Why the checking matters more than the design
A woman in a waiting room, or the mother holding her hand, is about to hand a doctor a page from a website nobody signed. The only thing that page has going for it is that it is right. So we treated it the way a careful editor treats a front-page story: assume every line is wrong until the source proves otherwise.
Two independent reviewers, working from a blank slate and with no knowledge of how the pages were written, read the Mirror as a gynaecologist, a statistician, a regulator, the mother who prints it, and a hostile journalist, and wrote down everything that would embarrass it. Round one found 17 problems, 4 of them serious enough that the page should not have been shown to anyone. Round two, on the rebuilt page, found 12 more, none of them breaking, 11 of which have been fixed. The 12th is below.
What round one found, and what changed
| What the reviewer found | What is on the page now |
|---|---|
| All eight features were called 'a strong recommendation'. Only some are; missing school is a weak recommendation, and two rows were not in the guideline's list at all. | Seven features, each carrying its own grade from the ESHRE 2022 guideline, on the tick-box, on the result, and in the doctor page. |
| The headline '62–75% of teens with resistant pain have endometriosis' used the wrong group and dropped the review's own warning. | 75% (237 of 314) for treatment-resistant girls who had surgery, with the review's warning that this may be an overestimate, shown only when her history matches. |
| The 100 dots were attributed to a paper that never gives the figure, and her own dot placed her among the diagnosed. | Re-sourced (and corrected again in round two, below). None of the dots is her. |
| Fonts were loaded from Google, so every visitor's address and the page she opened went to a third party, on a page promising 'nothing sent'. | Fonts are served from this site. Verified live: zero requests to anyone else. The build now refuses any page that loads from outside. |
| '28% / 62% pain between periods' was pinned on a 2025 Rome cohort; those are 1997 Boston numbers. | Both shown, each on its own paper. |
| The heavy-bleeding row rested on a source that does not say it. | Removed. |
| 'Untreated disease progresses in 45%' oversold a trial of about 20 women. | Says so: one small trial, about 20 adult women, and in about one in five it had improved on its own. |
| 'Seven-fold' family risk with no range. | Says that with 149 comparison women the true figure could be anywhere from 2 to 24 times; ACOG's 7-to-10-fold beside it. |
| Five 'quotations' were trimmed or paraphrased inside quotation marks. | Every quotation is the exact source text, odd grammar included. |
| No base rate: a girl who ticks three boxes was left to assume the worst. | ACOG's own sentence is on the count card: most adolescents with painful periods have primary dysmenorrhoea, not endometriosis. |
| The Print button produced a blank page (found in our own testing, not by the reviewer). | Prints the doctor page alone; verified as a real two-page PDF. |
What round two found, and what changed
| What the reviewer found | What is on the page now |
|---|---|
| The 100 dots were pinned on a 2002 paper of 3,680 members that never gives an age-of-onset figure. | The dots cite the US Endometriosis Association's registry ('over 7,000 confirmed cases', its founder's own words), and the card says the figures are 'as quoted by' a 2020 review and Endometriosis Australia, because the association's reports are not freely readable online. |
| The wait card drew her years as a bar beside the diagnosed women's, seating a 16-year-old in a queue she is probably not in. | Her bar is gone. Her years are stated in a sentence; the 4–11 years and the 10 are labelled as other women's, and a teenager reads that most teenagers with painful periods do not have endometriosis. |
| The doctor page turned 'painkillers' into 'first-line treatment has been tried'. Painkillers are not ESHRE's first-line; hormonal treatment is. | A new question asks what was tried: painkillers only, the pill or another hormonal treatment, or both. The letter says exactly that. Laparoscopy appears only after a hormonal treatment has failed. The build refuses the old phrase. |
| The letter said she was 'never offered treatment' when she was only asked what she had tried. | 'Tried', everywhere. The build refuses 'offered' in the letter. |
| The untreated ask named hormonal contraception and skipped ACOG's own first step, anti-inflammatories. | Both quoted: ESHRE's first-line, ESHRE's good-practice point on NSAIDs, and ACOG's 'empiric treatment with NSAIDs or hormonal suppression, or both'. |
| The letter carried the date the site was built, so a letter printed in October would say 22 September. | Dated by her own phone on the day she prints it. The build refuses a pre-filled date. |
| 'Strong recommendation' hid that the evidence behind it is very low quality, by ESHRE's own rating. | Every grade reads, for example, 'strong recommendation · very-low-quality evidence'. The guideline is firm about asking the question, not about what the answer means. |
| Three wait mis-statements: '4–11 years in the US' (ACOG did not say US); 'nearly 7,000 women diagnosed' (self-reported); one cohort's mean shown as the average. | 'ACOG cites a range of 4 to 11 years'; 'who reported a diagnosis'; the French 10 against 'the 7 years its authors say is commonly described'. |
| No contact address. A page nobody signs gives the people it is about no way to send a correction. | Not yet fixed. It is a decision about who stands behind this site, and it is being made. Until then, this page is the correction record. |
| Eight smaller items: a blank 'did it help' produced a false sentence; onset could not be entered below 10 though ESHRE cites cases at 8; school absence was widened beyond the guideline; and others. | All eight fixed. |
| The privacy sentence was larger than what the site can strictly promise. | Shortened to what is true: no ads, nothing for sale, no analytics. One honest limit is below. |
| Make the letter editable so it is her request, not the site's. | Every sentence of the doctor page can be changed with a tap before printing. It is hers; the guideline text beside it is theirs. |
A mistake the reviewers did not catch
While correcting the dots, we wrote the authors of the 2020 review from memory as one team. The cached source says Sieberg, Lunde and Borsook. Nobody caught it but us, an hour after being told about the same class of error. It is written here because a site about being believed should show how it checks itself, including the misses.
What the software refuses to build
Every number and quotation on the Mirror carries an anchor: the exact phrase in the source it rests on, and a cached copy of that source. If any of the 38 anchors is missing from its source text, the page does not build. We tested it by changing '10 years' to '12 years': refused.
The build also refuses: a figure from a secondary source unless the page says 'as quoted by'; the words 'offered' or 'first-line treatment has been tried' in the doctor page; a pre-filled date; a question count that does not match the questions on the page; any name, tracker, cookie, storage, or file loaded from an outside host; and any page without the words 'not medical advice'.
Before every publish, five pre-filled scenarios (a teenager who has tried the pill, one who has tried only painkillers, an untreated daughter, an adult with late onset, a long history with 'did it help' left blank) are run through a real browser and each doctor page is read back, checking what it must say and must never say. On its first run this caught a real bug: laparoscopy suggested to the painkillers-only girl.
What nobody here can vouch for
The full text of the 2003 Endometriosis Association report (only the abstract is public). The body of ACOG's 2026 clinical practice guideline (paywalled; the site quotes its press release and says so). Printing from an iPhone: the doctor page was verified as a PDF from a desktop browser. And one honest limit of the privacy promise: the hosting company injects no analytics into these pages (checked live: zero), but like every site on its network it sends two headers that ask the browser to report failed connections, not visits, back to it. We cannot turn that off from here and are finding out whether a custom address would allow it.