A note from Eden
I didn’t set out to appraise the Cass Review. But over the past few months it has been the piece of work I’ve been asked for most often.
A full re-review of all the evidence behind the Review is more than one researcher can do well, so I chose something narrower: to appraise the appraisal. That means asking whether the reviews the Review commissioned were soundly done, and whether the conclusions drawn from them follow.
I’m not neutral. I support responsible access to puberty blockers and hormone treatment for adolescents, and I’m saying so first because it matters for how you read what follows. I can’t promise that holding that view won’t affect my judgement; no one in my position honestly could. What I can do is fix the rules before I read, publish every judgement with its reasoning, and make it possible for you to check my work and see where you’d have called it differently.
The results will be what the method finds. Some may support my view and some may cut against it; both will be reported at full weight. My position and interests are set out in full below.
Eden
GUIDANCE APPRAISAL · GA-002 · IN PREPARATION
Eden Openly is carrying out a structured meta-appraisal of the Cass Review’s final report, published on 10 April 2024, and of the research it commissioned from the University of York. No findings have been published. None will be until the whole appraisal is complete, and then they will be released together.
What we are doing
The Review commissioned seven systematic reviews and an international survey of gender services. A meta-appraisal appraises an appraisal. The question is not what the right treatment is for gender-questioning young people. It is narrower. Were the commissioned reviews soundly conducted? Did the Review’s conclusions follow from what they found? How did the state act on them?
We keep three layers apart:
- Evidence: what the York reviews found, and how well they were done.
- Interpretation: what the Review concluded from that evidence.
- Policy: what government and the NHS did next.
What it does not do
- It does not re-review the clinical evidence, re-run searches, pool studies or produce new estimates of benefit or harm. It is not a systematic review.
- It does not judge the Review by evidence published afterwards. The test is whether each conclusion was a defensible inference on the information available at the time.
- It makes no inference about the motives of any person or organisation. Methods and reasoning are assessed; people are not.
Where things stand
The appraisal runs in four stages, in this order:
- Methods registrationThe full protocol, frozen publicly before any scoring.We are here
- Claim registrationThe list of the Review’s claims to be appraised, frozen before any is scored.
- AppraisalScoring and judgement, in a random order, under the registered rules.
- Single releaseEvery result, record and correction, published together.
The methods are being finalised. Before any scoring begins, the full protocol will be frozen as a public registration on the Open Science Framework, with a published checksum, and mirrored on Zenodo with a DOI. A second registration then fixes the list of claims to be appraised before any of them is scored.
Until the methods are frozen, reading of the report and the York reviews is restricted by a written rule to what the method design needs. Every read, including a small feasibility pilot, is logged for publication. Registration does not mean we come to the Review unseen. It has been public and widely debated since 2024, and prior knowledge is recorded in an appendix of the registration.
The protocol
The protocol is not yet public. It is a working draft, and drafts are never posted, so that the only public version is the one frozen before scoring. When it is registered, its link, DOI and checksum will appear here.
If this page and the registered protocol ever differ, the protocol governs.
Our position, and why we don’t claim neutrality
We don’t claim to be neutral. Eden is a transgender woman, with a personal stake in this subject.
Eden is a dermatologist. She holds a declared prior in favour of adolescent access to puberty blockers and gender-affirming hormones; the method is designed to make any resulting bias more visible and auditable, and findings that cut against that position are reported at full weight.
Nothing in this appraisal rests on Eden’s clinical standing. Every judgement is published with its reasoning, for anyone to check.
She has further interests. All of them are listed, with the date each began, in our register of declared interests, and the protocol declares them in full. Those that bear on this appraisal are:
- She founded and funds a non-profit transgender healthcare and education organisation and a transgender health research institute.
- Through Eden Openly, she has taken public positions on the pause of gender surgery in Wales and on the Health Bill’s single patient record provisions.
- Her own published evidence reviews have graded some of the same evidence: Puberty Suppression in Adolescents (ER-014, with its Evidence Note EN-014) and Masculinising and Feminising Hormones in Adolescents (ER-015, with EN-015), published in July 2026. The complete list of her earlier publications that grade any of the same studies is fixed and sealed before registration, and until the appraisal’s judgements are committed, Eden Openly publishes no new grades on that evidence.
She also co-owns a commercial skincare business and is planning a general aesthetic hospital. That interest is in the register, and does not bear on this appraisal.
The appraisal is unfunded. Eden accepts no third-party funding for research, publication or appraisal, and pays every cost, including external review, herself.
Most of these interests create a plausible pull toward findings critical of the Review or of the policy that followed it. Declaring them doesn’t neutralise them. The method has to do that.
How the method guards against it
- Rules before reading, claims before scoring. Two public registrations fix the method, then the claim list, before any judgement is made.
- Established instruments, applied in the open. ROBIS for the treatment and exposure reviews; the JBI checklist for the descriptive, care-pathway and guideline reviews; AXIS for the survey; and a registered audit of how York applied AGREE II. Every item-level judgement is published with its reasoning. No invented overall scores are produced.
- The same rules in both directions. Benefits and harms are judged by identical methodological rules. A gap in reporting is never scored as a defect. Before any claim is judged, the arguments for and against it are listed in full, and each judgement is checked for whether it would stand if its direction were reversed.
- Graded, not binary. Each of the Review’s conclusions is placed in one category: concordant, defensible, overextended, unsupported or indeterminate. “Defensible” exists for conclusions within the evidence where a materially different conclusion is equally defensible.
- Blind re-scoring. Every ROBIS judgement, and a fixed sample of 100 conclusion-level judgements, are re-scored with the original masked, at least four weeks later. Every discrepancy is published.
- Order and samples drawn at random. Claims are scored in a random order, and the re-scoring samples are drawn at random. The randomness comes from a public source, and only after the relevant work has been committed, so no one can know or choose it in advance. See Fixing things in advance, below.
- AI assistance, disclosed. In the Cass Review appraisal, an AI model makes its own calls on some judgements, as a check on Eden’s. These are the judgements that need subject expertise, and any judgement that turns on a study Eden has graded before. Eden records her call first; the model then makes its call without seeing hers, and the two are compared only once both are fixed. Eden’s call is the one that counts. Where she makes no call on such an item, the model’s call is recorded and marked as the model’s. Every disagreement is published, and agreement is never counted as support. See How We Use AI.
- External review. The protocol underwent four rounds of independent, adversarial review in the endocrinology domain by a confidential reviewer; the domain review was commissioned (paid). This is domain review, not a claim of systematic-review methodology credentialing.
- A single author, stated plainly. These safeguards mitigate the risk of a one-person appraisal. They do not remove it.
- Errors in the open. Errors in our own records are logged by the direction they ran and published with the appraisal. Corrections are never made silently.
- A publication hold. From registration until every judgement is committed, Eden Openly will publish no new grades on evidence the appraisal covers.
Fixing things in advance: commitments and public randomness
Several of the safeguards above depend on one thing: being able to show that something was fixed before an outcome could have influenced it. These include the claim list, the scoring order, the re-scoring samples, and each judgement before it is compared. We do this in two ways, and both can be checked by anyone without trusting us.
Commit now, reveal later
When a record has to be fixed before an outcome it constrains, for example a list of judgements before they are compared, we publish its fingerprint first: a SHA-256 hash, a short string that changes completely if a single character of the record changes. The record itself is published later, with everything else. Anyone can then hash it and confirm that it is the record we committed to. A commitment is not a publication. It shows what we had fixed and when, without revealing it early.
- A chained, append-only manifest. Each new entry records the hash of the committed record and the hash of the manifest before it. Deleting, inserting or rewriting any earlier entry would break the chain.
- Timestamps anchored in Bitcoin. Each state of the manifest is timestamped with OpenTimestamps, which records its hash in the Bitcoin blockchain. The proof shows that the manifest existed in that form no later than a given block. It can be checked with free software and any Bitcoin node, without relying on us or on any website.
- Where to find it. The manifest lives in a public repository, github.com/EdenOpenly/cass-commitments, with a fallback page on this site if the repository cannot be reached. Snapshots are archived on Zenodo, each with its own DOI, and in Software Heritage. These are places to find the record. The evidence is the timestamp proof itself.
- Nothing is committed yet. The repository is set up. The first commitments are made when the methods are registered.
Public randomness
Where the method needs a random choice (the order in which claims are scored, or which judgements are re-scored), the random number must not be ours to pick. It is taken from a public randomness beacon: a service that publishes a new, unpredictable, cryptographically verifiable value at fixed intervals.
- The source. We use drand, run by the League of Entropy, a group of independent organisations. Each drand value is produced jointly by many of them, so no single operator can predict or bias it, and anyone can verify it against drand’s public key.
- The order of events. For each draw, the work it applies to is first completed or frozen. A declaration of the draw is then committed and timestamped. The value used is the first beacon value published at least ten minutes after that timestamp. It did not exist when we committed, so we could not have chosen it.
- Fallbacks, fixed in advance. The registered protocol says what happens if drand cannot be reached within a set time. It names the NIST Randomness Beacon, if that is verified as operational when the methods are frozen, and then a Bitcoin block hash as a last resort, under a mechanical rule. The protocol discloses the Bitcoin source’s known limit: a miner who finds a qualifying block could, at a cost, choose not to publish it.
- Reproducible. The script that turns a beacon value into an order or a sample is published with the registration, with its tests and test values. Anyone can rerun it and get the same result.
These steps prove timing and chance. They do not prove that a judgement is right. That is what the published reasoning, the blind re-scoring and the open error log are for.
The wider debate
Published criticism and defence of the Review are gathered by a registered search that runs identically for both sides. Each source’s declared interests are published as that source declared them, as context for readers, never as a substitute for engaging its argument.
Corrections and questions
If you’ve seen a claim about this project that doesn’t match this page, or you think we’ve got something wrong, please get in touch through our editorial contact page. Corrections are logged on our corrections page.
Eden Openly is an independent clinical health-education publication. It receives no funding from, and has no commercial relationship with, any provider of gender-related healthcare, any pharmaceutical company, or any supplier of NHS digital systems. This page describes a protocol in preparation; once the protocol is registered, it governs. Our full position on funding and interests is set out in our statement of independence, and every interest is listed, with its date, in our register of declared interests.