Editorial Policy

Three principles govern everything published.

Measurement over assertion. Where a claim can be expressed as a ratio, a threshold or a count, we express it that way and name the source. "A heart-shaped face has a wide forehead" asserts. "A heart classification needs bitemporal width above bigonial width by a stated margin" measures.

Stated uncertainty. Facial classification is continuous, not categorical. Where a result sits near a boundary, or the literature splits, the article says so rather than offering a clean answer nobody can support.

Independence from revenue. Display advertising funds the site. No advertiser, agency or commercial partner sees a page before publication or influences what gets commissioned.

2.1 What we commission

Topics arrive from three places: gaps in our own documentation, questions reaching [email protected], and search demand for questions that published material answers badly.

Advertising value never justifies a topic on its own. Where a subject carries high commercial value and thin evidence, and most "best face shape" framings land there, we either decline it or reframe it around something measurable. Whatever survives goes into a written brief.

2.2 The brief

Briefs specify the question a piece must answer, the measurements it uses, the sources the writer consults, the claims out of scope, and the disclaimers required. For tool methodology pages a brief adds the model version, the landmark indices in use and the current thresholds.

3.1 Acceptable sources

In descending order of preference:

1. Peer-reviewed research in anthropometry, craniofacial morphology, orthodontics and vision science, cited with author, year and publication.
2. Standard reference works on facial measurement convention.
3. Primary technical documentation from the model vendors whose systems we run, principally Google's MediaPipe Face Landmarker docs and its published landmark topology.
4. Direct measurement by our own team, described in enough detail for a reader to repeat it.
5. Named practitioner expertise, attributed to a person with stated credentials and marked in the text as professional opinion rather than evidence.

3.2 Sources we do not accept

Other tools sites, content farms, undated listicles, unattributed social posts, brand marketing and chatbot output supply no factual claim here. An uncited article never becomes a citation. Where a widely repeated claim has no traceable origin, and the "seven face shapes" taxonomy is the clearest case, the text calls it convention rather than evidence.

3.3 Citation practice

Claims a reasonable reader could dispute carry an inline attribution naming the source. Links point at the primary work, never a secondary write-up. Paywalled material gets flagged as paywalled, so readers know what they'll hit before clicking.

Four stages stand between a draft and publication.

That third stage exists because the most damaging mistake this site can make is describing a calculation that differs from the one running in your browser. Nothing describing a tool's behavior ships without it. A human reviewer signs off on everything, and there is no automated publication path.

Readers deserve a straight answer here.

5.1 What AI tools are used for

Large language models handle research triage, surfacing candidate sources for a human to read. They also assist with outlining, copy-editing suggestions on human-written drafts, and test cases against our classification logic.

5.2 What AI tools are not used for

Model-generated text never ships as finished copy. No model counts as a source of fact: any claim one surfaces gets verified against primary material first, and that primary material is what gets cited. Citations are never machine-generated, because fabricated references are a known failure mode. Nor do we synthesize images presented as photographs of real people.

5.3 Human accountability

Accountability rests with people. Every page has a named human author and a named human reviewer answerable for its accuracy, whatever tools assisted production. Those names appear in the byline described in section 11.

Separation here is structural, not a promise. The Face Shape Detector is free and funded by display advertising served through Google AdSense.

Advertisers select no topics, review no drafts, influence no conclusions and affect no ordering. Nobody selling ad space speaks to editorial, because nobody here sells ad space. Placements sit apart from article copy and carry labels clear enough that no reasonable reader mistakes a unit for editorial. If an ad renders in a way that blurs the line, report it to [email protected] as a bug.

Classification results are not for sale. No payment makes a tool return a chosen answer.

Sponsored posts and paid placements don't currently run here. Should that change, these rules apply without exception.

• Label any sponsored or paid item "Paid content" or "Sponsored" in the heading area, above the fold, at the same visual weight as the title.
• Disclose any affiliate link at the top of the page before the first such link appears, never in a footer.
• Exclude sponsored items from every recommendation, ranking and comparison.
• Refuse to let a commercial relationship change a conclusion. Anything we wouldn't recommend unpaid stays unrecommended paid.

Undisclosed commercial interest is what those rules prevent, so contributors disclose theirs.

Contributors disclose in writing any financial interest, employment, consultancy or family relationship with a company, clinic or brand named in a piece they work on. A disclosed conflict triggers reassignment, or an on-page disclosure where that is impractical. Gifts, free treatments and paid travel from the beauty, cosmetic surgery, orthodontic and styling sectors are refused.

Every page carries a visible last-reviewed date. Baseline cadence is a full review at least once every 12 months, and every 6 months for tool methodology pages.

Reviews also happen ahead of schedule, triggered by any of these: a change to the MediaPipe model version or landmark topology; a changed classification threshold in our code; a reader report of a factual error; new peer-reviewed research contradicting a published claim; or a change in law or platform policy affecting what a page must disclose.

Substantive changes get a note at the foot of the page saying what moved and why. Cosmetic edits refresh the reviewed date without one. Unsupportable pages are retired with an explanation, never silently deleted. Reader reports are the trigger we act on fastest.

Corrections run openly, and history never gets quietly rewritten.

1. Report. Email [email protected] with the page URL, the passage you believe is wrong, and where possible the source contradicting it. The site contact form reaches the same inbox.
2. Acknowledge. A reply arrives within 5 working days confirming receipt and whether we're treating it as a factual error, a clarity issue or a difference of interpretation.
3. Verify. Someone other than the original writer re-checks the claim against primary sources.
4. Fix. Confirmed factual errors are corrected within 10 working days. Anything that could mislead a reader about privacy, safety or data handling is corrected within 48 hours.
5. Disclose. Corrections to a factual claim carry a dated note at the foot of the page stating what was wrong and what it now says. Typographical fixes need none.

Where a page cannot be made accurate, we unpublish it and say so at that URL, under the byline of whoever owned it.

Bylines name an author, the reviewer where a specialist read the piece, a publication date and a last-reviewed date. Invented personas and stock author photographs appear nowhere here.

Editor note: before launch, replace this paragraph with the registered trading name of the site operator and the name and contact route of the responsible editor, and publish author biography pages listing relevant credentials.

Questions about this policy, corrections, source requests and complaints all reach [email protected]. Personal data is covered by our privacy policy, and site use by our terms of use.

This policy exists so the process behind a measurement-led page is as checkable as the measurement. Three principles set the floor: numbers over assertions, uncertainty stated rather than hidden, revenue kept away from conclusions. Topics arrive from documentation gaps, reader mail and badly answered search demand, then travel through a written brief.

Peer-reviewed anthropometry and craniofacial morphology sit at the top of the source order, chatbot output nowhere on it. Four stages check a draft, one against the shipped code. Language models triage and outline; people write, verify and sign. Google AdSense funds the work and touches none of it, sponsored placement would carry a label above the fold, conflicts get disclosed in writing.

Pages are reviewed at least yearly, sooner when a threshold or a reader report demands it. Errors are acknowledged in 5 working days and fixed in 10. Send yours to [email protected].