Evidence first. Limits visible.

1. Start with the question and its scope.

A useful article states which question it addresses, for whom and under what conditions. Clinical interpretation, research methods, instrument documentation and commercial announcements are different categories. A conclusion should not silently move from one category to another.

Clinical, diagnostic, MRD, regulatory and product-comparison claims require careful source checking, explicit limitations and an accountable human editorial decision. If a claim cannot be adequately supported, it should be withheld or narrowed. CFCM does not present this editor-led process as independent scientific validation. Educational material does not replace local procedures, assay validation, manufacturer instructions or professional clinical judgement.

2. Make the evidence traceable.

Factual claims should point to identifiable original sources where available: original studies, official technical documents, consensus recommendations, regulators or the responsible organisation. A manufacturer is the primary source for its own specification or announcement; that does not make its claim an independent performance evaluation.

Separate peer-reviewed work, preprints, abstracts, technical documentation and marketing. Record the relevant source date and version. A recently visited webpage may still describe an older study. Accessing a landing page is not the same as examining the full publication.

3. Show the boundary of the conclusion.

State the specimen, population, method, instrument configuration and intended use when they determine applicability. Distinguish association from causation, analytical performance from clinical benefit, and a numerical calculation from a validated measurement procedure.

Include important contrary findings and limitations. Where sources disagree, describe the disagreement rather than disguising it as consensus. When evidence is missing or inaccessible, say what could and could not be checked.

4. Label interpretation and teaching constructions.

CFCM analysis should be recognisable as analysis, not attributed to a source that did not make it. Worked numbers, plots and cases constructed for teaching should be labelled synthetic. Synthetic material must not be presented as patient data, a clinical validation dataset or a published experimental result.

Research examples and historical teaching can be valuable when their limits are visible. An old classification should not be presented as the current diagnostic framework.

5. State who reviewed the material.

In the initial phase, the founder is responsible for editorial review and the decision to release each exact version. An AI drafting assistant cannot provide that human approval. The founder's own review is not independent external peer review, and source checks do not make it so. Independent scientific review is planned for a later phase; it will only be claimed for material that has actually undergone that process. Substantive changes require another editorial check.

Review should include the whole public output: body text, titles, summaries, figures, captions, references, search previews, metadata and tool outputs. A verified source URL does not by itself verify every sentence beside it. A review of one item does not authorise an unrelated whole-site release.

6. Use AI assistance transparently.

AI may assist with drafting, discovery, code and formatting. It is not an author credential, a cited scientific source or a substitute for accountable human review. References, quotations and calculations still need checking. Private patient information must not be entered into public teaching tools or an unapproved processing workflow.

The seven initial Academy lessons and the teaching-calculator prototypes were prepared with AI assistance. Juan Manuel Ojeda has completed the editorial review of the seven lessons and approved the calculator change delivered as version 0.1.2. They remain unpublished pending completion of the technical release process. They have not undergone independent scientific peer review. Local software tests do not make them clinically validated, and no scientific-review badge or date is assigned on that basis.

7. Make interests and corrections visible.

Relevant employment, consulting, funding, ownership, sponsorship and other interests should be disclosed at the appropriate author or item level. A neutral tone alone does not establish independence. Commercial relationships must not be hidden behind a scientific-looking summary.

The correction process retains the original version, records what changed and why, and displays a meaningful update note when a correction affects interpretation. A source-check date, substantive revision date and original publication date are distinct; none should be invented to create an appearance of longevity.

8. Define what a tool does not do.

Each educational tool should state its formula, units, assumptions, supported inputs, limitations and tested scope. Software arithmetic tests and browser checks are not clinical validation. No universal pass/fail threshold should be implied by the number of digits displayed.

Tools should minimise data collection and avoid requesting identifiers that are unnecessary for teaching. A statement about one tool's code must not be mistaken for a complete privacy assessment of the hosting website or its external services.