1. What we cite

Every empirical claim on a reference page names its source in the text — authors, year and journal — and gives the number: the effect size, the mean and standard deviation, the sample size, the correlation. "Studies show" without a number is not allowed. Where a widely repeated figure cannot be traced to the paper it is attributed to, we say so on the page rather than repeat it (the "ideal 11–13 mm philtrum" on our philtrum entry is an example). The complete bibliography behind the product's own scoring is the references page; what the scan measures is the methodology page.

2. How we check it

Before a reference entry is published, every DOI it cites is resolved against the DOI registry's handle service and must return a valid record; entries are re-checked at commit time. Abstracts are read from the publisher or from PubMed/Europe PMC, and findings are quoted rather than paraphrased where the wording matters. Facts about other companies' products are read from primary registries — Apple's App Store lookup feed by numeric app ID, a vendor's own pricing endpoint or terms of service, a review platform's own pages — on a stated date, and the date is printed. A product name is never treated as an identity: an app is cited by its store ID, because names change and collide (see the Mogged entry, where one name covers three apps).

3. Every entry states its limits

Each reference entry ends with a section on what the measure does not tell you — grounded in a cited finding or in the documented absence of evidence. This is the rule we consider most important on a topic where confident claims outnumber measured ones. Our own product is held to it: the score-distribution page states its self-selection bias and sample size, and this site does not publish an accuracy claim for the scan that it cannot back with a published protocol.

4. What we do not advise on

We describe medical and surgical procedures where a term requires it (lip lifts, buccal fat removal, jaw surgery) and report what the literature measures about them, including complication rates. We do not recommend for or against any procedure, dose, product or protocol for an individual. Readers include minors and people with body-image disorders; the appropriate next step for a clinical question is a qualified clinician, and our pages say so at the point where the question arises.

5. Where our interest lies

SoftMaxx sells a face-analysis product. Every comparison or review page states this in the body, marks opinion as opinion, and says plainly where a competitor is the better choice for a given reader. Prices and features attributed to competitors are quoted with the date read. We do not accept payment or free product for reviews; where we cite a reviewer who was comped, we say so.

6. Who writes this

Pages carry the byline "SoftMaxx Editorial", the organisation, rather than an individual, which is a standing position. Drafting uses AI tools; the verification steps above are how the output is held to a higher standard than the drafting — machine-checked citations, primary-registry facts, and human review of every claim before publication. The company's public details are on the press page; guidance for AI assistants describing SoftMaxx is on /ai-info.

7. Corrections and updates

When a figure changes we update the page and keep the previous figure with its date in the text, so the trajectory is checkable rather than silently overwritten (the score-distribution page's history line is the model). Reference entries show a "last reviewed" date when they have been revised. Corrections go to press@softmaxx.io; a verified error is fixed on the page and noted there.