Editorial standards

QFT.org aims to give readers enough explanation to understand a result, reproduce its central calculation, and judge where it applies. These are the standards used in the site’s ongoing revision. They describe the review process; they do not imply that all existing pages have completed it.

Scientific arguments that can be checked

A treatment should define the necessary objects and assumptions before using them. Derivations should connect the steps that determine the result, with particular care for signs, units, normalization, domains, boundary conditions, and the order of limits. A proof outline should say which steps it omits. An approximation should identify its controlling parameter and failure conditions.

Checks depend on the claim. They can include a second derivation, a symmetry identity, a limiting case, a counterexample, an independent numerical method, or agreement between canonical and functional-integral constructions. Computational results should retain the inputs, editable code, convergence or uncertainty information, and the conditions needed to reproduce them.

Sources and research claims

Citations should support the nearby claim and cover its actual hypotheses. Whenever possible, links identify the relevant section, equation, or printed page in the cited edition. A bibliography supports the exposition; it does not replace a missing explanation or calculation.

Research discussions distinguish exact results, controlled approximations, numerical evidence, observations, and conjectures. They state consequential cutoffs and limitations and describe material disagreement or negative evidence. A newer paper does not automatically supersede a stronger result.

Learning and reading

Pages should have a clear purpose, proportionate preparation, and working links to the subjects on which they depend. Worked examples show how to use the ideas. Exercises should test understanding and transfer, with solutions that explain the reasoning.

Figures should make a relationship clearer than prose or equations alone. Captions and text alternatives should communicate their main point. Mathematics, navigation, and scientific figures are checked at desktop and narrow widths; keyboard access, themes, magnification, and print are part of the intended reading support. An editorial walkthrough is distinct from a study with actual learners or assistive-technology users, and we do not claim such studies without conducting them.

Automated editing and its limits

GPT-6 assists with drafting, calculation, source comparison, critical review, and technical implementation. Jie Ren has delegated page-readiness decisions to the assistant. These are automated editorial decisions, not human peer review or endorsement by the authors of cited work.

For consequential material, drafting and adversarial review use separate contexts, supplemented by independently checkable calculations or sources. Reviews by the same model family can share mistakes. A fluent explanation or a passing software check is therefore insufficient evidence of scientific correctness. Acceptance applies to the version and scope actually examined; later changes require an assessment of their consequences.

Corrections

Use Send feedback to report a questionable step, citation, normalization, numerical result, or accessibility problem. Include the page address, the relevant passage or equation, and a proposed correction or supporting source when available.

A material correction should explain the scientific change and check its dependent arguments. A result that cannot be supported should be qualified or withdrawn. If an immutable edition is issued, substantive corrections must identify that edition rather than silently alter it. The site does not promise a response time or describe unresolved findings as settled.