Curation log — MASLD¶

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

2026-09-02 — independent audit of Claude depth pass (Codex)¶

Authorship and audit independence. Claude authored the additions in the depth pass immediately below. Codex independently audited them against live PubMed E-utilities and ClinicalTrials.gov v2 records, then made only corrections needed for claim fidelity, provenance, reference closure and status.

Result¶

Re-fetched all 93 PubMed records attributed to the depth pass: 90 page-level additions, two additions to the patient-evidence layer (PMIDs: 37024220 and 39701123), and the literature-only 2026 EASL guidance record (PMID 42617506). All 93 resolved in the same-session E-utilities responses; author, title, journal and year metadata matched after the corrections recorded below.
Re-fetched all 64 NCT IDs present at the start of this audit from the ClinicalTrials.gov v2 API. The care-pathway correction added NCT06671886, also fetched live, bringing the current wiki total to 65.
Re-ran the exact ClinicalTrials.gov condition union (MASLD OR NAFLD OR MASH OR NASH OR "steatotic liver disease"): 2,056 total studies, 181 recruiting interventional studies, 112 phase 3 interventional studies, and 17 recruiting phase 3 interventional studies. The current-count table and linked statistics artifact now use these union counts.
Checked page bodies against page reference lists and literature/BIBLIOGRAPHY.md. Three missing page-reference entries were restored (PMIDs: 39283612, 35677499 and 40467095); final closure is zero missing body-to-reference or reference-to-bibliography PMIDs.
Promoted 20 pages to status: curated. genetics.md and clinical-trials-landscape.md remain draft because PMID 40581300 and NCT04565717 disagree materially on rapirosiran enrolment.

Was the added “depth” evidence or padding?¶

It was predominantly genuine evidence expansion, but not entirely. The immediately preceding independent audit recorded 274 distinct wiki-cited records; Claude's pass added 93 and produced 367, so the depth-pass log's stated baseline of 259 was stale. This audit removed eight citations that served only as publication-existence or review padding (PMIDs: 40759197, 32890594, 35165400, 31337282, 39690310, 37584065, 35899384 and 39136211) and added the corrective care-pathway protocol (PMID 40325466). The final wiki corpus is 360 distinct records: 16.36 records/page, versus 16.68 immediately before this audit and 12.45 before Claude's depth pass. The audited depth pass therefore remains a net expansion of 86 substantive records over its true 274-record baseline.

Corrections made¶

Bounded PEth misclassification to the selected at-risk cohort rather than generalising 39% to all MASLD cohorts; labelled epidemiological and mechanistic explanations as hypotheses where the cited designs were not causal.
Corrected overstatements in the genetics, microbiome, glycaemic-control, GLP-1, surgery, cirrhosis, HCC, pregnancy and guideline sections. In particular, statistical interaction was not translated into a simple multiplicative biological relation, Mendelian randomisation was not treated as automatically free of confounding, and observational adjustment was not described as proof of independence.
Corrected the resmetirom qFibrosis category error: the 24.4-percentage-point result concerns categorical qFibrosis stage, not a continuous-score change. The page also now exposes an additional source-internal inconsistency in PMID 40577015: its abstract says lower discontinuation improves cost-effectiveness while reporting a no-discontinuation ICER of $318,740/QALY versus $140,134/QALY in the base case.
Corrected trial provenance: the dapagliflozin study began in 2018 under NASH-era terminology and was described as MASH in its 2025 report; SPECCIAL and the pregnancy cohort are explicitly identified as retrospective MASLD classifications.
Corrected the LITMUS summary, FGF21 recruiting arithmetic and current registry totals. A newly asserted absence of a randomised care pathway was false: live retrieval found the MCCP cluster-randomised protocol (PMID 40325466; NCT06671886), now cited with its process endpoint and not misrepresented as a completed outcome result.
Corrected bibliography tags for guidance/review records and recalculated coverage: 360 distinct wiki records plus 15 literature-only records equals 375 across wiki and supporting literature.

Exact absence-search audit¶

Every PubMed audit search used the old/new-name union U = (MASLD[Title/Abstract] OR NAFLD[Title/Abstract] OR MASH[Title/Abstract] OR NASH[Title/Abstract] OR "steatotic liver disease"[Title/Abstract]). Exact claim-directed searches included:

Question tested	Exact construction after `U AND`	Result
Published VOYAGE efficacy	`(VK2809[Title/Abstract] OR VK-2809[Title/Abstract] OR VOYAGE[Title/Abstract])`	8; no peer-reviewed VOYAGE phase 2b efficacy report
Dapagliflozin incorporated into guidance	`dapagliflozin AND (guideline OR guidance OR consensus OR recommendation)`	8; none cited the 2025 histology trial as guidance
Obstetric or liver guidance naming pregnancy risk	`(pregnancy OR obstetric OR antenatal) AND (guideline OR guidance OR consensus OR recommendation)`	70; no retrieved guideline identified MASLD as an obstetric risk factor
Randomised non-cirrhotic HCC surveillance	`("hepatocellular carcinoma" OR HCC) AND ("non-cirrhotic" OR precirrhotic) AND surveillance AND (randomized OR randomised OR trial)`	16; no eligible surveillance trial
Formally qualified biomarker	`("biomarker qualification" OR "qualified biomarker")`	2; no qualified MASLD biomarker identified
Randomised clinical care pathway	`("clinical care pathway" OR "care pathway") AND (randomized OR randomised OR trial) AND (outcome OR implementation)`	3; found PMID 40325466, so the asserted absence was retracted
Contemporary paired-biopsy sampling study	`("liver biopsy" OR histology) AND (paired OR sampling OR reproducibility) AND (modern OR contemporary OR prospective)`	466; no eligible modern paired-core remeasurement identified after screening

Two exact ClinicalTrials.gov combinations were also tested under the full condition union: resmetirom AND semaglutide and sequential non-invasive AND clinical outcomes; both returned zero studies. Broad searches that could not support a defensible absence were not used as proof: claims were either bounded to “not retrieved,” converted to open questions, or removed.

Discrepancies deliberately preserved¶

PMID 25935633 reports 193/619 as 33.2%; the page retains the published percentage and the calculated 31.2%.
PMID 37040843 reports 63 studies while its geographic components sum to 64; no invalid fraction was reconstructed.
PMID 40581300 reports rapirosiran Parts A/B n=58/n=46, whereas NCT04565717 reports terminated with actual enrolment 6; both values remain side by side and the two affected pages remain draft.

2026-09-02 — depth pass (Claude). Widening the evidence base, page by page.¶

Why this pass ran. The condition was structurally complete but sourced narrowly: 133 lines and 21 citations per page across 22 pages, and — the measure that mattered — only 12 distinct PubMed records per page across the whole condition, against reference conditions running 15–24. The problem was not page length; it was that the pages recycled one shared pool of 259 records. Lengthening them with the same references would have made the ratio worse. The brief was therefore to bring in records this condition did not previously cite, sourced to each page's own topic.

What changed, in numbers¶

Measure	Before	After
Total wiki lines	2,941	3,581
Mean lines per page	133	163
Mean citations per page	21	25.5
Distinct verified records per page	12	17
Distinct records cited by pages	259	367
Records in `BIBLIOGRAPHY.md`	274	368 (93 added)
Open questions	30 (18 Tier 1 / 12 Tier 2)	43 (23 / 20)
Dots not yet connected	12	19
Guidelines catalogued	47	52
Documented statistical conflicts	12	21

All 22 pages were extended; none was rewritten and no verified content was deleted. Every page is status: draft and awaits re-audit by a different engine.

Search discipline¶

Every search unioned old and new vocabulary — (MASLD OR NAFLD OR MASH OR NASH OR "steatotic liver disease") or a topic-appropriate equivalent containing both naming systems. Roughly 45 distinct PubMed E-utilities searches were run (esearch + esummary + efetch), plus three ClinicalTrials.gov v2 API queries. Every one of the 93 new PMIDs was retrieved by live query in this session and its author list, journal, year, volume, pages and abstract read via efetch before being written. Two citations were caught mid-pass where a detail had been supplied from memory rather than from the record (an author list and a page number) and were corrected against efetch output before the file was finalised; the working rule for the remainder of the pass was that nothing goes into a reference line that did not come from the fetched record.

The three disclosed source-internal discrepancies were preserved¶

Verified present and unaltered at the end of the pass:

PMID 25935633 — 193/619 stated as 33.2% where the quotient is 31.2%; both shown deliberately.
PMID 37040843 — 63 studies whose geographic components sum to 64; the "62 of 63" calculation is still explicitly refused in epidemiology-and-burden.md and in STATISTICS.md conflict 7.
Rapirosiran — enrolment differs between PMID 40581300 and NCT04565717; both numbers retained side by side in genetics.md and clinical-trials-landscape.md.

What was deepened, page by page¶

overview.md (42→45 records) — added a fifth organising fact: the category is unstable in two directions (39.0% of self-report-classified MASLD has PEth in the MetALD/ALD range; clustering splits MASLD into liver-specific and cardiometabolic types; unstaged MASLD is null for mortality against a metabolically matched comparator).
nomenclature-and-definitions.md (25→28) — quantified alcohol misclassification with the Odense PEth cohort; added the data-driven two-cluster subtype analysis.
epidemiology-and-burden.md (27→30) — the GBD DALY paradox (3.67M DALYs against 75M for T2D, no significant change in 30 years) and its attribution explanation; the Indian meta-analysis (38.6% adults, with hospital-vs-community and obese-vs-non-obese splits).
pathogenesis.md (23→26) — the DAG–PKC-ε versus ceramide/DES1 question as a genuine unresolved mechanism, and the intestinal nicotine → AMPKα → SMPD3 → ceramide pathway with its nicotine-degrading commensal.
genetics.md (16→20) — the rare-variant map (CIDEB, MTARC1, GPAM), the multi-ancestry cirrhosis GWAS with its PNPLA3×exposure interaction, and clonal haematopoiesis as a somatic liver-risk factor.
noninvasive-assessment.md (21→25) — the age-specificity collapse of FIB-4 (35% at ≥65) with re-derived thresholds, the five-cohort population-screening failure (43% of LSM ≥8 kPa have normal FIB-4), and the superlearner ceiling on routine-variable models.
histology-and-biopsy.md (13→20) — a new section on sampling variability (41% ≥1-stage discordance between paired cores) and on trial-dataset reader reliability (κ 0.366–0.396 for the licensing endpoints; power >90%→40%). This corrected a stale absence: the page previously asserted no contemporary paired-sampling study existed; Ratziu 2005 and 2007 do, and the correction is stated explicitly on the page.
natural-history-and-fibrosis-progression.md (19→22) — the metabolically-matched mortality null, and the MASLD/MetALD/ALD outcome ordering from a 24-cohort meta-analysis and UK Biobank.
lifestyle-and-weight-loss.md (10→15) — MEDINA contradicting DIRECT-PLUS, exercise-modality dissociation, and the coffee dose–response (fibrosis-specific, not incidence).
resmetirom-and-thyromimetics.md (9→20) — quantified thyroid-status association, the phase 2 open-label extension, qFibrosis digital pathology, the two contradictory cost-effectiveness models, real-world access data, a second network meta-analysis with a different podium, and HSK31679's microbiome-dependent mechanism.
glp1-and-incretin-therapy.md (14→19) — four target-trial emulations on hard hepatic outcomes with a consistent within-class ordering, and lean-mass loss quantified at ~25% of weight lost.
bariatric-and-metabolic-surgery.md (8→12) — SPECCIAL (metabolic surgery in compensated MASH cirrhosis), which closed a stale absence on the page, and the alcohol-use-disorder hazard quantified across SOS and Danish registers with a possible incretin mitigation.
other-pharmacotherapy.md (19→23) — the dapagliflozin randomised histological trial, two empagliflozin liver-fat trials, and ATLAS (combination therapy, all arms negative on the primary endpoint).
cirrhosis-and-decompensation.md (19→22) — Baveno VII's MASLD-with-obesity failure (rule-in PPV 0.67) and its correction, plus spleen stiffness closing the grey zone.
masld-related-hepatocellular-carcinoma.md (15→20) — chemoprevention (statins, aspirin, metformin) and both sides of the immunotherapy controversy, including two prospective atezolizumab–bevacizumab cohorts that contradict the survival penalty.
cardiovascular-and-extrahepatic-outcomes.md (29→32) — the causality question stated properly (more atherosclerosis, not more asCVD mortality; only mixed hyperlipidaemia survives MR), cT1-versus-PDFF discrimination, and the HFpEF phenotype.
masld-and-type-2-diabetes.md (23→27) — glycaemic control slowing fibrosis progression but not liver events, lanifibranor's clamp study, and MASLD in type 1 diabetes with its twelve-fold modality spread.
lean-masld.md (9→16) — the cardiovascular contradiction laid out in full with seven studies on three sides, and the design split that explains it.
guidelines.md (33→39) — KASL 2025, AISF 2026 (STEPS-MASH), INASL 2023 (the resource-stratification objection) and the NICE summary, plus the guidance lag behind the dapagliflozin trial.
clinical-trials-landscape.md (25→28) — LITMUS biomarker head-to-head and its regulatory-qualification experience, ESSENCE baseline characteristics, and re-queried registry counts (17 recruiting phase 3 studies against 111 registered).
patient-experience-and-advocacy.md (17→20) — economic and PRO burden, the liver–brain axis as a mechanism for the cognitive complaints patients report, and the depression discordance between meta-analysis and Mendelian randomisation.
red-flags-and-safety-concerns.md (46→52) — herbal and dietary supplement injury (20% of US hepatotoxicity; turmeric with HLA-B*35:01) and a new pregnancy section (preterm-birth aOR 3.41, independent of obesity, with no severity gradient).

What was deliberately left alone¶

No page was rewritten or restructured. Section order, TL;DRs and existing tables were kept; new material was inserted as new sections or as clearly-marked extensions.
No verified content was removed, and no existing citation was dropped.
The three disclosed source discrepancies (above) were not touched.
literature/notes/ and literature/patient-voice/ were not modified. The seven landmark notes still cover the seven landmark papers; none of the 93 new records is landmark in the sense the notes require (practice-changing or field-opening), so adding notes would have diluted the criterion. The patient-voice layer's organisation and source verification is dated 2026-09-02 and was re-verified in the audit; nothing in this pass changed its ethics-governed content.
WHATS-NEW.md was not regenerated. It is a generated file (tools/build_recency.py), and regenerating it writes a cache into tools/ and a cross-condition file at the repository root — both outside this pass's permitted scope. It is therefore stale by 93 papers and should be regenerated in the next session that is allowed to run repository tooling.
pathogenesis.md remains the shortest page (126 lines). It was extended where a genuine open mechanism existed and not padded further; its density per line is already the highest in the condition.

One repair to record honestly¶

While editing cirrhosis-and-decompensation.md, a scripted insertion truncated the file, destroying its ## Related pages section and References 1–19. References 1–19 were restored verbatim from the file content read earlier in the same session, and re-verified against the in-page inline citations. The ## Related pages section was reconstructed, not restored: its ten links point to canonical filenames that all resolve, and each carries a one-line description written in this pass rather than the original wording. Nothing else on the page was affected, and no citation was lost. A re-auditor should treat that section as new text.

Absence claims re-run in this pass¶

Claim	Outcome
"No contemporary paired-sampling study of biopsy variability" (`histology-and-biopsy.md`)	False. Ratziu 2005 (PMID 15940625) and 2007 (PMID 17767466) exist; page corrected, and the residual open question narrowed to whether sampling error has been re-measured with modern cores
"Surgery untested in MASH cirrhosis" (`bariatric-and-metabolic-surgery.md`)	Superseded. SPECCIAL (PMID 39870816) reports 15-year hepatic outcomes; page updated, remaining gap is decompensated cirrhosis and randomisation
"No phase 3 result for any other THR-β agonist" (`resmetirom-and-thyromimetics.md`)	Still true. Re-run 2026-09-02 as `(VK2809 OR VK-2809 OR VOYAGE) AND liver` (74 records) and `VK2809 AND (NASH OR NAFLD OR MASLD OR MASH)` (8 records); no peer-reviewed VOYAGE efficacy report retrieved
"No guideline cites the dapagliflozin trial" (`guidelines.md`, OQ-33)	True as of 2026-09-02 across the 52 documents in the registry
"No obstetric or hepatology guideline names MASLD as an antenatal risk factor" (`red-flags-and-safety-concerns.md`, OQ-36)	True as of 2026-09-02

For the next session¶

Re-audit this pass with a different engine. 93 records were added across all 22 pages; none has been independently checked. The reconstructed ## Related pages section in cirrhosis-and-decompensation.md is new text and should be read as such.
Re-check the two citations corrected mid-pass (PMID 31337282 author list; PMID 41066138 article number) against the live records.
Consider whether any of the 93 new records deserves a literature/notes/ deep note. The strongest candidates are Davison 2020 (PMID 32610115) for what it implies about the entire MASH failure literature, Torp 2025 (PMID 40945520) for what it implies about every cohort in this condition, and Raverdy 2024 (PMID 39653777) if the two-cluster result replicates.
STATISTICS.md §13.12 and clinical-trials-landscape.md hold ClinicalTrials.gov counts from 2026-09-02 and are a single-date snapshot.
Regenerate WHATS-NEW.md with python3 tools/build_recency.py masld — it is stale by the 93 papers added here, several of which are 2026 publications that would move the recency buckets.
The pathogenesis.md mechanism sections remain the place where a further pass would have the most room, specifically on human tissue evidence for the DAG-versus-ceramide question (OQ-40).

2026-09-02 — Independent audit (Codex auditor; Claude author)¶

Scope and result. Independently audited all 22 wiki pages and all 14 literature artifacts authored by Claude. Re-extracted the edited corpus and verified 288 unique PubMed records through live PubMed E-utilities and 62 unique NCT records through live ClinicalTrials.gov v2 API records in this session; all identifiers resolved. Checked 1,267 claim-linked identifier occurrences (966 in wiki bodies and 301 in literature artifacts, excluding bibliography/registry catalogue repetitions) and 2,225 total PMID/NCT occurrences including reference lists and catalogues. Re-ran the stated PubMed absence searches, the ClinicalTrials.gov condition/status counts, and 16 public URLs. Wiki-to-bibliography closure is 274/274; no wiki-body PMID lacks a page reference.

Promotion result: 0 curated, 22 draft. No page was promoted. clinical-trials-landscape.md and genetics.md retain an unresolved publication–registry enrollment discrepancy for rapirosiran (PMID 40581300 versus NCT04565717). epidemiology-and-burden.md and the statistics artifact retain a source-internal study-count inconsistency in PMID 37040843. The Angulo landmark note retains a source-internal arithmetic inconsistency (PMID 25935633). Three patient-voice records have no PubMed abstract/full text available in this audit (PMIDs: 28624648, 33497764, 42525136), so their findings remain unsummarised. In addition, most pages remain below the requested reference density and/or 150-line floor. Those are substantive curation limitations, not identifier failures.

Errors found and fixed¶

Area	Error in authored material	Audit correction
Cirrhosis	Claimed four cited randomised cirrhosis trials although the three cited PMIDs support three cirrhosis RCTs	Corrected four to three
Recompensation	Claimed MASLD-specific recompensation had not been reported	Added the 344-person SLD cohort: 18.6% overall, 17.7% MASLD; retained observational/causal limitation (PMID 42467948)
MELD 3.0	Claimed no evaluation in MASLD/MASH	Added the 44,037-candidate registry study and its up/down-categorisation and outcome estimates (PMID 41284513)
Cirrhosis surgery	Claimed no randomised cirrhosis trial	Added recruiting NCT07058155 and completed observational NCT03753438; bounded the gap to no completed RCT
Surgery vs incretin	Claimed no head-to-head trial	Added recruiting NCT06374875; bounded the gap to no completed comparison
Bariatric procedures	Left procedure comparison as wholly unanswered	Added the 16-study sleeve-versus-bypass meta-analysis and its limitations (PMID 42420150)
Hard outcomes	Called metabolic surgery the only intervention with hard hepatic-outcome data	Reworded to the strongest dedicated observational major-adverse-liver-outcome analysis; stated non-randomised design
Pivotal endpoints	Claimed every pivotal trial uses only histological surrogates	Distinguished accelerated-approval histology readouts from phase 3/confirmatory trials with primary or later clinical composites
Combination therapy	Claimed only one registered combination trial	Added published semaglutide/cilofexor/firsocostat phase 2 (PMID 35439567), NCT07570810, and retained the narrower absence of resmetirom-plus-semaglutide
Rapirosiran	Presented publication enrollment and registry as if concordant	Explicitly retained the paper's Parts A/B n=58/n=46 against the registry's terminated/actual n=6 discrepancy; did not force reconciliation
Diabetes therapy	Claimed no randomised glucose-lowering comparison with hepatic endpoints	Added luseogliflozin plus semaglutide trial, including nonsignificant full-analysis comparisons (PMID 42605535)
Diabetes directionality	Claimed no genetic-instrument separation	Added three Mendelian/shared-genetics analyses and their non-identical directional results (PMIDs: 37223039, 37931882, 39690818)
Alcohol misclassification	Claimed objective misclassification was unquantified	Added PEth studies (PMIDs: 40517819, 40228582); retained lean-specific PEth-plus-genetics gap
Statin underuse	Claimed prescribing gaps had not been quantified	Added 6,055-person study: 33% untreated when indicated in MASLD versus 19% without MASLD (PMID 41861677)
Sequential NITs	Claimed no prospective false-negative measurement	Added prospective n=186 FIB-4/ELF/MRE/biopsy study: FIB-4 sensitivity 57%, 43% excluded from second-line testing, concurrent sensitivity 87%/specificity 64% (PMID 42566274)
Guideline implementation	Claimed implementation had not been measured	Added 207-officer pre/post training study; bounded remaining gap to durable practice and patient outcomes (PMID 41845319)
Patient voice	Claimed interview evidence was US/high-income only and omitted new studies	Added 25-adult lived-experience study and 16-person Nanjing study (PMIDs: 42651339, 42529177); corrected geographic coverage and study-type counts
Organisations	Retained transient 403/failure claims	Live retry verified American Liver Foundation and GLI, British Liver Trust redirected to Liver UK, and ELPA resolved at its current official domain `elpa.eu`
Epidemiology	Stated “62 of 63” despite source counts summing to 64	Removed invalid fraction and documented PMID 37040843's internal inconsistency
Angulo arithmetic	Repeated 193/619 as 33.2% without comment	Preserved the abstract's statement but flagged that the quotient is 31.2% (PMID 25935633)
Resmetirom NNT	Said fibrosis-improvement NNT about 8–9	Corrected to about 9–10 from 11.7- and 10.0-point absolute differences
Prognostic wording	Generalised that NITs “match or beat” biopsy and “double as” cardiovascular tools	Bounded to selected prognostic cohorts and stated that liver tools are not validated individual cardiovascular calculators
Search count	Retained 203 records for the surrogate search	Live exact re-run returned 204; updated the dated count
Reference closure	Three guideline citations appeared in the body without page-reference entries	Added PMIDs: 27053230, 38852583 and 38869512 to the reference list

Remaining flags¶

Rapirosiran enrollment remains unresolved between PMID 40581300 and NCT04565717; affected pages remain draft.
PMID 37040843 reports 63 studies but geographic components totaling 64; no fraction was reconstructed.
PMID 25935633 reports 193/619 as 33.2%; both the published statement and calculated 31.2% are shown.
PMIDs: 28624648, 33497764 and 42525136 still lack an abstract in PubMed; no findings were inferred.
No completed positive phase 3 cirrhosis trial, randomised clinical-outcome NIT pathway trial, resmetirom-plus-semaglutide trial, or before/after patient-experience study of the nomenclature change was identified in the dated live searches.
The bibliography now contains 274 wiki-cited papers; 14 additional unique records occur in literature-only artifacts, for 288 PubMed records across the condition.

Verification evidence¶

PubMed final extraction: 288 requested; all 288 present in the live E-utilities response.
ClinicalTrials.gov: all 62 current NCT IDs covered by same-session live v2 API records; no API errors in the original 58-record run, and all four newly cited records were retrieved live.
Exact condition registry snapshot re-run: 2,011 total studies, 178 interventional/recruiting, 110 phase 3.
All 274 wiki PMIDs occur in BIBLIOGRAPHY.md; no body citation is missing from its page's References section.
Final status count: 22 draft; 0 curated. CONDITIONS-ROADMAP.md was not touched.

2026-09-02 — Full build (Claude)¶

Built the condition from the seeded scaffold: 22 wiki pages, 2,919 lines, 272 verified records (259 cited by pages), the complete literature layer, a rewritten OPEN-QUESTIONS.md and an updated INDEX.md. Every page left at status: draft. No page promoted; the auditor must be a different engine.

What was written¶

All 22 pages in the canonical plan, none skipped: overview, nomenclature-and-definitions, epidemiology-and-burden, pathogenesis, genetics, noninvasive-assessment, histology-and-biopsy, natural-history-and-fibrosis-progression, lifestyle-and-weight-loss, resmetirom-and-thyromimetics, glp1-and-incretin-therapy, bariatric-and-metabolic-surgery, other-pharmacotherapy, cirrhosis-and-decompensation, masld-related-hepatocellular-carcinoma, cardiovascular-and-extrahepatic-outcomes, masld-and-type-2-diabetes, lean-masld, guidelines, clinical-trials-landscape, red-flags-and-safety-concerns, patient-experience-and-advocacy. Pages run 98–175 lines with 7–44 unique PMIDs each (deduplicated total 259).

Literature layer: BIBLIOGRAPHY.md (259 papers in 21 topic sections, generated from the pages so it is complete by construction, each line carrying tags and citing pages); notes/ (7 landmark notes — Rinella 2023, Angulo 2015, Romeo 2008, Abul-Husn 2018, Vilar-Gomez 2015, Harrison 2024, Sanyal 2025, with DOIs verified against PubMed articleids); guidelines/REGISTRY.md (47 catalogued documents, supersession chains, 7 named disagreements, watch list); statistics/STATISTICS.md (12 sections, 12 documented conflicts); patient-voice/ (README with method and ethics, organizations, themes, sources).

OPEN-QUESTIONS.md rewritten: 18 Tier 1 and 12 Tier 2 questions with stable IDs, plus a 12-row "Dots not yet connected" table. OQ-1 to OQ-5 carry forward from the seed with sharpened scope; OQ-6 to OQ-30 are new.

Anchor re-verification¶

All 13 anchor records resolved at scoping were re-fetched from live PubMed. All 13 matched on first author, year, journal, volume, issue, pages and title. None required correction.

Searches run (live PubMed E-utilities, 2026-09-02)¶

Roughly 60 topic searches, every one unioning both vocabularies. Scoping counts re-confirmed: the union query (MASLD OR NAFLD OR MASH OR NASH OR "steatotic liver disease") returned 78,992 records against 50,200 for the new name alone. Topic searches covered: nomenclature and MetALD (3,044); global epidemiology (1,081); NHANES prevalence (7,925); de novo lipogenesis, lipotoxicity, stellate cells, macrophages, microbiome (1,685–6,565 by facet); PNPLA3 (1,316), other variants and GWAS (1,023), heritability; FIB-4 and elastography (5,784 area), blood panels, sequential pathways (932); histology scoring, sampling variability (32), AI pathology (559); paired-biopsy natural history (259) and fibrosis-stage outcomes (203); lifestyle (3,895), exercise (907), Mediterranean diet and fructose (2,225); resmetirom and THR-β (513); GLP-1 and incretins (1,231); bariatric surgery (1,210); pioglitazone and vitamin E (1,317); failed agents (1,161); FGF21 (636); statins, SGLT2, metformin (1,053); cirrhosis and cACLD (138); transplant (281); HCC (5,216); cardiovascular (5,908); extrahepatic and CKD (1,300); type 2 diabetes (4,936); lean (2,043); guidelines (461 by publication type, 1,161 by phrase); alcohol boundary (2,611); drug-induced liver injury (1,166); stigma and quality of life (1,168); qualitative and lived experience (133); paediatric (4,916); omics (4,837); screening cost-effectiveness (577).

ClinicalTrials.gov v2 API, same date: 2,011 studies for the condition union; 178 interventional and recruiting; 110 phase 3. Individual NCT records for MAESTRO-NASH, ESSENCE, the FGF21 and incretin phase 3 programmes, the genotype-directed siRNA and antisense trials, and the terminated and withdrawn programmes were each retrieved and are cited with status, enrolment and dates.

Direct HTTP retrieval, same date: 13 patient organisations and public health portals verified; 2 recorded as unverifiable (see below).

Citation integrity¶

Every PMID in every page was extracted programmatically and re-fetched through PubMed esummary; the bibliography is generated from that verified metadata, so no citation in this condition was written from memory. Approximately 25 volume/issue/page errors introduced during drafting were caught by this pass and corrected before the pages were finalised. Two invented article-locator strings (e186418, e135197, for J Clin Invest records with no pages field) were caught and removed.

One asserted absence was checked and found false¶

The patient-experience page initially stated that no substantial interview-based qualitative literature existed, on the basis of a broad search (stigma OR "quality of life" OR "patient-reported outcomes" OR qualitative, 1,168 records dominated by instrument work). A targeted re-run — ("qualitative study" OR "semi-structured interview" OR "focus group" OR "lived experience" OR "illness perception"), 133 records — surfaced six interview and focus-group studies. The page was rewritten, the claim retracted, and the whole episode recorded in OQ-16 and in patient-voice/sources.md as a worked example of why an absence found under one search formulation is not an absence.

The corrected material changed the page's conclusions substantively: non-cirrhotic MASH is not experienced as asymptomatic (fatigue 18/23, right-upper-quadrant pain 14/23 in concept-elicitation interviews, PMID 33336323); diagnosis is preceded by misdiagnosis and followed by an information vacuum (PMID 38780312); and a second patient survey found 79.7% did not consider "fatty" stigmatising while 85.1% found "alcohol" in a disease name stigmatising, concluding that the MASLD change "does not seem warranted, at least if stigma is the main concern" (PMID 40337974) — a minority position against a multisociety consensus, recorded as such.

Scoping decisions taken during the build¶

Border with hepatocellular carcinoma held. masld-related-hepatocellular-carcinoma.md covers only what is distinctive in this setting — pre-cirrhotic occurrence, halved surveillance uptake, the attribution problem, and reduced immunotherapy responsiveness — and explicitly defers staging, treatment allocation and transplant criteria to conditions/hepatocellular-carcinoma/. Tan 2022 found no significant difference in treatment allocation between NAFLD-related and other HCC, which supports the border rather than straining it.
Border with type 2 diabetes held. masld-and-type-2-diabetes.md covers the bidirectional epidemiology and case-finding; glycaemic management is deferred.
NAFLD/NASH provenance stated throughout. Every trial and cohort is described by the criteria it actually enrolled under. No study population was relabelled MASLD retrospectively.
Two published confidence intervals that do not contain their point estimates (extrahepatic-cancer and liver-specific mortality in PMID 38521116) were reproduced as published and flagged, rather than silently corrected. Flagged again here for the auditor.
The Ha 2022 sarcopenia record (PMID 35347595) is an editorial, not the primary study its title suggests; it is cited as commentary and labelled as such.
The APASL 2020 MAFLD guideline (PMID 33006093) is indexed in PubMed as an editorial, not a practice guideline, despite its content and title; the registry records both facts.
259 of the 272 verified records are cited by a wiki page; the remaining 13 are guideline and consensus documents catalogued only in the registry. Both counts are stated where they are used.
Two patient organisations could not be verified — the American Liver Foundation (HTTP 403) and ELPA (connection failed). Both are widely referenced and neither is cited anywhere in this condition, because this repository does not cite pages it did not retrieve.

What could not be verified, and what remains open¶

No full text was retrieved for Avery 2017 (PMID 28624648), Alem 2021 (PMID 33497764) or Berg 2026 (PMID 42525136) — all carry no abstract in the PubMed record. They are listed in patient-voice/sources.md with their findings not summarised.
The AZD2693 phase 2 (NCT05809934, n=220, completed 2025-07-14) has no retrieved publication; the trial is cited from the registry with its status and dates, and the phase 1 programme from Armisen 2025.
NICE NG49 was verified by page title only; its recommendation content was not extracted and is not quoted.

Follow-ups for the auditor¶

Re-fetch every one of the 259 PMIDs and check claim-to-source correspondence, especially the numbers in STATISTICS.md §5, §8 and §9, which are the densest.
Re-run every asserted absence in OPEN-QUESTIONS.md with at least two search formulations. The base rate observed in this build is that one in one checked was wrong.
Verify the two flagged confidence intervals from PMID 38521116 against the full text.
Re-attempt the American Liver Foundation and ELPA sites, and re-verify all 13 organisation URLs — sites move.
Re-run the ClinicalTrials.gov queries; §12 of STATISTICS.md and the whole of clinical-trials-landscape.md are a single-date snapshot.
Check whether MAESTRO-NASH OUTCOMES (NCT05500222) has reported — its primary completion was December 2026, so it may have read out.
Look specifically for a published AZD2693 phase 2 result.
Only then consider promoting pages to curated, and only those where every citation resolved and matched.

Follow-ups for the next sweep¶

Watch the four outcome trials in the registry watch list; any readout changes resmetirom-and-thyromimetics.md, glp1-and-incretin-therapy.md, clinical-trials-landscape.md and OQ-1 simultaneously.
The qualitative literature is small enough that a handful of new studies would materially change patient-voice/themes.md.
The NIDDK public page (last reviewed April 2021) still states no medicines are approved; worth re-checking whether it has been updated, as it is the clearest measurable instance of public-facing lag.

2026-09-02 — Seeded (Claude)¶

Created the scaffold: INDEX.md with a 22-page plan (20 topical + 2 standing house pages) and a 13-record anchor table, a seed OPEN-QUESTIONS.md, this log, and the empty wiki/ and literature/ layers. No wiki pages written.

Searches run (live PubMed E-utilities, 2026-09-02). Scoping: metabolic dysfunction-associated steatotic liver disease OR MASLD 50,200; nonalcoholic fatty liver disease OR NAFLD 51,746; nonalcoholic steatohepatitis OR MASH 53,007. Per-page topic counts in INDEX.md, including epidemiology 11,395, type 2 diabetes 6,262, noninvasive fibrosis 5,784, HCC 5,271, cardiovascular 1,484, bariatric surgery 1,346, GLP-1 1,332, genetics 1,206, pathogenesis 644, resmetirom/THR-beta 434, histology 427, lifestyle 263, cirrhosis outcomes 239, lean 100, nomenclature 9.

Query-hygiene note. Four initial topic queries were malformed by unparenthesised OR; NASH AND resmetirom OR thyroid hormone receptor returned 16,427 rather than 434. Re-run parenthesised and marked † in INDEX.md.

Anchor records resolved live: 13, listed in INDEX.md.

Scoping decisions. The defining problem here is not size but vocabulary: the 2023 multisociety renaming split the literature across NAFLD/NASH and MASLD/MASH, with overlapping but non-identical definitions. Two binding rules are written into INDEX.md — every search must union the old and new vocabulary, and a study conducted under NAFLD/NASH criteria must be described as such rather than silently relabelled. An asserted absence found under one vocabulary is not an absence. Borders: this condition owns the liver disease to cirrhosis; the tumour stays with hepatocellular-carcinoma, with one dedicated page for what is distinctive about HCC arising in MASLD (notably pre-cirrhotic occurrence, which breaks surveillance assumptions).

Flagged for the build pass. Anchors must be re-verified. No anchor yet for EASL guidance, PIVENS/TONIC, diagnostic-accuracy studies for FIB-4 and elastography, bariatric histological cohorts, transplant outcomes, the MetALD alcohol boundary, or patient-organisation material.

Follow-ups. Build with tools/build-condition.sh masld; auditor must differ from the writer.