Curation log — melanoma¶

Newest entries first. Every content-changing session appends an entry: date, what changed, what was searched, follow-ups for next time. See CLAUDE.md.

2026-09-01 — Independent full audit (auditor: codex; author: claude)¶

Independently audited every one of the 26 wiki pages and all 14 literature artefacts written by claude. codex re-fetched the final identifier set from the official live APIs, checked the records against the citations and the cited claims, re-ran searches behind categorical absence statements, corrected the corpus, and then repeated the identifier and structural checks on the edited end state. All 26 pages passed and were promoted from draft to curated.

Audit volume and final verification¶

Check	Result
Wiki pages audited	26 / 26
Literature artefacts audited	14 / 14
Unique PubMed records in final audited scope	636 / 636 resolved live
PubMed record types	635 `PubmedArticle` + 1 `PubmedBookArticle` (GeneReviews)
Unique ClinicalTrials.gov records in final audited scope	63 / 63 resolved live
Claim–PMID pairs checked	2,212
Claim–NCT pairs checked	134
Total claim–identifier pairs checked	2,346
Bibliography completeness	636 / 636 scope PMIDs represented exactly once
Wiki body-to-reference completeness	26 / 26 pages; 0 missing and 0 unused page references
Non-journal web endpoints	24 / 24 returned HTTP 200 in the final live pass
Page density	26 / 26 pages at 150–179 lines and 25–55 references
Local Markdown links	0 broken
Bare verification markers remaining	0

Pair counts treat each PMID or NCT occurrence attached to narrative or tabular evidence as one claim–identifier pair and exclude the wiki ## References lists and the master bibliography. The final clean PubMed E-utilities EFetch pass returned exactly the 636 identifiers in the edited scope, with no missing or extra record. The final ClinicalTrials.gov v2 pass returned the identification module for exactly all 63 NCT identifiers. Bibliography titles, first authors and publication years were compared with the live PubMed metadata; ePub/print year offsets and PubMed typography such as BRAFV600 versus BRAF(V600) were treated as equivalent.

Errors found and fixed¶

#	Error in the authored material	Correction made
1	The ctDNA section categorically stated that no melanoma trial had randomised management on ctDNA.	Added the live 2026 CAcTUS report (21 patients; PMID 42168180; NCT03808441), which randomised a ctDNA-triggered treatment-switch strategy. Reframed the gap precisely: feasibility is shown, but no progression-free- or overall-survival benefit from acting on ctDNA is established. Added NCT06470880 (not yet recruiting) and NCT04901988 (terminated before randomisation).
2	The surgical and trials pages stated that no randomised trial addressed thinner-primary margin de-escalation.	Added recruiting phase 3 NORMA 2 (NCT07530887; planned n = 1,749), testing no re-excision versus 1–2 cm re-excision after an initially clear ≥1 mm diagnostic excision for pT1b–pT4b melanoma. Preserved the narrower, dated gap that no melanoma-in-situ-specific margin RCT was located.
3	The staging/statistics layer stated that no PET meta-analysis existed.	Added the 24-study meta-analysis (PMID 19727717): scan-level specificity 0.86, positive likelihood ratio 5.86 and diagnostic odds ratio 37.89. Clarified its obsolete search windows and retained the dated gap for a contemporary PET-CT meta-analysis.
4	The patient/advocacy layer stated that prevention campaigns lacked evaluated outcomes.	Added a hospital campaign series (PMID 2390498) and the NSW mixed process/outcome evaluation (PMID 39663828). Distinguished awareness, shade and behavioural outcomes from the still-unmeasured melanoma-incidence and mortality effects.
5	The molecular page overstated the absence of prospective evidence for the 31-gene expression profile.	Added prospective multicentre decision-impact studies (PMIDs 36617959 and 39754143), which reduced sentinel-node biopsy use. Reframed the unresolved question as absence of demonstrated recurrence, survival or adjuvant-selection benefit.
6	The guideline registry omitted a dedicated practical melanoma-in-pregnancy guideline.	Added Crisan 2016 (PMID 27240064) and corrected the pregnancy-gap wording.
7	The guideline registry omitted recommendations covering checkpoint therapy in chronic infection, immunosuppression and solid-organ transplantation.	Added the Spanish Melanoma Group recommendations (PMID 33782108) and corrected the special-population gap wording.
8	The transplant section stated that no prospective strategy study existed.	Added recruiting phase 1/2 NCT05896839 (planned n = 16 kidney-transplant recipients; nivolumab/ipilimumab with sirolimus/prednisone). The page now says that a prospective strategy is recruiting and has no results.
9	The active phase 3 landscape was stale and materially incomplete.	Re-ran the live phase 3 registry search and added NCT04674683, NCT05155254, NCT05625399, NCT06246916, NCT05751928, NCT06488482, NCT05078047, NCT07530887, NCT07552597, NCT05502900 and NCT06581406. Reframed the page as a selected efficacy/practice-changing landscape that explicitly excludes legacy completed-primary, biosimilar/PK and basket studies.
10	The uveal-melanoma phase 3 count was four and omitted two active efficacy studies.	Corrected the count to six active phase 3 or phase 2/3 efficacy trials, plus expanded access, and added NCT05502900 and NCT06581406 to the page and tebentafusp landmark note.
11	The text implied that nivolumab–relatlimab had never been compared with nivolumab–ipilimumab in any trial structure.	Added NCT03724968, which contained both arms but was non-randomised, terminated and enrolled two patients; distinguished it from a valid head-to-head randomised comparison. Added the Harmony anti-LAG-3-combination comparison where relevant.
12	The AI diagnosis gap was expressed as an unqualified absence of patient-level prospective testing.	Added recruiting cluster-randomised NCT06932172 (planned n = 3,000) and stated the exact dated gap: its endpoint is diagnostic yield and no reported patient-morbidity or mortality result was located.
13	The safety page said pigment-dependent tools “fail” in amelanotic melanoma.	Replaced the adjective with the cited pooled estimates: dermoscopy sensitivity/specificity 61%/90% and reflectance confocal microscopy 67%/89% (PMID 31747045).
14	The prevention page said behavioural counselling had not reduced sunburn.	Corrected this to “has not consistently reduced sunburn; melanoma outcomes are essentially unmeasured,” matching the evidence review (PMID 29558557).
15	The GeneReviews citation was malformed as a journal-style `1993` record.	Corrected it to Kraemer et al., Xeroderma Pigmentosum, GeneReviews, 2003 [updated 2022], PMID 20301571, and retained its `PubmedBookArticle` handling in validation.
16	The patient layer linked to the superseded OcuMel UK domain.	Verified the organisation's rebrand and replaced the URL with the current Ocular Melanoma UK endpoint, `https://omuk.org/`, in both affected artefacts.
17	The targeted-therapy page silently used the publication enrolment figure for NCT01307397 although the current registry total differs.	Recorded both figures explicitly: 3,226 enrolled / 3,222 treated in the publication versus 3,219 in the current registry record.
18	A ctDNA rewrite left an unused colorectal ctDNA reference in the clinical-trials page reference list.	Removed the unused page reference and renumbered the list; retained the paper only where it is actually cited in the surveillance page.
19	Four live-verified registry papers were missing from the master bibliography.	Added PMIDs 18983550, 22997461, 26314774 and 31220881 with registry tags and citing paths; the bibliography now covers all 636 scope PMIDs exactly once.
20	One PubMed title contained a Unicode paragraph separator that split PMID 29149136 across logical lines.	Normalised the character in the targeted-therapy page and bibliography, restoring one-record-per-line parsing.
21	The neoadjuvant page fell below the requested reference-density floor at 23 unique papers.	Live-searched PubMed, added the pre-S1801/NADINA Cochrane synthesis with its very-low-certainty effect estimates (PMID 36648215) and a 2026 efficacy/toxicity meta-analysis (PMID 41478279), and brought the page to 25 unique references without padding.
22	The neoadjuvant page repeated its practical-implementation synthesis.	Removed the duplicated paragraph while retaining the cited real-world cohort and Delphi evidence in the dedicated implementation section.

Asserted-absence searches re-run¶

Live PubMed E-utilities and ClinicalTrials.gov API searches were re-run for population melanoma-screening randomisation and mortality, melanoma-specific surgery versus radiosurgery for brain metastases, leptomeningeal randomised comparisons, nivolumab–relatlimab versus nivolumab–ipilimumab, thin-primary and melanoma-in-situ margin trials, ctDNA-guided management, AI diagnostic patient outcomes, 31-GEP treatment selection, prospective checkpoint-toxicity selection, nicotinamide with melanoma endpoints, germline-surveillance mortality, transplant strategies, occupational-UV subtype attribution, risk-model-directed surveillance, advocacy-campaign outcomes, and neoadjuvant-versus-adjuvant trials with overall survival as the primary endpoint. A broad active phase 3 melanoma registry query was also reconciled against the trials page.

Remaining flags and dated evidence gaps¶

No unresolved substantive flag remains. Genuine gaps are now written as positive, dated statements rather than bare absence claims: no randomised population-screening mortality result; no melanoma-specific randomised surgery-versus-radiosurgery comparison; no randomised leptomeningeal comparison; no melanoma-in-situ-specific margin RCT located; no demonstrated outcome benefit from ctDNA-guided management; no prospectively validated checkpoint-toxicity selection rule; no melanoma-endpoint nicotinamide trial; no randomised germline-surveillance mortality evidence; no reported AI-associated morbidity or mortality result; and no modern neoadjuvant-versus-adjuvant trial with overall survival as the primary endpoint. These are time-sensitive evidence gaps and must be re-searched at the next curation sweep.

2026-09-01 — Full build (author: claude; auditor: pending, must be a different engine)¶

Built all 26 canonical wiki pages from live literature, plus the complete literature layer, a rewritten OPEN-QUESTIONS.md, and an updated INDEX.md. Status raised from seeded to built in CONDITIONS-ROADMAP.md. No page was promoted beyond status: draft — promotion is the auditor's decision and the auditor must be a different engine.

What was produced¶

Artefact	Result
Wiki pages	26 of 26, 150–179 lines each, 26–55 citations each, 3,984 lines total
Unique PubMed records cited	622, every one retrieved live in this session and re-resolved in a single end-of-session verification pass (622/622 resolved, 0 missing)
NCT identifiers	46, every one resolved live against the ClinicalTrials.gov v2 API on 2026-09-01 (46/46 resolved)
Landmark notes	7 (`literature/notes/`)
Guidelines registry	30 document rows with supersession chains, 6 web resources verified by HTTP fetch, a disagreements section, a stated-gaps section and a watch list
Statistics sheet	117 records across 14 tables, plus a 14-item "known conflicts and caveats" section
Patient-voice layer	4 files; 18 organisations verified live by HTTPS fetch on 2026-09-01; 8 themes each supported by ≥2 independent sources; no quotation used at all (stricter than the 15-word convention)
Open questions	21 questions across three tiers, a 12-row "dots not yet connected" table, and 9 seed items explicitly closed rather than deleted
Internal links	All cross-page links checked programmatically; 0 broken

Searches run¶

257 distinct live PubMed E-utilities queries were logged during the build (esearch + esummary, with efetch for abstracts). Scoping counts re-verified on 2026-09-01: melanoma 184,619; uveal melanoma 8,280; mucosal melanoma 3,476; acral melanoma 2,099 — unchanged from the seeding session. ClinicalTrials.gov v2 was queried by condition, by phase-and-status filter, and by individual NCT identifier.

Query clusters, one per planned page, covered: global burden and GLOBOCAN/GBD vintages; incidence and mortality trends by country and birth cohort; overdiagnosis, diagnostic scrutiny and biopsy rates; population screening programs; UV exposure, phenotype, nevus count, sunbeds, sunscreen and chemoprevention; germline predisposition genes and their surveillance evidence; TCGA and precursor genomics, mutational signatures, resistance programmes; dermoscopy, confocal microscopy, AI reader studies, teledermatology and biopsy technique; Breslow/ulceration/mitotic rate/TIL/regression prognostics and immunohistochemistry; AJCC 8 validation cohorts and non-cutaneous staging systems; MSLT-I/II, DeCOG-SLT, nodal tumour burden and sentinel-node technique; margin trials, Mohs and lentigo maligna, in-transit disease, metastasectomy; adjuvant and neoadjuvant trial programmes and their critiques; CheckMate 067/KEYNOTE-006/RELATIVITY-047/DREAMseq; BRAF/MEK trials, resistance and pyrexia management; TIL therapy, oncolytic agents and neoantigen vaccines; brain and leptomeningeal metastases; immune-related adverse events by organ, fatality, chronicity and corticosteroid effect; uveal genomics, COMS, tebentafusp and liver-directed therapy; acral/mucosal genomics, efficacy gap and disparities; transplant, pregnancy, paediatric, older-adult, HIV, autoimmune and unknown-primary populations; follow-up schedules, ultrasound adherence, ctDNA and second primaries; guideline documents from ten bodies; and the patient-experience and advocacy literature.

Scoping decisions held and made¶

Border, not breadth — held from the seed. The condition owns melanoma of all primary sites and the systemic-therapy story melanoma established; it does not own keratinocyte carcinomas or immune toxicity outside melanoma cohorts. Where a study belongs to both this condition and another, it is cited here for its melanoma endpoint and the other condition is linked.
Uveal melanoma kept inside melanoma as a dedicated page. The build confirms the reason: the contrast is the content. Uveal melanoma's 5-year relative survival was flat at 82.8% from 1975 to 2016 while cutaneous metastatic 5-year survival rose from 16% to 35% (PMID 40225965; PMID 41528114), and its first-line agent works by a different mechanism entirely (PMID 34551229). Filing them apart destroys that comparison.
The retinoblastoma border was resolved as instructed. The melanoma-specific figures the seed lacked were located: 50-year cumulative melanoma incidence 4.5% in heritable versus 0.7% in non-heritable retinoblastoma survivors, with onset ~20 years earlier (Kleinerman 2021, PMID 34153328). The Schonfeld 3.1–17 figure is reported honestly as a range across melanoma, CNS, oral-cavity and breast subsequent neoplasms, never as a melanoma-specific SIR (PMID 33473166). "Retinoblastoma survivors" is never written as a single risk group anywhere in this condition. Retinoblastoma is cross-linked, not restated.
The Olivier critique of SWOG S1801 is presented as a live disagreement, as the seed required, with its three specific objections stated and NADINA's phase 3 design noted as answering only the third (PMID 38621314; PMID 38828984).
clinical-trials-landscape.md is dated and verified rather than illustrative. Every NCT record was fetched from the v2 API on the build date and its status, enrolment and primary-completion date recorded as of that date.
Patient-voice ethics applied more strictly than the convention requires: public sources only, no private individual named or profiled, and no quotation at all rather than the permitted ≤15 words.

Notable content decisions¶

The overdiagnosis controversy is treated as content, not as a problem to resolve. Both readings are presented with their own evidence, and the junction that neither literature has crossed is recorded as OQ-1 and D1.
The stage IIC/IIIA survival inversion is stated plainly across three independent validation cohorts, because it inverts the intuition that a higher stage number means worse disease and is the direct argument for stage IIB/IIC adjuvant therapy.
KEYNOTE-716's PRFS2 hazard ratio of 0.75 (95% CI 0.56–1.01) is given equal prominence to its RFS result. It is the number that determines whether adjuvant therapy in stage IIB/IIC produces durable benefit or mostly advances the timing of treatment, and it is rarely quoted.
The European guideline's own disclaimer is quoted: sentinel node biopsy "shall be offered… although there is as yet no clear survival benefit for this approach."

What could not be verified, and what is recorded as absence¶

No trial was found comparing modern neoadjuvant against modern adjuvant therapy with overall survival as the primary endpoint. The active phase 3 landscape was queried by condition and phase/status filter on 2026-09-01; the absence is dated and recorded as OQ-3 and D3.
No melanoma trial randomising management on ctDNA was located (OQ-15). The comparable design exists in colon cancer (PMID 40055522).
No randomised trial of population melanoma screening exists, and none appears in the active phase 3 registry (OQ-19).
No evaluation of a melanoma advocacy or awareness campaign against an outcome was located; recorded in literature/patient-voice/organizations.md and sources.md.
No current guideline document was located for melanoma in pregnancy, in transplant recipients, in children as a dedicated document, or for acral melanoma specifically; the Australian ocular/periocular document dates from 2008. Recorded in the registry's gaps section.
No acral- or mucosal-specific patient organisation was located, and no organisation from Africa, South America, South Asia, East Asia or the Middle East returned a verified page — recorded as a directory gap, not as evidence of absence of need.
Two ocular-melanoma organisation URLs did not return a verified page on 2026-09-01 and were therefore omitted rather than listed unverified.
NCCN guideline content sits behind registration. The two NCCN resources are cited by URL and access date only; the version number behind the login was not read, and this is stated in the registry.
PMID 38752346 (Olsen, Denmark) has no abstract in PubMed. It is cited at title level only, for the existence and direction of ecological evidence, never for a figure.
PMID 36946230 (Ni, Australian second-primary cohort) has no abstract in PubMed. Cited at title level only.
Two collective-author records (PMID 26091043 TCGA; PMID 25555246 AJCC Ophthalmic Oncology Task Force; PMID 32135640 Chinese Anti-Cancer Association) return "Anonymous" from the esummary author field; their collective names were retrieved from the efetch XML and corrected by hand.

Follow-ups for the auditor¶

Re-fetch every one of the 622 PMIDs and 46 NCT IDs; check author, year, journal and every quoted number against the record. Tolerate ePub-vs-print year offsets.
Re-run the searches behind every asserted absence listed above — stale absences are the most common real error, and four of them are load-bearing for open questions.
Check the retinoblastoma border rules specifically: that no page writes "retinoblastoma survivors" as one group, and that 3.1–17 is never presented as a melanoma-specific point estimate.
Check the inline author attributions. Several were corrected during the build after the citation resolved to a different first author than assumed; a systematic check against the References list of each page is warranted.
Verify the two title-only citations (PMID 38752346; PMID 36946230) are used only for existence and direction, not for figures.
Re-verify the 18 organisation URLs and 6 guideline web resources; web resources rot faster than PubMed records.
Check that no page exceeds its scope border into keratinocyte carcinoma or non-melanoma immune toxicity.
Promote to curated only pages where every citation resolves and matches with no unresolved substantive issue; otherwise leave draft and record why.

Next sweep¶

Re-check: whether any neoadjuvant-versus-adjuvant overall-survival trial has been registered; whether any melanoma ctDNA management trial has opened; whether MelMarT-II (NCT03860883) has reported; whether any guideline has addressed the KEYNOTE-716 PRFS2 result; and whether the EADO/EDF/EORTC guideline, which states validity to the end of 2026, has been updated.

2026-09-01 — Seeded (Claude)¶

Created the condition scaffold: INDEX.md with a 26-page canonical plan (24 topical + 2 standing house pages) and an anchor table, a seed OPEN-QUESTIONS.md (7 questions, 6 dots), this log, and the empty wiki/ and literature/ layers. No wiki pages written. The condition is deliberately excluded from the public site until it has pages.

Searches run (live PubMed E-utilities, 2026-09-01). Scoping counts: melanoma 184,619; melanoma[MeSH Major Topic] 98,409; cutaneous melanoma 24,442; uveal melanoma 8,280; mucosal melanoma 3,476; acral melanoma 2,099. Topic counts recorded per planned page in INDEX.md, including incidence/epidemiology 30,191, special populations 13,154, adjuvant therapy 8,972, screening/early detection 7,553, brain metastases 6,843, BRAF/NRAS/NF1 6,765, checkpoint agents 6,525, sentinel node 4,740, dermoscopy 3,504, targeted agents 3,402, Breslow/ulceration prognosis 3,034, TIL therapy 2,435, irAEs 2,223, UV risk factors 1,502, margins 1,286, neoadjuvant 1,231, ctDNA 581, survivorship 306.

Anchor records resolved live: 29, listed in INDEX.md with year, first author and journal. Retrieved and read abstracts rather than citing from memory for PMID 35353115 (Arnold, global burden) and 39282897 (Wolchok, CheckMate 067 10-year); every figure quoted in the INDEX.md overview comes from a retrieved abstract.

Scoping decisions made. 1. Scoped by border, not by breadth — the same treatment given to hypertension, and for the same reason. At 184,619 records this is the second-largest literature in the repository and the failure mode is sprawl. The condition owns melanoma of all primary sites and the systemic-therapy story that melanoma established; it does not own keratinocyte carcinomas or immune toxicity outside melanoma cohorts. 2. Uveal melanoma kept inside melanoma as a dedicated page, not split into its own condition. It has 8,280 records and is biologically distinct enough to justify separation on the lung-adenocarcinoma/squamous precedent. It is kept together because the contrast — checkpoint blockade transforming cutaneous and not uveal melanoma — is one of the most informative things in the field and is destroyed by filing the two apart. Recorded in CONDITIONS-ROADMAP.md as a candidate future standalone condition; if promoted, this page becomes the overlap page. 3. Acral and mucosal melanoma share one page (2,099 and 3,476 records), scoped around their distinct genomics and their under-representation in the trials that set the standard of care. 4. Screening and overdiagnosis is a page, not a paragraph, because the incidence–mortality divergence is a live unresolved controversy and this repository treats those as content. 5. A quantified shared border with retinoblastoma, seeded in the same session (PMID 33473166). Both conditions carry the border in INDEX.md with a rule to cross-link rather than restate.

Flagged for the build pass. - Schonfeld's abstract gives 3.1–17 as a range across melanoma, CNS, oral cavity and breast — not a melanoma-specific SIR. Retrieve the melanoma-specific figure or report the range as a range. Do not invent a point estimate. - Olivier (PMID 38621314) raises time-related biases in the SWOG S1801 analysis. Present the disagreement rather than resolving it by preference. - No anchor was resolved for: stage IIB/IIC adjuvant trials (KEYNOTE-716 and counterparts), DeCOG-SLT, KIT-directed therapy in acral/mucosal disease, melanoma in pregnancy, transplant/immunosuppressed cohorts, ctDNA surveillance, patient-organisation material, or current NCCN/ESMO documents. These are live-search tasks, not inheritances. - Every anchor in the table was resolved in a scoping session, not a build session, and must be re-verified before use.

Follow-ups. Build with tools/build-condition.sh melanoma; auditor must be a different engine from the writer.