How we research, grade and verify.
AntiCirc argues a position. That only counts for something if the facts underneath it hold up when someone checks them — so here is exactly how the data on this site is sourced, graded, labelled and corrected.
The short version
Every factual claim on this site is meant to trace back to a source you can open yourself. Sources are graded, not treated as interchangeable. Claims carry a confidence level. Where the evidence genuinely conflicts — and on several questions in this field it does — we show the conflict instead of quietly picking the side that helps our argument.
The uncomfortable corollary: some things we would like to state, we cannot. There is no reliable measured national complication rate for circumcision in the Philippines, so we do not publish one. The evidence on circumcision and erectile function contradicts itself, so we report the contradiction. An advocacy site that invents the missing numbers is easy to dismiss, and deserves to be.
How we grade a source
Not all citations are equal, so we do not display them as if they were.
Peer-reviewed studies · medical-body guidelines · government data · court documents and statutes
What we lead with. A claim that carries real weight on this site should trace back to one of these, and we link the source so you can check it yourself rather than take our word for it.
International-organisation reporting (UN, UNAIDS, WHO) · NGO reports · academic books
Reliable and usually well-documented, but institutional positions can carry a policy stance. We attribute them by name rather than laundering them into 'experts say'.
News reporting · historical sources · datasets of uncertain provenance
Good for establishing that something happened, weaker for establishing how often or why. Used for events and context, not for medical claims.
Personal testimony · editorial and advocacy analysis — including our own
Never presented as evidence of prevalence or causation. Where we publish lived experience, we label it as testimony on the page it appears, because a real account of one life is not a measurement of many.
You will see this grading surface directly in the page furniture — the tier chips beside citations on our guides, and the source list at the foot of every long-form page. Where a source is used for background only (a general encyclopaedia entry, say) it is labelled as background and is not treated as medical authority.
How confident we are
Structured claims in our research layer carry an explicit confidence level.
Multiple strong sources agree, and we are not aware of serious contrary evidence.
Supported, but with limitations — sample size, study design, or narrow applicability.
Thin or preliminary evidence. Stated as a possibility, never as a finding.
Credible sources genuinely disagree. We show the disagreement rather than pick the side that suits us.
We do not have adequate sourcing. We say so instead of filling the gap.
"Disputed" is a real answer here, not a hedge. Circumcision research contains well-funded work on both sides, and a systematic review of sexual-function studies can find some reporting harm, some reporting benefit and most reporting no measurable difference. We would rather show you that mess than flatten it into a slogan.
The Incident Registry
Documented harm cases, each carrying its own verification tier.
Surfaced by AI deep-research from public reporting and not independently verified by us. Labelled as such on every case that carries it — visibly, not in a footnote.
Backed by named press or official reporting that we have read and linked. This is the default tier for the cases in the registry.
Corroborated by member reports. Useful signal, weaker provenance than press or court records — and tiered accordingly.
Directly confirmed by AntiCirc against primary documentation. The smallest tier, deliberately.
What the registry is not: it is a record of cases that were documented, which is not the same as a measurement of how often harm occurs. Cases enter the registry because someone reported them — so the registry reflects reporting, media attention and our own coverage as much as underlying incidence. Read it as evidence that these harms are real and documented, never as a rate.
Browse the registryCountry data & the world map
Which figures are measured, which are estimates, and which are ours.
Circumcision prevalence
Per-country prevalence estimates, held in one queryable store rather than scattered through the code so that a figure can be traced and corrected in one place. These are estimates drawn from published sources — national prevalence data in this field is often old, patchy or self-reported, and should be read as an order of magnitude rather than a precise count.
HIV & epidemiological context
Hardened to UNAIDS estimates, or national surveillance for high-income countries UNAIDS does not model. Each country stores the year and the issuing body alongside the figure, so a number on a page can be checked against its vintage. Prevalence and trend are shown separately — a country can have low prevalence and a fast-growing epidemic at the same time, and collapsing the two misleads.
Composite scores are editorial
The sex-education gap, prevention context, policy environment and stigma indices are AntiCirc editorial composites — our own summary judgements on a 0–100 scale, not measured international indices. So is the categorical profile (whether routine infant circumcision is common, uncommon or rare; whether it is treated as medically necessary or contested). We build them because they are useful for comparison, and we label them because passing them off as measurement would be dishonest.
Legal status is maintained separately from these indicators, against statutes, rulings and government sources, and joined to each country by ISO code. See the law tracker and the world explorer.
The news pipeline, and where AI is used
Deterministic ingestion, human moderation, and an honest account of the machine's role.
How news gets here
Items are ingested automatically from a public global news dataset, matched deterministically — by query, not by model judgement — and land in a moderation queue as unpublished. A human approves or hides each one before it appears. We store the publisher's own headline and a short lead excerpt under fair use, link out to the original, and never republish a full article body.
Where AI is — and is not — involved
AI assists with research sweeps, drafting and translation. It does not decide what is true. The news pipeline's AI processing is switched off at the feature-flag level, so ingestion stays deterministic. Where AI deep research surfaced a case, that case carries the AI-compiled tier for anyone to see. Nothing is published as verified because a model asserted it.
Language editions
Tagalog and Cebuano are full editions, not machine output dropped on the page.
Our Tagalog and Cebuano pages are separately authored and human-reviewed, with their own sources and their own URLs, because a translated medical page that reads like machine output does not get trusted and does not deserve to be. Where a translation has had only a first pass and still needs native review, we treat that as an open task rather than a finished page. If you speak Tagalog or Cebuano and something reads wrong, telling us is genuinely useful.
What we will never do
The lines that make the rest of it worth anything.
- Invent a statistic, a percentage or a complication rate. Where a reliable measured figure does not exist — such as a national complication rate for the Philippines — we say that plainly instead of publishing a number that looks authoritative.
- Present documented individual cases as if they were measured rates. A registry of harm cases counts reports, not incidence.
- Fabricate testimonials, reviews or personal stories. Every account on this site came from a real person or a real cited source.
- Present AI-compiled research as independently verified. It carries the AI-compiled tier wherever it appears.
- Quietly delete an error. Corrections are made in the open, on the page that carried the mistake.
Corrections
We get things wrong. When you find one, tell us — with the page and, if you have it, a source — and we will fix it on the page that carried the error rather than quietly deleting it. If a claim cannot be re-sourced when challenged, it comes down. That rule applies to claims that support our position as strictly as to ones that don't; a correction that costs us an argument is still a correction.
Frequently asked questions
How does AntiCirc research its data?
Every claim is traced to a source, and the source is graded. Strongest are peer-reviewed studies, medical-body guidelines, government data and court documents; below those sit international-organisation and NGO reporting; then news and historical sources for context; and finally testimony and analysis, which is never used as evidence of prevalence or causation. Each structured claim also carries a confidence level — high, moderate, low, disputed, or unknown — and where credible sources disagree we show the disagreement rather than choosing the side that suits our position.
Where do the country figures come from?
Circumcision prevalence and HIV figures are stored per country with their provenance attached — HIV estimates come from UNAIDS, or national surveillance for high-income countries UNAIDS does not model, and each figure carries the year and the issuing body. Legal status is maintained separately against statutes and rulings. Some composite scores shown on country pages — the sex-education gap, prevention context, policy environment and stigma indices — are AntiCirc editorial composites, not measured international indices, and are described that way wherever they appear.
Does AntiCirc use AI to write its content?
AI is used as a research and drafting assistant, never as an unchecked author. The news pipeline is deterministic — items are ingested from a public news dataset and moderated by a human before publication, with AI processing switched off by feature flag. AI deep-research compilations are carried in the incident registry under an explicit AI-compiled verification tier so you can see exactly which cases have not been independently verified. Nothing is published as verified because a model asserted it.
How is the incident registry verified?
Each case carries one of four verification tiers: AI-compiled (surfaced from public reporting, not independently verified), Reported (backed by named press or official reporting we have read and linked), Community (corroborated by member reports), and Verified (directly confirmed by AntiCirc against primary documentation). The tier is shown on the case. We do not promote a case to a higher tier without doing the work the tier describes.
How do I report an error on AntiCirc?
Tell us and we will fix it. Use the contact page or the feedback board, and include the page and, where possible, a source. Corrections are made openly on the page that carried the error rather than quietly removed. If a claim cannot be re-sourced, we withdraw it.