncertain or conflicting information visible instead of filling gaps with a plausible answer.
The Prompts direct the AI to use only the relevant checks instead of working through the entire reference section from beginning to end.
Several reports tracing back to one claim are treated as one underlying source unless independent confirmation exists.
Each Prompt defines the question, the permitted checks, the answer format and the conclusions or recommendations the AI must not invent.
27 research records, with the source trail kept visible.
The current Atlas contains 27 research records across football research, decision science, contextual modelling and AI reliability. Each record identifies the underlying paper or research source and provides a direct DOI, publisher, author or institutional source link; where an open version is available, the Atlas also provides a direct reading link. The research is used only where it supports a decision method or AI reliability control. It is not treated as a football transfer rule or as proof of a live-case conclusion.
Inspectable by design. The buyer does not have to treat the methodology as a black box: the research records identify the underlying source and provide direct links so the basis for the relevant controls can be checked.
Early numbers, reputations and familiar comparisons can shape later judgment. The Atlas includes checks that ask whether the case still stands when the anchor is removed.
Tversky & Kahneman, 1974 →A persuasive case narrative can crowd out relevant outside evidence. The Atlas asks whether a defensible reference class exists before using one.
Kahneman & Lovallo, 1993 →A good result does not prove that the earlier decision was good, and a bad result does not prove the opposite. Decisions are reviewed from the information available at the time.
Baron & Hershey, 1988 →Research has shown that language models can produce false but plausible answers and may use long documents unevenly. Those findings support the Atlas's approach of directing the AI to only the relevant parts of the PDF, keeping uncertainty explicit and asking for focused answers.
TruthfulQA, 2022 →Lost in the Middle, 2024 →
The papers shown above are selected examples from the 27-record research basis, not the full bibliography. The Atlas keeps the source trail with the relevant research record so the underlying source can be checked directly. Research informs these checks; live transfer conclusions still depend on the current case evidence.
Leave the gaps where they are.
The Atlas separates a supported fact from a party claim and from an explanation of why something happened. It also keeps five clear status labels available: CONFIRMED, OPEN, UNKNOWN, DISPUTED and N/A.
This matters because a missing fact can be commercially important without being knowable yet. A disputed explanation can be repeated widely without becoming established. The method keeps those differences visible until better evidence arrives.
When credible accounts conflict, the Atlas preserves the strongest common factual ground rather than forcing one narrative to win.
Use history to find the question, not to manufacture the answer.
Real transfers can reveal useful mechanisms: a seller-side dependency, a deadline that removes an option, a commitment that is narrower than it appears, or a player who cannot yet be used as intended.
The Atlas uses historical transfers only when they help explain why a question matters. A past transfer does not establish the facts, rule or outcome of your current transfer. Rule-sensitive examples require the current framework, date, jurisdiction and scope to be checked again.
When you already have a usable earlier analysis, update only what the new information actually changes.
The UPDATE Prompt is for cases where your team already has a usable earlier Atlas analysis. It compares new information with that earlier analysis before using any deeper Atlas reference material.
It can return NO MATERIAL CHANGE, identify supported changes, preserve directly implicated items that remain unchanged and reopen questions only where the evidence supports that dependency.
If there is no usable earlier analysis, use another Prompt first. UPDATE is optional and is not a final stage that every case must reach.
See the update in the example case →Use AI for structure, not authority.
The Atlas is decision support. It does not choose the player, guarantee a valuation or make the final transfer decision. Where a legal, medical, regulatory, tax, accounting or other specialist conclusion is required, that conclusion remains with the relevant specialist.
The method is designed to make the team's reasoning easier to inspect and update. Verification, approvals, negotiation and the final decision remain with the customer.
Read the Terms of Sale and Licence →