Eden Openly

Evidence Transmission Register

Editorial · Evidence Transmission Register

This register records defects found in published documents — national evidence reviews, journal articles, a systematic review, a specialist news summary, news articles, and our own work — together with what was reported, to whom, on what date, and what has happened since.

The error is not the measurement. The measurement is the interval between reporting a defect and anything changing, and whether anything changes at all. A register of other people’s mistakes would be a different document with a different purpose.

It sits alongside the Deposit Standard, which governs how our own documents are built and checked, and the Corrections policy, which governs what we do when a reader tells us we are wrong.

Why this is public. A finding about someone else’s document has two possible lives. It can be published as a criticism, or it can be sent to the people who can correct it. Only the second produces a fact rather than an allegation — and it produces one either way. If the document is corrected, the defect was real and is now gone. If it is not, the sentence reported on this date, unchanged as of this date is itself the finding.

The cost of the second route is time. The rule we work to is short: report first; escalate only if ignored; never both at once.

How to read an entry

Each defect is one entry. A defect that appears in more than one document is split by document, with a letter suffix — ET-001a, ET-001b, ET-001c — so that a publisher correcting one document and not another can be recorded exactly.

Entries and items are different counts, and this page never uses one for the other. An item is what was put in a notice. An entry is that item in one named document. The ascertainment defect below is one item and three entries.

Identifiers are permanent. Numbers are never reused and never renumbered. An entry we get wrong is struck and stays visible. An entry that was a drafting placeholder is marked superseded and says what replaced it. The R-series comes from the earlier internal register and keeps its own numbering, because renumbering it would break the audit trail it exists to provide.

Levels

Level describes the effect of a defect on the document it is in. It is never the effect on policy, on patients, or on anyone’s decision. A clerical error in a national evidence review and a clerical error in a newsletter are the same level, because the level describes the defect and not the stakes. The moment the scale absorbs consequence it stops being a measurement and becomes an argument.

Clerical — the document says something other than what it evidently means. Correcting it changes no statement.
Substantive — a statement is wrong as written, and correcting it changes that statement.
Conclusion-affecting — a statement the document’s reasoning rests on is wrong or missing, such that correcting it changes or qualifies what the document concludes.

Statuses

Reported — notice delivered to the address. Nothing back.
Acknowledged — a human replied with something substantive. Not agreement.
Correction directed — the publisher has said a correction will be made. The document still carries the defect.
Corrected — the document itself has changed, verified by opening it.
Declined — the publisher has given reasons for not correcting. Reasons are recorded in full.
Unchanged — a dated check found the defect still present.
Struck — we were wrong. The entry stays visible, struck through, with the date and the reason.

Correction directed and corrected are never merged, and the first never ages silently into the second. A publisher’s undertaking is not a corrected document. Readers are served the document, not the undertaking.

Delivery is recorded. Opens and clicks are not. Our mail platform logs opened and clicked events; neither establishes that a person did it. Letters sent on 28 September were logged as opened within seconds of delivery, and one as clicked fifteen seconds after — timings that indicate a security gateway, not a reader.

What is in this register, and what is not

In: defects in published documents, ours and everyone else’s, in one sequence and on the same terms. Including accusations of ours that turned out to be wrong. Those have their own section and are not hidden among the rest.

Not defects, and recorded separately: findings that are true but are consequences of design rather than mistakes, and which cannot be fixed by editing a document; and methodological criticism, which is arguable and which a competent person could disagree with. Neither was ever put in a letter. Both are listed below so that this page does not imply everything found was an error.

Out — requests for data. A defect has a correct-or-decline axis and a latency from report to change. A data request has a statutory clock and a disclosure outcome. In one register the status would mean two different things. Those are recorded separately.

Out — confidential enquiries. Approaches made to us in confidence are not recorded here in any form: not by name, not by role, not by channel, not as a dated line with the content withheld. Anonymising does not work at this scale — in a small field, a role and a date identify a person to exactly the people who will read this page. The only reliable protection is absence.

Institutions are named. Individuals are not. The finding never needs the name: a corresponding author, notified 13 July, no reply as of 28 September carries precisely what is being measured.

NHS England — evidence reviews published 9 March 2026

Reported: 21 items · Entries: 23 · Resolved: 0

Reported to the Clinical Effectiveness Team in two notices: sixteen items on 12 July 2026 concerning URN 2417h (oestrogen monotherapy) and URN 2417j (testosterone monotherapy) with related PICO documents, and five further items on 14 July 2026. Acknowledged on 13 July, within twenty-four hours, and escalated to the Medical Director for specialised commissioning and to the independent evidence review provider, with a full response undertaken. That is the most engaged response of the notices sent, and it should be recorded as such.

Both counts are derived from the notices as sent. The first contains sixteen defects; the second states five and opens by acknowledging NHS England’s own count of the first — “escalating the sixteen items”. Twenty-one items across twenty-three entries, because one defect appears in three documents.

Documents opened and read on 12 August 2026 and again on 1 September 2026. On both dates every reported item was present and unchanged, with no correction notice, erratum, revision statement or changelog. The resolved count is derived from those two checks.

The acknowledgement covers the first notice only. The sixteen items of 12 July 2026 were acknowledged on 13 July. The five items of 14 July 2026 were not acknowledged, and have not been since. The 14 July notice opens by thanking NHS England for their reply of the previous day and for escalating the sixteen items — so the acknowledgement on file predates the second notice and cannot cover it.

The entries reported on 14 July, and therefore unacknowledged, are ET-001a, ET-001b, ET-001c, ET-036, ET-037, ET-040 and ET-041. Every entry in this section is unresolved and has provenance ER-015.

Item ET-001 — one defect, three documents

Reported 14 July 2026. All three reviews reproduce Valentine’s outcome definition — outcomes “defined as having either a diagnosis or at least two abnormal measurements” — and none records what the paper says about how often those measurements were taken.

ET-001a

URN: 2417h
Issue: Valentine’s outcome definition reproduced without its ascertainment limitation. The paper reports that transgender and gender-diverse youth were tested far more often than matched controls — total cholesterol 39.1% against 11.6%, HDL 38.8% against 11.1%, triglycerides 39.1% against 12.3% — and states this “could lead to more opportunities to receive a cardiometabolic-related diagnosis”. The authors use it to explain their own results. Where an outcome is defined by two abnormal laboratory values, a threefold difference in testing rates bears directly on it.
Level: Conclusion-affecting · Status: Unchanged · Reported: 14 July 2026

ET-001b

URN: 2417l · Issue: As ET-001a, in this document.
Level: Conclusion-affecting · Status: Unchanged · Reported: 14 July 2026

ET-001c

URN: 2417n · Issue: As ET-001a, in this document.
Level: Conclusion-affecting · Status: Unchanged · Reported: 14 July 2026

Carrying this limitation forward would weaken an apparent adverse association, not strengthen one. We said so in the letter, unprompted, and we say it here.

The same notice records two things in NHS England’s favour, and they belong here too. URN 2417n’s limitations section is notably thorough — it identifies the absence of temporality, the missing dose and duration data, the possible inclusion of out-of-scope individuals, the unreported co-interventions and the impossibility of adjusting for smoking or alcohol. And 2417n reports Valentine’s dyslipidaemia odds ratio as 3.7 (95% CI 2.0 to 6.7), which is the figure in the paper’s Results. Their reviewers read the Results rather than the abstract. In the same period, we did not.

Item ET-002 — superseded

ET-002 — SUPERSEDED 28 September 2026

A bundle entry covering “twenty further items”, used as a drafting placeholder while the notices were unavailable. Expanded into ET-023 to ET-042 below, one entry per defect, derived from the notices as sent. It was never a finding, so it is marked superseded rather than struck.

URN 2417h — oestrogen monotherapy

Reported 12 July 2026. ET-023 to ET-026 concern the reporting of Grannis et al. (Psychoneuroendocrinology 2023;156:106319) and Allen et al. (Clin Pract Pediatr Psychol 2019;7(3):302–311).

ET-023

Issue: Three non-significant comparisons reported as statistically significant. The review states participants receiving oestrogen monotherapy had “statistically significantly higher” depressive symptoms, suicidality and social anxiety, each p<0.05. Grannis Table 1(D), AMAB column: depression F = 0.03 (1,29), suicidality F = 0.38 (1,29), social anxiety F = 0.20 (1,29), none asterisked. Grannis §3.1: “no GAHT effect was found for TNBAMAB.”
Level: Substantive · Status: Unchanged

ET-024

Issue: One significant comparison reported as not significant. The review reports body image dissatisfaction as “not statistically significant”. Grannis Table 1(D), AMAB column: F = 4.88 (1,29), asterisked, means 85.33 against 88.53 — lower dissatisfaction in the oestrogen group. All five significance designations in the review correspond to Grannis’s Omnibus column, which reports the sex-assigned-at-birth × treatment interaction, not the within-group effect.
Level: Substantive · Status: Unchanged

ET-025

Issue: Confounder adjustment stated absent where present. The review states Grannis “lacked identification of and adjustment for potential confounding factors”. Grannis §2.4.1: “age was included as a covariate in all models”. The review’s own Table 1 reproduces the age difference and its own GRADE footnote reads “Adjusted for age group”.
Level: Substantive · Status: Unchanged

ET-026

Issue: A statistical comparison reported as not having been made. The review states twice that Allen et al.’s “results were not compared statistically”. Allen reports suicidality significantly lower after treatment, F(1,44) = 15.09, p < .001, and well-being significantly higher, F(1,44) = 11.39, p < .002.
Level: Substantive · Status: Unchanged

ET-027

Issue: The comparator described as receiving no hormones. The executive summary and result headings describe the contrast as “oestrogen monotherapy vs no hormones”. The review’s own evidence-table footnote records that 10 of the 17 comparator participants were receiving GnRH analogues or spironolactone.
Level: Substantive · Status: Unchanged

ET-028

Issue: A standard deviation inconsistent within the document. The GRADE profile gives the body image dissatisfaction SD as 16.21, matching the source; the results section gives 16.31.
Level: Clerical · Status: Unchanged

ET-029

Issue: Kramer et al. placed under “oestrogen monotherapy”. The review’s own footnote records that the authors “do not specify the GAH used” and that hormone-use data were missing for 63 participants; the paper describes the exposure as “prior use of gender-affirming hormones”.
Level: Substantive · Status: Unchanged

ET-030

Issue: Grannis et al. cited under three different years within one document — 2023 in the study list and bibliography, 2022 in the results statements, 2019 in the under-15s section.
Level: Clerical · Status: Unchanged

URN 2417j — testosterone monotherapy

Reported 12 July 2026.

ET-031

Issue: Testosterone ester names transposed. The review gives “3-weekly intramuscular testosterone undecanoate” (n=63) and “3-monthly enanthate or mixed esters” (n=78). Moussaoui et al. (Int J Transgend Health 2024;25(1):10–18) reports 3-monthly undecanoate (n=78) and 3-weekly enanthate or mixed esters (n=63). Intervals and counts are paired correctly in both; the drug names appear to have been exchanged, so each ester carries the other’s interval and denominator. Undecanoate is the long-acting ester.
Level: Substantive · Status: Unchanged

ET-032

Issue: Final adult height reported as lower than baseline height — “mean 164.1 (SD 6.8) cm vs 164.7 (SD 6.7) cm, respectively”.
Level: Clerical · Status: Unchanged

ET-033

Issue: A height Z-score of 21.0 — “mean 0.21 (SD 1.0) vs 21.0 (SD 1.0), respectively”.
Level: Clerical · Status: Unchanged

ET-034

Issue: Statistical significance asserted where no statistical measures were reported. Both height statements conclude “no statistically significant difference” and then state “No statistical measures were reported.”
Level: Substantive · Status: Unchanged

ET-035

Issue: Haemoglobin reported in mg/dL. Values of approximately 13.2–14.2 mg/dL appear in the results and GRADE tables; the conventional unit is g/dL.
Level: Clerical · Status: Unchanged

URN 2417l

Reported 14 July 2026.

ET-036

Issue: The included study cited with two different years. Section 4 reads “One study was identified for inclusion (Valentine et al 2024).” Table 1, the Results and the Executive Summary all cite Valentine et al 2022. Same study; the 2024 date corresponds to no publication.
Level: Clerical · Status: Unchanged

ET-037

Issue: A results heading that does not match its result. Under “Oestrogen with GnRH analogues vs no hormones”, a bullet headed Dyslipidaemia is followed by a hypertension result — OR 2.1 (95% CI 1.2 to 3.6), p<0.01. No hypertension heading appears, and no dyslipidaemia result is reported anywhere in the review. The equivalent structure in 2417n is correct.
Level: Substantive · Status: Unchanged

PICO, consultation and excluded-papers documents

ET-038

Documents: PICO documents for 2417j and 2417n
Issue: “Feminising medicines” in the masculinising PICOs. The cognitive-outcomes rationale reads “…because feminising medicines can negatively impact cognitive processes”. Both are masculinising PICOs; the evidence reviews read “masculinising medicines”.
Level: Clerical · Status: Unchanged · Reported: 12 July 2026

ET-039

Document: Consultation guide, Appendix C
Issue: Population described as “children and young adults” for items 7 and 8. The PICOs concern people up to their eighteenth birthday.
Level: Clerical · Status: Unchanged · Reported: 12 July 2026

ET-040

Document: Excluded papers, PRN02420_iv
Issue: A duplicate table label in the Contents. Table 4 is 2417h; Table 10 is also given as 2417h. Table 10 sits between Table 9 (2417m) and Table 11 (2417o), so on the sequence it can only be 2417n. As printed, 2417h appears to have two exclusion tables and 2417n none.
Level: Clerical · Status: Unchanged · Reported: 14 July 2026

ET-041

Document: Excluded papers, PRN02420_iv
Issue: The same paper excluded twice. Entries 63 and 64 are both Baltieri DA et al., J Sex Med 2009;6(11):3214-20 — same journal, volume, issue and pages. Two database records of one paper, screened and excluded as two.
Level: Clerical · Status: Unchanged · Reported: 14 July 2026

ET-042

Document: Glossary, across the review series
Issue: Two definitions. p value is defined as the “probability of obtaining these results by chance”; it is the probability of observing data at least as extreme as those observed, assuming the null hypothesis and the model are correct. Statistical significance is defined as a result “assessed as being due to a true effect rather than random chance”; significance does not establish that a true effect exists. Raised because the reviews rely substantially on significance language.
Level: Substantive · Status: Unchanged · Reported: 12 July 2026

One further point was put as a query rather than a correction and is not counted as an item. In Appendix D of 2417j, Chu et al. is recorded as excluded with the reason “Larger, higher level evidence identified reporting on incidence of acne”, while the retained result is described in the same document as “progression of acne” in 77 of 119 participants. Incidence and progression are not equivalent. Clarification was requested; none has been received.

NICE

Reported: 2 items · Entries: 2 · Declined

Reported 12 July 2026. Both concern López de Lara et al. 2020 (An Pediatr 2020;93(1):41–48), the single study on which the review’s entire gender dysphoria outcome rests.

ET-004a

Document: Evidence review: Gender-affirming hormones for children and adolescents with gender dysphoria (October 2020), prepared for NHS England · Ref: EH-348012
Issue: The Utrecht Gender Dysphoria Scale was administered in two different forms. The review reports a reduction from 57.1 (±4.1) to 14.7 (±3.2) points at 12 months, “below the threshold (40 points) for gender dysphoria”. The paper discloses in its Methods that the scale was completed for the sex assigned at birth at T0 and for the self-identified gender at T1. The UGDS exists in two sex-specific forms with differently framed items, so the two scores are not on the same instrument; the difference is not a change score, and the 40-point cut-off cannot be applied across two forms.
Level: Conclusion-affecting · Status: Declined
Reported: 12 July 2026 · Acknowledged: 15 September 2026 · Resolved: 15 September 2026, declined

ET-004b

Document: As ET-004a
Issue: The study described as having no control group. The GRADE footnote reads “high risk of bias (poor quality overall; lack of blinding and no control group)”. The paper is a prospective study of 23 transgender adolescents and 30 cisgender controls, and the authors’ lead finding is that the groups differed at baseline and equalised at one year. If what was meant is that there is no untreated comparator from the same population, that is correct and more important — but it is not what the footnote says, and the study has been described as uncontrolled downstream of this review.
Level: Substantive · Status: Declined
Reported: 12 July 2026 · Acknowledged: 15 September 2026 · Resolved: 15 September 2026, declined

Reasons given, in full: the reviews were commissioned by NHS England and produced by NICE; they are no longer on the NICE website because the evidence service programme closed on 31 March 2022; they were superseded by the systematic reviews commissioned by the Cass Review from the University of York; comments will be retained on file; there is no intention to update the documents or issue any correction or addenda.

Interval from report to reply: sixty-five days. Replied to the same day, nine minutes after the reply arrived.

This is the most structurally informative entry in the register, and it should be read as a finding rather than a refusal. The document remains in circulation and citable. The programme that produced it no longer exists. No successor body has taken ownership of its contents. A defect reported into that document has nowhere to go — not because anyone declined to act, but because acting is nobody’s job. An orphaned document with no owner is a different failure from an uncorrected one, and it is invisible to every reader of the document itself.

Our reply added one fact and requested no action: the review is no longer on the NICE website, but copies remain in circulation, the companion review on GnRH analogues appears among the materials on NHS England’s consultation platform, and the gender-affirming hormones review was cited with a direct link to a mirrored copy as recently as June 2026. “Superseded” describes a document’s status within an organisation. It does not describe whether the document is still being read and relied upon.

European Journal of Endocrinology

Entries: 2 · Correction directed

ET-003

Document: van der Loos et al., The natural course of bone mineral density in transgender youth before medical treatment, Eur J Endocrinol 2024;191(4):426–432 · DOI: 10.1093/ejendo/lvae126
Issue: The abstract and the Results disagree for three of the four regions of interest, AMAB, unadjusted, aged 12 to 22.
Level: Substantive · Status: Correction directed
Reported: 13 July 2026 to the author; 2 September 2026 to the journal
Acknowledged: 4 September 2026 by the author; 8 September 2026 by the journal
Resolved: Not yet — corrigendum directed 9 September 2026, article unchanged

Lumbar spine
Abstract: −0.13 (95% CI −0.17; −0.10) · Results: −0.13 (95% CI −0.17; −0.09)

Total hip
Abstract: −0.05 (95% CI −0.08; −0.02) · Results: −0.04 (95% CI −0.08; −0.01)

Femoral neck
Abstract: −0.06 (95% CI −0.10; −0.03) · Results: −0.06 (95% CI −0.09; −0.03)

Total body less head
Identical in both, as are the figures for individuals assigned female at birth.

ET-003b

Document: As ET-003
Issue: For individuals AMAB aged 22 to 25, the total hip P value is printed as “P = 0.” with no digits following, and without the comma preceding the other three P values in the same sentence. A truncation at typesetting.
Level: Clerical · Status: Correction directed
Found: 1 September 2026 · Reported: 2 September 2026

The corresponding author was written to on 13 July 2026; an automatic reply indicated absence for personal reasons and no follow-up was sent. The journal was written to on 2 September with the author copied. The author replied on 4 September, undertaking to trace how the discrepancy occurred. The journal acknowledged on 8 September, and on 9 September wrote to both parties stating a corrigendum needs to be initiated. As of 30 September 2026 the published article still carries the abstract as printed. Verified against the article page that day: the abstract gives LS −0.13 (95% CI −0.17; −0.10), TH −0.05 (95% CI −0.08; −0.02) and FN −0.06 (95% CI −0.10; −0.03), unchanged, and no corrigendum notice appears on the page. Six days from letter to acknowledgement; twenty-one days from the corrigendum being directed, with the article unchanged.

New England Journal of Medicine

Entries: 4 · Resolved: 0

Chen et al., Psychosocial Functioning in Transgender Youth after 2 Years of Hormones, N Engl J Med 2023;388:240–250, doi 10.1056/NEJMoa2206297. Corrected 18 October 2023 (N Engl J Med 2023;389:1540, doi 10.1056/NEJMx230007). The article page carries a separate revision note dated 19 October 2023; the two are different events and both are true.

ET-005

Document: The free author manuscript deposited in PubMed Central
Issue: The version of record was corrected on 18 October 2023. The deposit carries the entire pre-correction text and no notice on the manuscript itself that a correction exists.
Level: Substantive · Status: Unchanged
Reported: 13 July 2026 · Acknowledged: None

Positive affect, annual increase
Deposit: 0.80 (95% CI 0.08 to 1.54) · Corrected: 1.12 (95% CI 0.37 to 1.89)

Observations in the analytic sample
Deposit: 6114 · Corrected: 6138

Severe depression sub-analysis
Deposit: of 27 participants, 18 (67%) · Corrected: of 35, 20 (57.1%)

Table 3
72 of 80 cells differ. Positive affect and life satisfaction differ in all sixteen cells each; the eight identical cells are almost all in the appearance congruence column. Both tables were transcribed in full and compared by script; the corrected side was then checked cell by cell against page 246 of the printed article.

The deposit contains a value that cannot be right on its face. It prints a point estimate of −0.23 with a 95% interval of (−0.08 to 0.36) — an interval that does not contain its own estimate. The corrected record gives (−0.81 to 0.35).

This is not a failure of the journal, which corrected the paper, or of the repository, which stores what it is given. Only the authors can replace a deposit of that kind. As of 28 September 2026 the two versions report materially different numbers to different readers, and the version that is free is the one that is wrong. Seventy-seven days from our notice; two years, eleven months and ten days from the correction being published.

ET-045

Document: Chen et al., version of record — Results text
Issue: The corrected Table 3 does not support a statement the text still makes. The Results read “Depression and anxiety scores decreased among youth designated female at birth but not among those designated male at birth.” In the corrected table the effect of designated sex at birth on the growth slope is 1.86 (0.27 to 3.43) for depression, 1.53 (−0.04 to 3.11) for anxiety and −1.92 (−3.44 to −0.41) for life satisfaction. The depression and life-satisfaction intervals exclude zero; the anxiety interval does not. The correction replaced the table and the appendix but did not alter the text.
Level: Substantive · Status: Reported
Reported: 13 July 2026 · Acknowledged: None

ET-018

Document: Chen et al., corrected version of record, Table 3, page 246
Issue: Conditional model, Time, Slope mean, Positive Affect is printed as 4.29 (95% CI 2.38 to 61.20). An upper bound of 61.20 against a point estimate of 4.29 and a lower bound of 2.38 is not a credible interval, and is an order of magnitude larger than any other bound in the table. The adjacent Life Satisfaction cell reads 4.44 (2.26 to 6.32). The same cell in the deposit reads 1.79 (0.14 to 3.43), so the value was introduced by the correction rather than carried through it.
Level: Substantive · Status: Reported
Found: 28 September 2026 · Reported: 28 September 2026

ET-019

Document: Chen et al., Table 3, page 246 — present in both versions
Issue: Time-invariant effects on intercept, Baseline age, Appearance Congruence is printed as 0.02 (95% CI –002 to 0.07). The lower bound is printed without its decimal point. It predates the correction and survived it.
Level: Clerical · Status: Reported
Found: 28 September 2026 · Reported: 28 September 2026

ET-018 and ET-019 were found and reported on the same day, and the specifics were withheld from this page until the letter was sent. Both were first read from extracted PDF text, which is what produces dropped decimal points, and neither was reported until they had been checked against the printed page.

JAMA Dermatology — and a notice of ours that was wrong

Entries: 2, one struck · Resolved: 0

ET-006 — STRUCK 28 September 2026

Issue: Five figures in a news summary of Smith et al. differ from the source paper, and a sixth cannot be located in it.

Struck because it is false. On 13 July 2026 we told a specialist news outlet that figures in its coverage did not match the paper. Every one was a correct quotation of the paper’s Results. We had compared them against the paper’s abstract. The sixth figure, 7.6%, is in the paper as the moderate-to-severe cumulative incidence in the gender-affirming hormone therapy subcohort; our notice compared it against 5.9%, which is the whole-cohort figure for the same outcome — a different row. The notice was withdrawn in full and unreservedly on 28 September 2026, seventy-seven days after it was sent.

ET-006b

Document: Smith CA, Kaabi O, Manatunga AK, et al., Acne Incidence and Severity in Transgender Individuals, JAMA Dermatology, published online 21 January 2026 · DOI: 10.1001/jamadermatol.2025.5597
Issue: For the gender-affirming hormone therapy subcohort, five of the six hazard ratios are reported differently in the abstract and in the Results.
Level: Substantive · Status: Reported
Found: 28 September 2026 · Reported: 28 September 2026

Transmasculine, first year after testosterone, vs matched cisgender male
Abstract: 8.29 (7.11–9.68) · Results: 8.56 (7.29–10.05)

Transmasculine, first year, vs matched cisgender female
Abstract: 2.63 (2.33–2.97) · Results: 2.63 (2.33–2.97) — identical

Transmasculine, after 1 year, vs matched cisgender male
Abstract: 5.29 (4.45–6.28) · Results: 6.22 (5.07–7.62)

Transmasculine, after 1 year, vs matched cisgender female
Abstract: 1.69 (1.46–1.96) · Results: 1.66 (1.43–1.93)

Transfeminine, after estradiol, vs matched cisgender male
Abstract: 1.56 (1.31–1.84) · Results: 1.64 (1.37–1.96)

Transfeminine, after estradiol, vs matched cisgender female
Abstract: 0.53 (0.46–0.62) · Results: 0.54 (0.46–0.64)

The ordinary explanations were tested first. The comparisons are named the same way in both passages and drawn from the same subcohort; the paper splits the transmasculine analysis at one year because the proportional hazards assumption was violated, and both passages describe that same split. There is no second model, no adjusted-versus-unadjusted pairing and no change of denominator. The whole-cohort ratios — 4.27, 1.50, 2.00 and 0.68 — appear only in the Results.

The outlet is not named, and that was their decision to make rather than ours. Our rule is that institutions are named. Applied here it would put an outlet’s name in an entry recording that we accused them wrongly, so on 28 September 2026 we sent them both entries in full, set out the case for and against naming — including that the entry says plainly their reporting was correct — and told them the default, absent any reply, is that they are not named. That is where it stands.

Primary papers — abstract and Results disagreements

Entries: 4 · Resolved: 0 · All reported 13 July 2026, none acknowledged

Superseding ET-008. An earlier bundle entry recorded “four primary papers, eight items”. The letters as sent do not reconcile to that count, and the entries below are derived from the letters rather than from the summary. The original count is not carried forward.

ET-046

Document: Valentine et al., Multicenter Analysis of Cardiometabolic-related Diagnoses in Transgender and Gender-Diverse Youth: A PEDSnet Study, J Clin Endocrinol Metab 2022;107(10):e4004–e4014
Issue: The abstract and the Results give different confidence intervals for the same estimate. Dyslipidaemia in those prescribed testosterone with a GnRH agonist — abstract 3.7 (95% CI 2.1–6.7), Results 3.7 (95% CI 2.0–6.7). The point estimate and upper bound agree; the lower bound does not.
Level: Clerical · Status: Reported

ET-049

Document: Persky et al., Pubertal suppression in early puberty followed by testosterone mildly increases final height in transmasculine youth, J Endocr Soc 2024;8:bvae089
Issue: The abstract pairs a positive height difference with a wholly negative confidence interval — “FAH was 7.95 cm greater in the GnRHa + T group than in T-only group (95% CI −10.85, −5.06)”. The Results show the regression coefficient is coded for the testosterone-only group (β = −7.95), meaning that group was estimated to be 7.95 cm shorter. The magnitude is unaffected and the Results are internally consistent; the abstract’s point estimate carries one reference group and its interval the other.
Level: Substantive · Status: Reported

ET-047

Document: Kramer et al., Disordered eating in transgender and gender-diverse adolescents, Clin Child Psychol Psychiatry, doi 10.1177/13591045231184917
Issue: Compensatory exercise, transgender female, printed as “4 (29.6%)” alongside “10 (71.4%)”. Four of fourteen is 28.6%. The correct value is printed elsewhere in the same table, where the subjective binge episode row gives the identical cell as “4 (28.6%)”.
Level: Clerical · Status: Reported

ET-048

Document: As ET-047
Issue: Transgender male, printed as “104 (8.19%)” alongside “23 (18.1%)”. One hundred and four of one hundred and twenty-seven is 81.9%, and the figure is printed as “104 (81.9%)” in another row of the same table.
Level: Clerical · Status: Reported

ET-047 is the clearest single illustration of what this register measures, and it runs through us. The 29.6% is the paper’s own typographical error. NHS England’s URN 2417h transcribed it faithfully into its own evidence table — their transcription is accurate, and the error is not theirs. It also appeared in a draft of our own correction notice, and was caught only when the arithmetic was recomputed; it is recorded below as rejected item R-3. One mistyped percentage, copied without recalculation, into three separate documents by three separate parties.

Cass-commissioned systematic review

Entries: 2 · Resolved: 0 · Neither acknowledged

Superseding ET-007, a bundle placeholder. Both entries are derived from the letters as sent.

ET-050

Document: Taylor et al., Masculinising and feminising hormone interventions for adolescents experiencing gender dysphoria or incongruence: a systematic review, Arch Dis Child 2024;109(Suppl 2):s48–s56
Issue: The review’s entire gender dysphoria synthesis rests on López de Lara et al., and records neither the instrument used nor the limitation the paper discloses — that the Utrecht Gender Dysphoria Scale was administered in two different sex-specific forms, one before treatment and the other after. The scores are therefore not on the same instrument and the difference is not a change score. The word “Utrecht” does not appear in the review.
Level: Conclusion-affecting for that outcome · Status: Reported
Reported: 13 July 2026

ET-051

Document: As ET-050, Supplementary Table 2
Issue: Several of the 53 included studies appear to be separate reports of the same cohort, and the review does not say so. Klaver 2018 and Klaver 2020: both n=192, same centre, same subgroups, identical inclusion wording. Burke 2016 and Beking 2020: both n=21, mean age 16.1, same two control groups of the same size and means. Fourteen of the fifty-three list the same centre. The word “overlap” does not appear. The reported total of 40,906 participants is a sum across studies and inherits any duplication.
Level: Substantive · Status: Reported
Reported: 14 July 2026

A judgement recorded against ourselves, on this entry. ET-051 changes nothing in the review: the synthesis is narrative, the appraisal is study-level, and no conclusion depends on the participant totals. The review also decomposes the 40,906 figure honestly in its own next clause. The letter was drafted before anyone asked whether the finding was worth sending — the same failure the rest of this register documents: finding something true and reaching for the strongest available action without first asking what it was worth. It was sent and is not retracted, because it is true and hedged. No third letter will be sent. The rule arising: verification means the document was opened, and then asking whether it matters. An error is not a reason to write to someone. A consequence is.

Haematocrit thresholds — a second evidence review

Entries: 7 · Reported 25 September 2026 · None acknowledged

These arise from an evidence review of haematocrit thresholds in trans-masculine people on testosterone, doi 10.5281/zenodo.22830379, and are the first entries here from a review other than ER-015. Every one was reported on 25 September 2026 and none has been acknowledged.

ET-053

Document: Ory J, Nackeeran S, Balaji NC, Hare JM, Ramasamy R, J Urol 2022;207(6):1295–1301 · DOI: 10.1097/JU.0000000000002437
Issue: In the sensitivity analysis restricted to haematocrit above 54%, the Results text reads “Odds of MACE remained higher in the polycythemia group when defined as hematocrit >54%” and prints OR 0.78 (95% CI 0.62–0.98, p=0.03) in the same parenthesis. An odds ratio of 0.78 with an interval entirely below unity is a lower odds. The sentence and the number disagree. The orientation is established for the main analysis — 301 of 5,842 against 226 of 5,842 gives 1.3499, printed as 1.35, where reversed coding would have printed 0.74 — but the paper does not state whether the same orientation was implemented in the sensitivity analysis, so which limb is in error cannot be determined from the printed number. No claim is made about which limb is wrong, only that the two disagree.
Level: Substantive · Status: Reported

ET-054

Document: Cannarella et al., Testosterone replacement therapy and vascular thromboembolic events, Asian J Androl 2024 · DOI: 10.4103/aja202352 · Supplementary Table 7
Issue: The Endocrine Society row is printed “Hematocrit<48%” under a column headed “Recommendation against TRT”. The 2018 guideline sets a haematocrit above 48% — or above 50% at altitude — as a relative contraindication to starting. The operator is inverted and the altitude qualifier is not carried.
Level: Substantive · Status: Reported

ET-055

Document: As ET-054
Issue: The AUA row is printed >54%. The AUA guideline’s instruction before starting is that above 50% clinicians should consider withholding testosterone until the aetiology is formally investigated; 54% is its on-treatment intervention level. The on-treatment value appears under the initiation heading, and the initiation value is absent.
Level: Substantive · Status: Reported

ET-056

Document: As ET-054
Issue: The AACE row is printed >50%. AACE names polycythaemia a relative contraindication without attaching a number; 50% is its level for decreasing or discontinuing therapy once started. The same substitution of an on-treatment value for an initiation position.
Level: Substantive · Status: Reported

An observation put to the authors and explicitly not raised as an error. Seven of the eight rows checked set an on-treatment action level different from their initiation position, and seven of those eight set it at 54%. Because the table holds one value per society, a reader sees a 48–54% spread and may conclude the societies disagree, when the disagreement is about initiation and the on-treatment threshold is very nearly unanimous. That is a property of any single-column format rather than a defect in theirs — and the same mistake was made in our own work and carried for a long time.

ET-057

Document: Nordestgaard et al., Arterial and venous thrombosis by high platelet count and high hematocrit, J Thromb Haemost 2019 · DOI: 10.1111/jth.14574 · Section 3.3
Issue: The paragraph opens with an adjusted hazard ratio for venous thromboembolism of 1.16 (1.01–1.33) for platelet counts of 183–230 × 109/L, and closes “neither high platelet counts, nor high hematocrit was associated with risk of venous thromboembolism.” The interval excludes unity. The Discussion resolves it — the association appears in the 6th to 24th percentiles and not dose-dependently, with confounding by low-platelet disease offered as the explanation — but the closing sentence, read alone, transmits “no association” from a paragraph whose own figure shows one, and the qualification that makes it defensible is four pages away.
Level: Substantive · Status: Reported

ET-058

Document: As ET-057 · Section 4.3, Limitations
Issue: The text reads “we cannot totally exclude that the lack of statistical significance for risk of arterial thrombosis in the heart in those with high hematocrit values represent a type II error.” Figure 2B gives that hazard ratio as 1.46 (1.06–2.00), and the abstract, the Essentials box, the Discussion’s opening and the Conclusion all report an approximately 1.5-fold risk. There does not appear to be a lack of significance to exclude. Three readings would each make the sentence coherent — a substitution of “platelet count” for “hematocrit”, or reference to the competing-risk estimate 1.38 (1.00–1.90), or to the cumulative-incidence estimate 1.34 (0.98–1.82) — and the paper does not disambiguate. None was assumed.
Level: Substantive · Status: Reported

Two further observations in the same letter, offered and explicitly not relied upon. Figure 1A gives a hazard ratio of 1.24 with an interval of 0.99–1.35, whose geometric centre is 1.156; all twenty-four hazard ratios in Figures 1 and 2 were checked and every other row agrees with its own interval to within 0.013, which suggests a single cell rather than a typesetting pattern. And Section 3.1 gives incidence rates of 40 and 25 per 10,000 person-years where Figure 2A gives the reference band as 26 (25–28) and the abstract gives 40 and 26. Which of the three numbers is wrong is not determinable from the paper, and no inference was offered.

ET-059

Document: EAU Guidelines on Sexual and Reproductive Health 2026 (limited update March 2026)
Issue: The guideline describes Ory’s sensitivity analysis at haematocrit above 54% as showing even higher risk. That is an accurate reproduction of Ory’s own sentence — and the opposite of Ory’s own number. The guideline carries the prose limb without the interval, so a reader has no way to see that the source disagrees with itself. The adjacent limb is handled differently: “No risk was observed when a 50% threshold was applied” is an accurate reading of Ory’s figure for that threshold, OR 0.90 (0.78–1.03). The panel followed the source’s figures for one threshold and its prose for the other, in a single sentence.
Level: Substantive, in the source rather than the guideline · Status: Reported

ET-059 is the clearest transmission entry in this register, and nothing in the guideline is careless. The panel reproduced its source faithfully. The defect is in the source; the guideline’s format — prose without the interval — is what makes it invisible to everyone downstream. A reader reaching the guideline has no way to see the disagreement, and a reader reaching the paper can see it only by comparing a sentence against the number in its own parenthesis. This is the mechanism the whole register is about, caught in one sentence of one guideline, with no one at fault at any hop.

Department for Education — KCSIE 2026 policy paper

Entries: 1 · Resolved: 0 · Not acknowledged

Reported to the Teacher Regulation, School Safeguarding and Safety Team on 3 September 2026, with a correction to our own description of the statutory text on 11 September 2026. This is the first entry here arising from a guidance appraisal rather than an evidence review: GA-001, doi 10.5281/zenodo.22703225. The full correspondence, including a request for data that is recorded there rather than here, is on KCSIE 2026 and children questioning their gender.

ET-064

Document: Department for Education, Including guidance on children who are questioning their gender in ‘Keeping children safe in education (KCSIE)’, policy paper, 12 February 2026
Issue: The policy paper states that schools must not permit children over the age of 8 to access toilets designated for the opposite biological sex, and children over the age of 11 to access changing rooms designated for the opposite biological sex. The statutory guidance in force from 1 September 2026 puts no age qualifier on either prohibition: paragraph 108 prohibits access to toilets designated for the opposite biological sex; the final sentence of paragraph 115 prohibits access to changing rooms designated for the opposite sex where a school is supporting any degree of social transition; and paragraph 276 prohibits access to toilets, changing rooms and boarding or residential accommodation designated for the opposite sex, with no exceptions. The ages in the statutory text attach to different duties — providing separate toilets for pupils aged 8 and over (paragraph 109), and a child aged 11 or older at the start of the school year undressing in front of a child of the opposite biological sex (paragraph 115). The paper remains published without annotation. A school relying on it alone would take a narrower rule than the one in force.
Level: Substantive · Status: Reported
Reported: 3 September 2026 · Acknowledged: None

Which kind of defect this is has not been established, and the entry does not assume it. The paper was published at consultation stage and describes what was then a proposal. If the proposal carried those thresholds and the enacted text dropped them, the paper is an accurate account of a superseded draft, and the defect is that it remains published as though current — the same class as the NICE entries above, a document that has outlived its own accuracy. If the proposal did not carry them, the paper misdescribes it. Our fourth question asks the Department which. Either way the paper and the statutory guidance disagree, and only one of them is in force.

Our own letter of 3 September described paragraph 115 incompletely, and described our appraisal as in preparation when it was already published. Both are recorded below, as ET-061 and ET-060. The correction narrowed nothing: it made the disagreement sharper, because the statutory text prohibits access to changing rooms without any age threshold.

International Journal of Transgender Health — a scoping review

Entries: 2 · Resolved: 0 · Not acknowledged

Reported to the corresponding author on 11 October 2026. Found while reading the review and both primary papers in full on 11 October 2026, for our evidence to the UN, which cites the review.

ET-069

Document: Fuller C, et al., Gender-affirming healthcare for incarcerated transgender and gender diverse people: An international scoping review, International Journal of Transgender Health 2026;27(1):143–161, doi 10.1080/26895269.2024.2443753
Issue: The Results (p. 147) say that in McCauley et al. (2018) “seven out of 10 trans women reported HRT use within prison and of these seven, six experienced a disruption of their medication upon intake.” McCauley et al. interviewed women in a county men’s jail, and the seven participants’ hormone use was before they entered custody; six had it disrupted on intake (McCauley, Results, p. 38). The review’s own Table 1 (p. 149) gives the setting correctly as a county jail.
Level: Substantive · Status: Reported
Reported: 11 October 2026 · Acknowledged: None

ET-070

Document: As ET-069
Issue: Table 1 (p. 148) gives Drakeford (2018) a sample of 6450, and the Results (p. 147) give the review’s range of sample sizes as “10 to 6450”. 6,450 is the whole 2008–09 National Transgender Discrimination Survey sample. Drakeford analysed the 1,007 respondents who had ever been sent to jail or prison (p. 173), with regression models of N = 913 (Table 2) and N = 942 (Table 3).
Level: Substantive · Status: Reported
Reported: 11 October 2026 · Acknowledged: None

National Commission on Correctional Health Care — a position statement

Entries: 2 · Resolved: 0 · Routing reply only

Reported to the Commission’s general address on 11 October 2026, the only contact it publishes. Found while reading the statement in full on 11 October 2026, for our evidence to the UN, which cites it. The web version was checked the same day and carries the same definition.

ET-071

Document: National Commission on Correctional Health Care, Transgender and Gender Diverse Health Care in Correctional Settings, position statement, reaffirmed with revision 31 July 2026
Issue: The Definitions (p. 3) define a transgender woman/transfeminine individual as “a person with female gender identity who was recorded/designated female sex at birth.” It should say male. As printed, it repeats the sex-at-birth wording of the transgender man definition above it, and contradicts the statement’s own umbrella definition.
Level: Clerical · Status: Reported
Reported: 11 October 2026 · Acknowledged: None. Routing reply on 13 October 2026: forwarded “to the appropriate staff person for their careful review and consideration”.

ET-072

Document: As ET-071
Issue: Two DOIs in the references (p. 5) appear incomplete as printed: reference 1 (Murphy et al., 2023), 10.1089/jchc.21.09.00, and reference 3 (Kendig et al., 2019), 10.1177/107834581985711, one character shorter than the journal’s DOI format. We have not established the correct DOIs and did not supply them.
Level: Clerical · Status: Reported
Reported: 11 October 2026 · Acknowledged: None. Routing reply on 13 October 2026: forwarded “to the appropriate staff person for their careful review and consideration”.

Good Law Project and QueerAF — two news articles

Entries: 5 · Resolved: 0 · Not acknowledged

Content note: these entries concern figures for suicides, including the deaths of children.

Found while preparing Evidence Note EN-025, Five numbers, five ways of counting, which examines both articles against the documents they draw on. Reported to QueerAF on 12 October 2026, copying the QueerAF journalist who co-wrote both articles; Good Law Project was not separately contacted. On 13 October 2026 the journalist replied that a fuller reply would follow. EN-025 describes that reply as an acknowledgement, in the ordinary sense. Under this register’s definitions, where an acknowledgement is a substantive reply, it is a holding reply, and the entries stand at Reported until the fuller reply arrives. The articles are identified by title and not linked, following the Samaritans media guidelines on reporting suicide.

ET-073

Document: Good Law Project, Trans people are dying of suicide more than the general UK population, data shows, 25 April 2026 (produced with QueerAF)
Issue: The article gives 47 trans child suicides in “England and Wales”. The National Child Mortality Database response it draws on gives 46 deaths meeting NCMD’s umbrella definition of trans, in England only. The same article elsewhere gives 46 for England.
Level: Substantive · Status: Reported
Reported: 12 October 2026 · Acknowledged: None (holding reply 13 October 2026; see above)

ET-074

Document: As ET-073
Issue: The article describes the National Confidential Inquiry’s figure of 104 trans patients as covering “a seven-year time period”. The Inquiry’s period is 2016 to 2023, which is eight years.
Level: Substantive · Status: Reported
Reported: 12 October 2026 · Acknowledged: None (holding reply 13 October 2026; see above)

ET-075

Document: As ET-073
Issue: The article gives the National Confidential Inquiry’s figures as 16 in 2021, 34 in 2022 and 20 in 2023. None of the Inquiry’s annual reports for 2023 to 2026 gives a single-year count for trans patients. Subtracting successive running totals gives 16, 31 and 20, but the Inquiry revises earlier figures, so none is a reliable yearly count.
Level: Substantive · Status: Reported
Reported: 12 October 2026 · Acknowledged: None (holding reply 13 October 2026; see above)

ET-076

Document: As ET-073
Issue: Under the headline that trans people are dying of suicide more than the general population, the article sets trans people’s share of the population, “less than 1%”, beside trans under-18s’ share of child suicides, “7%”. The 7% is a share of child deaths in England classed as suicide or deliberate self-inflicted harm (46 of 647). The under-1% is an estimate for people aged 16 and over, under a different definition, covering England and Wales. Both figures are quoted correctly; set side by side, they cannot show a difference in suicide rates.
Level: Conclusion-affecting · Status: Reported
Reported: 12 October 2026 · Acknowledged: None (holding reply 13 October 2026; see above)

ET-077

Document: QueerAF, The deadly toll of the trans healthcare crisis: 21 suicides and over 600 other deaths, 10 October 2026
Issue: The headline gives “over 600 other deaths”. The article discloses at least 618 deaths in total, including at least 21 suicides or suspected suicides, so 597 deaths in the disclosed total were not among the 21. Because that total is incomplete, this does not show that the true number of other deaths is only 597.
Level: Substantive · Status: Reported
Reported: 12 October 2026 · Acknowledged: None (holding reply 13 October 2026; see above)

Accusations of ours that were wrong

Four · Three caught before sending · One sent and withdrawn

A register that contains its own failed accusations is the only kind anyone should believe. These four were found during the ER-015 verification build in July 2026 and are carried with their original identifiers.

R-1 — rejected before sending

Claimed: a comparator group of n=177.
True: n=17, followed by superscript footnote marker 7. A PDF-extraction artefact. Claimed twice by the reviewer.

R-2 — rejected before sending

Claimed: Valentine’s participant counts do not reconcile.
True: 832 + 349 + 106 + 125 = 1,412 exactly. The apparent discrepancy of 267 is the GnRH-analogue-alone group, and a blocker alone is not gender-affirming hormone therapy. This was headed for the top of the correction notice.

R-3 — rejected before sending

Claimed: an impossible 29.6% in NHS England’s evidence table.
True: Kramer’s own typographical error, transcribed faithfully by NHS England. It was in our draft correction notice and was killed by recomputing the arithmetic. See ET-047.

R-4 — rejected before sending

Claimed: Valentine’s adjusted results were unavailable.
True: the adjusted testosterone odds ratios were published; only the oestradiol analysis went null. Retracted by the reviewer.

Three of these four were caught before a letter went out. One was not, and it is ET-021. That is the sequence worth looking at: a process that rejects four accusations in a fortnight, then fails to reject a fifth and sends it. The failure is not that the register produced a wrong claim — it produced at least five — but that one of them escaped the check that caught the others.

Our own errors

Entries: 23 · Corrected: 18 · Superseded: 1 · Struck: 1 · Withdrawn: 1 · Reported: 1 · Open: 1

Ours are in the same register, with the same levels, the same dates and the same styling. A register that holds other people’s documents to a standard and keeps its own defects somewhere quieter is making an argument about who is trustworthy, rather than measuring what happens to errors.

ET-021 — a false notice

Issue: Our notice of 13 July 2026 to a specialist news outlet alleged discrepancies that do not exist and a figure that could not be found in a paper that contains it. All were correct quotations of the paper’s Results; we had compared them against the abstract. The paper was not held in our archive and had not been read.
Level: Conclusion-affecting · Status: Withdrawn
Found: 28 September 2026 · Resolved: 28 September 2026 — withdrawal sent, 77 days after the notice

ET-022 — the rule was already written

Issue: The published Deposit Standard already required what ET-021 failed to do: “Verification means the document was opened. An error is not an error until we have read the source and checked the claim against it. Anything unverified is published nowhere and reported to no one.” It also already names the exact failure: “‘Read in primary’ is not the same as ‘quoted from the primary’ … figures come from the Results. Where an abstract and a results section disagree, the results text governs.”
Level: Conclusion-affecting · Status: Open
Found: 28 September 2026

The Deposit Standard was first published on 13 July 2026. The false notice went out on 13 July 2026. The rule was published the same day it was broken.

That standard’s own second rule is a check that is built and not read is not a check, written from an incident in which a claim-to-reference map recorded a misalignment in plain sight and nobody read it. This is the same failure one level up: a rule written, published, and not applied by the person who wrote it. The gap is not in the rules. It is that correspondence has no build step, so nothing refuses.

Three changes follow, and none is a new rule:

ET-063 — we published a day before the date we gave

Issue: On 28 September 2026 we wrote to the outlet whose coverage we had wrongly accused, set out in full the entries concerning it, and said: “The page publishes on Thursday 1 October 2026. If anything above is unfair, or wrong, or you would like it worded differently, tell me before then and I will change it.” This page was published on 30 September 2026. The window we had offered was a day shorter than the window we stated, and it was shortened by us, to publish sooner, against the interest of the party we had already wronged once.
Level: Substantive
Found: 30 September 2026 · Status: Reported — disclosed to the outlet on 30 September 2026

There is no remedy available for this one. A day cannot be given back, and withdrawing the page to reinstate the window would be a second unilateral decision rather than a repair. What is left is to record it here, at the same level and in the same section as everything else of ours. It is also the second time in three days that this outlet has been on the wrong end of a decision we took without it — the first was a false accusation that stood for seventy-seven days, and this is a commitment withdrawn a day early. The pattern is the same one the register exists to measure: our own undertakings are the ones least likely to have a build step that refuses.

ET-066 — a protection the Bill does not contain

Document: Eden Openly, The Threshold (2/4): What the Conversion Practices Bill Actually Does, 25 June 2026
Issue: The piece said the draft Conversion Practices Bill defines a conversion practice as conduct intended to change “or suppress” a person’s sexual orientation or transgender identity. Clause 1(2) covers conduct intended to cause a person to have or not to have, or to believe that they have or do not have, an orientation or identity; it does not mention suppression. The piece also described the health care exclusion in clause 1(3) as applying to exploratory or affirming care, which has no conversion intention and is outside the definition anyway; the exclusion applies to conduct carried out with a conversion intention in the course of health care. Both errors made the Bill look more protective than it is, in a series arguing that its threshold is set too high. The first is the gap our written evidence to the Joint Committee asks Parliament to close. Found while drafting that evidence, when the Bill’s text was read against our own description of it.
Level: Substantive · Status: Corrected 10 October 2026
Found: 10 October 2026

ET-067 — frequencies we could not source, and pages out of step

Documents: Eden Openly, Done carefully enough, it is not a crime; The Conversion Practices Bill (overview); The Threshold, parts 1 to 3, its landing page and the briefing hub
Issue: Parts 1 and 3 of The Threshold, its landing page and the briefing hub said conversion practice takes quiet, sustained forms “most often” or “far more often”, that “a great deal” of it seeks suppression, and that “a majority of survivors” named family; no source we hold measures those proportions, and the family figure came from a general question in Galop’s survey. Our ET-066 correction itself said the Bill “does not cover” conduct intended to make someone hide who they are, where it may not cover conduct with that aim alone. The companion piece called Germany’s under-18 prohibition “outright” (parents are exempt unless they grossly breach their duty of care), listed Malta as naming suppression (its word is “repress”), misquoted clause 2(1), mislocated Galop’s methodology, described its own alignment with our draft evidence more fully than it was, and differed from our asks page on single conversations and consent. The overview said careful practice “sits under” the threshold rather than “can sit under” it. Found in an external review on 11 October 2026; several points had been raised in an earlier round and not fully carried through. A second review the same day found that the correction itself had introduced two overstatements: Galop’s finding that about four in five cases involved multiple attempts was restated as “only about one in five” being a single incident, which Galop does not say; and the 2021 research’s “may be providing” became “are carried out by” healthcare professionals. Both are corrected, with notes on the pages. On 12 October 2026 a further line in Part 2, saying the Bill “distinguishes” exploratory care from conduct with a conversion objective, was corrected to state that clause 1(3) also excludes the latter in health care unless it falls far below expected standards.
Level: Substantive · Status: Corrected 11 October 2026
Found: 11 October 2026

ET-068 — a clinical piece that stated its evidence more firmly than the sources do

Document: Eden Openly, Stopping Is Not Neutral (Anatomy of Good Care, 1/7), 29 June 2026
Issue: Re-reading every source while preparing our evidence to the UN found eleven defects. Apart from one citation error, every one made the evidence look more supportive of the piece’s argument than it is. The piece said the UCSF guidelines “recommend” bone-density testing after gonadectomy and five years without replacement; they say such people “should be considered” for it, and that the evidence to guide testing is insufficient. It omitted that the same guidelines find no evidence either way on continuing or stopping hormones in older trans women, and call it reasonable to consider stopping around 50 after gonadectomy. It said bone loss after stopping menopausal hormone therapy mirrors that in “early menopause”, where the review compares it with the first two years after menopause, and it omitted the review’s statement that short-term fracture risk does not seem to rise after stopping. It drew an inference from a Dutch fracture cohort without giving the authors’ own explanation or the design, and without labelling the inference as ours. It gave a low-bone-mass figure without its reference range. It cited a cohort for improvement in anxiety that the cohort did not find, with a source note claiming a statement the paper does not make. It described a trial’s suicidality result as clinically significant. It rested one sentence on a case series we do not hold and had not read. And it cited one paper without authors and with the wrong volume year. Found 10 and 11 October 2026 while preparing the submission; each item was logged as found.
Level: Substantive · Status: Corrected 11 October 2026 (v1.1), with a full note on the page. Three vasomotor sources have not yet been re-read against the piece; that check is outstanding and the page says so.
Found: 10 October 2026

Errors made in building this register

ET-016 — STRUCK 28 September 2026

Issue: the Chen correction dated 18 October 2023 in our records; the article page reads 19 October. This entry was itself wrong. 18 October is the correction’s publication date; 19 October is the article page’s separate revision note, and the footer read as evidence was the print issue date for volume 389 number 16. The question had already been adjudicated on 1 September and recorded “so it is not re-queried”. See ET-052.

ET-052 — a correction that corrupted a correct record

Issue: ET-016 reopened a settled question, reversed a correct finding, and published the wrong date into this page’s Chen entries. It was made by reading a page footer without checking whether the question had already been answered.
Level: Substantive · Status: Corrected 28 September 2026

ET-043

Issue: This page attributed the ascertainment defect to URNs 2417h, 2417j and 2417l. The notice as sent says 2417h, 2417l and 2417n. A URN was wrong and a document was missing, in an entry split by document precisely so each could be tracked separately.
Level: Substantive · Status: Corrected 28 September 2026

ET-044

Issue: This page stated ten own-error entries where there were eleven, and omitted the withdrawn status from the breakdown. The one count on the page derived from neither the page nor a source was the one that was wrong.
Level: Clerical · Status: Corrected 28 September 2026

ET-065

Issue: ET-060 dated our first letter to the Department for Education 3 September 2026 and its replacement 4 September, so a same-evening replacement read as a next-day correction. Both went on 3 September, forty minutes apart, at 21:25 and 22:05 UTC. The first date was taken in UTC and the second from the sending client’s +0900 date header. Our correction of 11 September refers to the replacement as “my email of 3 September 2026”, which is consistent with UTC; the error was in the entry, not the correspondence.
Level: Clerical · Status: Corrected 10 October 2026

The remaining entries

ET-009 — a review of ours quoted a paper’s abstract rather than its Results. Correcting that is what made the journal’s own discrepancy visible. Substantive · corrected.

ET-010 — five register defects corrected before publication: a published identifier that could not be sourced to anyone; a wrong digital object identifier, ours; an attribution to the wrong outlet, where the outlets named had reported the figure correctly; an item upgraded only after the file was opened; and an item struck for having no traceable provenance — in a register whose subject is claims entering documents without provenance. Clerical to substantive · corrected.

ET-011 — superseded by ET-021. Recorded the notice to the dermatology outlet as having overstated its case: claiming every value differed when one was identical, omitting a sixth figure, and asserting a discrepancy for a figure that could not be located. It did not overstate a real finding. There was no finding, so the entry is superseded rather than corrected.

ET-012 — a confidence interval under-recorded in our own working register, surfaced only when the archived paper was opened, having been drafted twice from summaries. Substantive · corrected before the letter was sent.

ET-013 — a letter logged as sent on the date it was verified rather than the date its header shows. One day, and it moved a scheduled re-check. Clerical · corrected.

ET-014 — a published page of ours described internally as wrong, with a same-day fix recommended. On fetching it, the sentence was date-anchored and true. No change was required or made; the error was in the advice. Substantive · corrected 7 September 2026.

ET-015 — a journal article’s digital object identifier transcribed wrongly in the first draft of this register, in the document whose subject is identifiers entering records unchecked. Clerical · corrected 28 September 2026.

ET-017 — the Chen table divergence recorded as ten differing cells, from cropped page images. The full comparison gives 72 of 80. Wrong by a factor of seven, and understating the finding. Substantive · corrected 28 September 2026.

ET-020 — a confidential approach carried as an item to be published, then proposed for inclusion “by role and channel” as an anonymisation. In a field this small that is a disclosure with a covering on it. Nothing was filed and nothing was published. Conclusion-affecting for scope · corrected 28 September 2026.

ET-060 — an email to the Department for Education of 3 September 2026 set out its points less precisely than intended, and described an appraisal of the guidance as “in preparation” when it was already published. Superseded by a corrected version sent forty minutes later the same evening, which states the error in its own opening paragraph rather than dropping it quietly. Substantive · corrected 3 September 2026. This entry previously dated the correction 4 September; see ET-065.

ET-061 — the same email described paragraph 115 of the statutory guidance as concerning a child aged eleven or older undressing in front of a child of the opposite biological sex, and contrasted that with the policy paper’s description of a rule governing access to changing rooms. The description was incomplete: the final sentence of paragraph 115 does provide that a school must not allow a child access to changing rooms designated for the opposite sex, and paragraph 276 prohibits access to toilets, changing rooms and residential accommodation with no exceptions. Corrected to the Department on 11 September 2026, with the reworded question supplied and the other five left standing. Substantive · corrected 11 September 2026.

ET-062 — this page stated at section level that every NHS England entry was acknowledged on 13 July 2026. Seven of them were reported on 14 July and could not have been acknowledged the day before. The acknowledgement on file answers the first notice only; the five items of the second have never been acknowledged. The error overstated the recipient’s engagement rather than their fault — it showed seven entries as answered which were not, on a page whose central measurement is whether anyone answers. Substantive · corrected 29 September 2026.

What the dates show about our own process

Three entries were found and reported on the same day: ET-006b, ET-018 and ET-019. In each, the source was open before the letter was written.

ET-021 was the opposite — the source was not held, the notice went anyway, and it took seventy-seven days to correct.

The latency from finding to reporting collapses when the archive gate is honoured, and the latency from error to withdrawal expands when it is not. That is measurable in this register, on our own entries, and it is the only part of the programme where we control both ends of the interval.

On the direction of our own errors

Of the entries that carry a direction, four overstate — declaring a correct page wrong, treating a confidential approach as publishable, and the false notice twice over. One understates, and badly. One, ET-052, reversed a correct finding into an incorrect one. ET-066 understated: it credited a Bill we were criticising with a protection it does not contain. ET-067 mostly overstated, in our own argument’s favour: it claimed frequencies no source measures, and made other countries’ laws look stricter than they are. The rest are clerical.

One, ET-062, ran the other way entirely: it overstated a recipient’s engagement, showing seven entries as acknowledged when they had not been. So the lean is predominantly, but not uniformly, toward finding more fault than is there. That is stated at exactly that strength. The Deposit Standard already warns that a record in which every error points the same way is a record nobody should believe — and the earlier internal register records that in the ER-015 build our errors ran the other way, overclaiming bone and height findings in the treatment-favourable direction. Across the whole programme the direction is genuinely mixed, and anyone reading only this page would get a narrower picture than the evidence supports.

The mechanism, unlike the direction, is uniform. Most of these are the same act: a claim made about a document without opening the document. In several cases the document was already in our own archive or our own sent folder.

Findings that are not defects

These are true, checkable, and were never put in a letter, because they cannot be fixed by editing a document. They are listed so this page does not imply everything found was an error.

Three exclusions are the clearest case. Three times, a study answering the question the policy turns on was excluded by a frame that could not accommodate it. Nobody did anything wrong. That is what makes it drift rather than misconduct — and it is why these are not in the register proper.

Methodological criticism

Arguable, and a competent person could disagree. Never sent in a letter, because a criticism is not a correction.

One argument is deliberately not made: that impossible values on one page establish a failure of validation. It is available to any reader. It is not ours to make.

What is not on this page yet

Seventeen unverified items. Held in the internal register, published nowhere and reported to no one, under a standing rule that nothing unverified may leave it. They are not listed because listing them would be reporting them.

Requests for data, and correspondence that is not a defect report. Submissions to Parliament, consultations, and requests for disclosure are recorded separately. They have a statutory clock and a disclosure outcome rather than a correct-or-decline axis, and putting them here would make the status field mean two different things.

Findings held for a methodological paper. Some of what this programme produced is an argument about how evidence moves rather than a defect anyone can correct. That belongs in a paper, under peer review, not in a register.

Levels for the two NICE items are our reading of the notice as sent and have not been separately reviewed.

What this register does not claim

Changes to this page
30 September 2026 — first published. The EJE entry was re-verified against the live article that morning; every other dated verification is stated on its entry and has not been repeated.
10 October 2026 — ET-064 added, a new section for the Department for Education. ET-060’s date corrected from 4 to 3 September, and the error recorded as ET-065. Own-error counts updated to match.
10 October 2026 — ET-066 added: an error of ours in The Threshold (2/4), corrected the same day. Own-error counts updated to match.
11 October 2026 — ET-067 added: errors of ours across our Conversion Practices Bill pages and The Threshold, corrected the same day. Own-error counts updated to match.
11 October 2026 — ET-068 added: errors of ours in Stopping Is Not Neutral, corrected the same day. Own-error counts updated to match.
11 October 2026 — ET-069 to ET-072 added, with new sections for a scoping review in the International Journal of Transgender Health and a position statement of the National Commission on Correctional Health Care, both reported that day.
13 October 2026 — ET-073 to ET-077 added, a new section for the Good Law Project and QueerAF articles examined in Evidence Note EN-025.
13 October 2026 — a routing reply from the National Commission on Correctional Health Care noted on ET-071 and ET-072; status unchanged.