Curation log — irritable bowel syndrome

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 (Codex; author: Claude)

Role and scope. Codex independently audited material authored by Claude. The audit covered all 20 files in wiki/ and all 13 artifacts in literature/: the bibliography, guideline registry, statistics sheet, six landmark notes and four patient-voice files. No shared roadmap or other condition was changed.

Live record checks. The final identifier set contained 378 unique PMIDs and 63 unique NCT identifiers (441 unique identifier records). Every PMID was re-fetched from PubMed E-utilities after the final edits; all 378 were returned as primary MedlineCitation/PMID records. Every NCT identifier was fetched separately from clinicaltrials.gov/api/v2/studies/<NCT> after the final edits; all 63 endpoints returned a record whose identificationModule.nctId exactly matched the requested identifier. The audit also checked 1,674 substantive claim–identifier pairs outside bibliography/reference lists: 1,589 claim–PMID pairs and 85 claim–NCT pairs across 32 content artifacts.

Searches re-run. A fresh ClinicalTrials.gov condition query returned 1,171 IBS records. Status and phase totals, recent phase 2/3 studies and intervention-specific counts were recalculated from that snapshot. Thirty-five targeted PubMed E-utilities searches were run to test negative assertions, including long-term dietary follow-up, mortality, consultation quality, sex-specific qualitative evidence, geographic coverage, criteria-stratified treatment, head-to-head therapies, long-term IBS-C efficacy, donor allocation in FMT, stigma interventions, response predictors, adult labelling replication, open-label placebo guidance and α2δ-ligand efficacy. Broad searches were narrowed when their Boolean scope admitted unrelated records. The three official NICE URLs and 17 unique patient-organisation/source URLs were also requested live.

Errors found and fixed. Every item below changed the audited material.

1. Four invalid NICE PMIDs. PMIDs 21656972, 31825575, 32073807 and 33939353 did not resolve to the institutional NICE documents claimed. They were removed from the guideline page, guideline registry and bibliography. NICE CG61 and NG193 are now cited by their live official URLs, with publication/update/surveillance dates taken from the NICE pages.
2. Bibliography incompleteness and generated count drift. Eight valid records already cited in the condition were absent from BIBLIOGRAPHY.md (PMIDs 11053260, 11570936, 15489576, 17488783, 21350320, 22489905, 29950604 and 37752291). They were added from live records. New evidence found during audit was also added, and all section labels were reconciled to their actual entry counts. The final bibliography contains 378 unique PubMed records plus two institutional NICE documents.
3. Wrong first author. PMID 35183112 was attributed to Björkman; the live record identifies Frändemark as first author. The bibliography, annotated patient-voice source and thematic synthesis were corrected.
4. Stale long-term low-FODMAP absence. The claim that controlled evidence did not extend beyond months was contradicted by a 12-month randomised follow-up (Tunali 2026, PMID 42623122). The diet page and statistics sheet now report 12-month response of 62.5% versus 34.5%, risk difference 28.0 percentage points (95% CI 4.2–47.7), while preserving the microbiome-specific limitation.
5. Stale mortality absence. A search found an NIH–AARP cohort of 132,697 participants, including 5,030 with claims-identified IBS (Gutiérrez-Torres 2025, PMID 41518229). The epidemiology page and statistics sheet now report the time-varying estimates—HR 0.75 (95% CI 0.64–0.88) at >2–5 years and 0.92 (0.87–0.98) after >17 years—without interpreting the association as protective.
6. Overbroad long-term IBS-C statement. The original “no trial beyond 26 weeks” wording ignored a 52-week lubiprostone safety study and a prospective 52-week linaclotide cohort (Yiannakou 2018, PMID 30302125). The page now distinguishes absence of placebo-controlled long-term efficacy from available open-label safety and observational effectiveness evidence; the linaclotide estimate is −70.7 IBS-SSS points (95% CI −95.0 to −46.5) among 76 paired observations.
7. Stale sex-specific qualitative gap. Searches retrieved a gender-focused 19-interview study (PMID 24224642) and a work-life study with 15 women and 8 men (PMID 35183112). Patient-voice sources and themes now state that men's accounts are sparse rather than absent.
8. Stale geographic qualitative gap. An Australian low-FODMAP qualitative study (PMID 42586418) and a Chinese social-media analysis (PMID 37093542) were retrieved. The eligible Australian study was incorporated. The Chinese study is recorded but excluded from synthesis under the existing no-social-media ethics rule. The residual geographic gap is scoped to regions not represented in the targeted search, rather than “non-Western evidence” generally.
9. Overbroad patient-knowledge gap. A 2016–2017 study of 204 laypeople (Sherwin 2018, PMID 29319900) documented continuing misconceptions. The page now asks for a recent comparable national patient survey rather than claiming there was no later evidence at all.
10. IBS Network status and link failure. The Charity Commission record now marks charity 1173208 as removed, and the historical domain serves unrelated commercial material. The organisation file now describes it as a former charity and explicitly warns against using the old site for signposting.
11. Consultation-quality overstatement. POACHER (PMID 41922986) evaluates nurse-led structured investigation and management, not consultation quality in isolation. It is now presented as adjacent implementation evidence, while the narrower isolated-intervention gap is dated. Cross-trial claims that consultation effects were “comparable to” or exceeded drug effects were removed.
12. Missed α2δ-ligand efficacy trial. The neuromodulator page said no efficacy trial followed the 2007 pregabalin mechanism study. A 330-participant randomised trial of the related ligand PD-217,014 found no benefit at either dose or in Rome II/IV bowel-habit subgroups (Houghton 2025, PMID 39812493). That negative class-level result is now included while retaining the narrower finding that pregabalin itself lacks a retrieved symptom-efficacy trial.
13. Clinical-trial search/count methodology. Intervention totals described as registry “mentions” were not reproducible as condition-plus-intervention API counts. The page and statistics sheet now state the query method and the reproducible counts: probiotic 133, FODMAP 97, CBT 50, FMT 37, linaclotide 27, rifaximin 26, hypnotherapy 22, acupuncture 18, vagus 10, tenapanor 9, amitriptyline 4 and ebastine 4.
14. Pipeline omissions and exclusivity claim. The live registry contained two recruiting Lilly phase 2 brenipatide studies omitted from the landscape (NCT07545759, planned n=531 in IBS-D; NCT07545772, planned n=342 in IBS-C) and an active phase 2 psilocybin-plus-psychotherapy study (NCT06206265). These were added. Wording implying camlipixant was the sole large current novel programme was removed.
15. Multi-condition enrolment overstatement. NCT04014413 reports n=450 for a multi-condition FMT platform, not an IBS-only cohort, and NCT05633706 reports n=300 across seven conditions plus controls, not 300 participants with IBS. Both rows now label total-study enrolment and the absence of an IBS subgroup target.
16. Undated permanent absences. Negative assertions about criteria stratification, cluster-based allocation, barostat-guided allocation, combined mechanism algorithms, missed microscopic-colitis outcomes, post-infectious prevention, head-to-head neuromodulator/behavioural therapy, secretagogue comparisons, response predictors, long-term antispasmodic efficacy, adult label replication, stigma intervention and open-label placebo guidance were rewritten as positively stated, dated search findings with their scope.
17. Comparative-efficacy overstatement. Claims that psychological therapy was the “largest” treatment effect and that consultation quality was quantitatively comparable with licensed drugs relied on indirect cross-trial comparisons. They were replaced with within-study estimates and explicit warnings that heterogeneous populations, controls and endpoints prevent direct ranking.
18. Guideline reference numbering. Removing the invalid NICE PMIDs left a four-number gap in the guideline reference list. The list was renumbered consecutively without changing the remaining records.

Internal consistency checks. Every PMID used in a wiki page or literature artifact is present exactly once in the bibliography's unique record set. Every PMID used in the body of a wiki page appears in that page's reference section. Live-record first-author checks found no remaining mismatch after correcting PMID 35183112. Bibliography section labels equal their actual entry counts. No [unverified], TODO, TBD, placeholder or unresolved identifier remains in wiki/ or literature/.

Remaining provenance limitations. Automated retrieval still receives HTTP 403 from IFFGD, About IBS and badgut.org. Their entries remain explicitly described as based on search-result excerpts rather than direct page text and are not used for numerical claims. The historical IBS Network site remains unsafe for signposting. These are access/provenance limitations, not unresolved PMID/NCT or substantive evidence claims.

Promotion result. 20/20 wiki pages promoted from draft to curated; 0 remain draft. Promotion occurred only after the final 378-PMID and 63-NCT live resolution passes and the claim/reference consistency checks above.

2026-09-02 — Full build (Claude)

What was built. All twenty planned wiki pages, the complete literature layer, a rewritten OPEN-QUESTIONS.md and an updated INDEX.md. The condition moves from seeded to built. No page was promoted beyond status: draft; the auditor must be a different engine.

Totals. 20 wiki pages, ~3,000 lines; 364 unique PMIDs, every one resolved live from PubMed E-utilities during this session; 38 NCT identifiers, every one retrieved live from the ClinicalTrials.gov v2 API on 2026-09-02; 6 landmark notes; a 32-document guideline registry; a ~250-row statistics sheet; a four-file patient-voice layer; 30 open questions (OQ-1 to OQ-30) and 12 dots-not-yet-connected.

Anchor re-verification. All twelve anchor records from the seed session were re-fetched and confirmed (author, year, journal, volume, pages): PMIDs 32294476, 32702295, 33315591, 29950604, 37858323, 39397570, 30177784, 40258375, 31859183, 38643782, 28846594, 21208106. None was wrong.

Gaps the seed flagged, and how they were closed. - Rome IV primary document — Drossman 2016 (PMID 27144617) and Palsson 2016 (PMID 27144634). Plus an unanticipated finding: Rome V was published in May 2026 (Drossman 2026, PMID 42031435), with a global epidemiology survey (Sperber 2026, PMID 42613194) and a first diagnostic-accuracy study (Staller 2026, PMID 42392123). This substantially changed the diagnosis and epidemiology pages and is the single most important thing the seed could not have known. - BSG/AGA/NICE guidelines — closed, plus JSGE, Seoul Consensus 2025, Italian joint consensus, UEG/ESNM, APAGE, Asian Consensus, AMG Mexico, WGO, German S3, BDA, ESPGHAN/NASPGHAN: 32 documents in the registry. - Gut-directed hypnotherapy trials — closed (Lövdahl 2022, PMID 35505463; Lövdahl 2025 two-year follow-up, PMID 40491242; Berry 2023 digital, PMID 37391055; Devenney 2024, PMID 37927144). - Linaclotide and tenapanor registrational trials — closed (Chey 2012, PMID 22986437; Chey 2021, PMID 33337659; plus plecanatide, lubiprostone, MD-7246). - Eluxadoline pancreatitis safety — closed (Cash 2017, PMID 27922029; Gawron 2018, PMID 28804032). - Open-label placebo trials — closed (Kaptchuk 2010, PMID 21203519; Lembo 2021, PMID 33605656; Nurko 2022, PMID 35099543; Ballou 2022 predictors, PMID 35412513). - Post-infectious IBS cohorts — closed (Walkerton series, Klem 2017, Porcari 2024, Svendsen 2019, Zanini 2012, Hanevik 2014, Villani 2010 genetics, Dunlop 2003 prednisolone). - Bile acid diarrhoea diagnostics — closed (Slattery 2015, Wedlake 2009, Valentin 2016, Dilmaghani 2025 sequestrants). - Patient-organisation material — closed by direct web retrieval, with failures recorded (see below).

Searches run (live PubMed E-utilities, 2026-09-02). Approximately 120 distinct queries across: Rome criteria and their validation; subtyping, Bristol stool scale and latent-class analysis; prevalence by region and criteria; sex differences; cost of illness; work productivity; coeliac, calprotectin, colonoscopy yield, microscopic colitis, bile acid diarrhoea, exocrine pancreatic insufficiency, giardiasis, lactose, ovarian cancer misdiagnosis, alarm features; visceral hypersensitivity and barostat methodology; brain imaging; mast cells, ebastine, nerve-fibre outgrowth; enterochromaffin cells and the female sex bias; GWAS and Mendelian randomisation; early-life adversity; microbiome composition, SCFAs, methanogens and hydrogen sulphide, probiotics by strain, FMT; post-infectious incidence by pathogen and its genetics and mucosal biology; anti-CdtB/anti-vinculin; low FODMAP and its comparators, reintroduction, fibre, gluten and wheat; peppermint oil and antispasmodics including mechanism and the labelling trial; every licensed IBS-C and IBS-D drug plus tegaserod's regulatory history; rifaximin, neomycin, SIBO prevalence and breath-test thresholds and their conflict-of-interest dispute; neuromodulators agent by agent; behavioural therapy networks, delivery modality, mindfulness and physical activity; placebo and nocebo magnitude, components, open-label placebo, endpoints and IBS-SSS; functional dyspepsia overlap, GORD overlap, central-sensitisation comorbidity, the duodenal hypothesis; guidelines worldwide; stigma, invalidation, patient educational needs, suicidality. ClinicalTrials.gov v2 API: total and status/phase breakdowns, recruiting studies, phase 2/3 studies started 2021 or later, and targeted queries for digital therapeutics, hypnotherapy, CBT, FMT, probiotics, FODMAP, rifaximin, linaclotide, tenapanor, amitriptyline, ebastine, acupuncture and vagus-nerve stimulation.

Query-hygiene note. The seed's caution about unparenthesised boolean OR was carried forward; all multi-term queries this session were parenthesised or issued as single phrases, and counts quoted in pages are stated with their retrieval date.

Findings that changed the shape of the build. 1. Rome V exists. Published May 2026, with global prevalence at 8.5% and DGBI overall unchanged from Rome IV (40.9% vs 40.5%), but only fair-to-moderate agreement with either predecessor (κ 0.36–0.55) and sensitivity of 66.1% — below Rome III and Rome IV — in its first accuracy study. Every guideline in the registry predates it. 2. Diet beats drugs in two independent randomised trials. CARIBS (specialist clinic, versus optimised medical treatment) and DOMINO (primary care, versus otilonium bromide, using a smartphone app). DOMINO was not in the seed anchor set and was found during the build. 3. A randomised 2×2 factorial trial separates drug from label. Mebeverine itself did nothing; the label "mebeverine" nearly doubled treatment success (OR 2.84, 1.52–5.34). This reframes the older antispasmodic meta-analyses and is now cited on three pages. 4. Eluxadoline's post-marketing pancreatitis signal is larger than the trial data suggest — 16.4% of 597 adverse-event reports, 53 hospitalisations, two reported fatalities. 5. The blinded FODMAP reintroduction control triggers symptoms in 26% of exposures, quantifying the nocebo share of dietary trigger identification. 6. Only 3% of rural primary-care patients with IBS report IBS alone without a comorbid central-sensitisation condition.

Contradictions recorded rather than resolved (all presented side by side with both sources): Rome IV prevalence lower than Rome III in three syntheses and higher in one; FMT effective in one large trial and null in two, with meta-analyses disagreeing on whether an effect exists; Escherichia probiotic strains rated highest-certainty benefit by one 2023 synthesis and ineffective by a 2026 strain-specific one; coeliac screening in IBS supported by one 2025 meta-analysis and questioned for North America by a 2017 one; SSRIs recommended against by AGA and found modestly effective by the 2025 meta-analysis; quality of life comparable to stroke by EQ-5D and better than rheumatoid arthritis by SF-36; peppermint oil ranked first in a network meta-analysis and failing both co-primary endpoints in its best single trial; breath-test SIBO in 35.5% of patients and 29.7% of controls.

Absences searched and dated (2026-09-02). No prevention trial in post-infectious IBS using the validated risk score; no head-to-head neuromodulator-versus-behavioural-therapy trial; no criteria-stratified treatment analysis; no cohort quantifying incident eating-disorder diagnoses after low-FODMAP instruction; no population mortality follow-up in criteria-defined IBS; no replication of the 2004 multi-tier suicidality survey; no four-arm open-label-placebo drug trial; no German S3 successor guideline; no Canadian IBS guideline; no phase 3 result for statin-based methanogen inhibition. Each of these is stated in the pages as a dated absence, not as a permanent fact — re-run before repeating.

Patient-voice layer: retrieval outcomes recorded honestly. Seven organisational URLs were retrieved successfully (Guts UK, IBS Patient Support Group, World IBS Day, MAGDA, PDS Belangenorganisatie, Rome Foundation, NHS). Three returned HTTP 403 to automated retrieval (IFFGD, aboutibs.org, badgut.org) and are described from search-result excerpts of the organisations' own pages, flagged as such. One finding is recorded deliberately: theibsnetwork.org resolved on 2026-09-02 to a cryptocurrency-casino review site, not to the UK IBS charity. The charity is independently verified as registered (Charity Commission 1173208) but its current live web address was not established this session and must be re-verified before the entry is used to signpost anyone.

Ethics. The patient-voice layer used public sources only; no social media, forums or individual stories; paraphrase by default with quotations capped at 15 words and attributed; no private individuals named (the single named person, Jeffrey Roberts, is a public organisational founder named on his organisation's own page); themes reported in aggregate with sample sizes given.

Judgement calls made during the build (recorded because no question could be asked mid-session): - Functional dyspepsia was written as an overlap page only, per the brief. Its own mechanism, subtypes and treatments are deliberately not curated. - Paediatric evidence was cited where it is the only or the decisive evidence (the mebeverine labelling trial, the open-label placebo trial in children, paediatric guideline documents) and explicitly flagged as paediatric each time; this knowledge base is adult-scoped. - Two adjacent guidelines (NICE chronic primary pain; ACG SIBO) were included in the registry as "adjacent" because they materially govern IBS practice. - The clinical-trials page carries fewer PMIDs than other pages because its primary evidence is the trial registry itself; 38 NCT identifiers are listed and dated instead. - Cross-condition links were made to anorexia nervosa, fibromyalgia, depression, generalized anxiety disorder, migraine and PTSD, all of which exist in this repository.

Not verified / flagged for the auditor. - The three organisation entries resting on search-result excerpts rather than direct retrieval (IFFGD, About IBS, badgut.org) should be re-tested; if retrieval succeeds, the entries should be upgraded. - The IBS Network's current URL is unresolved and must be established before use. - Two citation details were deliberately left partial rather than completed from inference: Ju 2024 (PMID 38892659) and Ng 2023 (PMID 37110143) are cited with volume only, because issue and page numbers were not present in the retrieved records. - The Kruis 1984 sensitivity/specificity pairs in STATISTICS.md are two operating points from one weighted score, not two independent estimates; this is annotated in the file's caveats section. - Several 2026 records are "online ahead of print" and carry no final volume/pages; they are cited with the retrieval note rather than an invented citation string. - The seed's per-page PubMed topic counts were not reused, because the exact query strings behind them were not recorded. Where a sub-literature size is quoted in a page, the query string is printed alongside the count and the date and was run in this session (e.g. irritable bowel syndrome AND (microbiome OR microbiota) = 2,797 on 2026-09-02, replacing the seed's 1,613). The two scoping counts in the seed (21,184 and 9,618) were re-run and confirmed unchanged.

Follow-ups for the next session. 1. Audit — a different engine must re-fetch every cited record, check each claim against it, re-run the searches behind every asserted absence, and only then promote pages to curated. The log entry must name both engines. 2. Re-test every URL in literature/patient-voice/sources.md section C before the layer is used for signposting. 3. Watch for guideline revisions incorporating Rome V, and for the camlipixant (NCT07519395) and Norwegian FMT (NCT04691544) readouts. 4. Consider whether functional dyspepsia warrants its own condition once IBS is audited; the overlap page currently carries more of its mechanism than a pure overlap page should. 5. CONDITIONS-ROADMAP.md was not modified this session, per the build brief (a concurrent agent may be editing it); its status entry for this condition needs updating to built by whoever owns that file.

---

2026-09-02 — Seeded (Claude)

Created the scaffold: INDEX.md with a 20-page plan (18 topical + 2 standing house pages) and a 12-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: irritable bowel syndrome 21,184; irritable bowel syndrome[MeSH Major Topic] 9,618. Per-page topic counts in INDEX.md, including epidemiology 3,607, microbiome 1,613, placebo response 813, low FODMAP 684, constipation pharmacotherapy 557, Rome IV 533, D-IBS pharmacotherapy 371, neuromodulators 337, visceral hypersensitivity 336, post-infectious 319, functional dyspepsia overlap 259, psychological therapy 112, bile acid malabsorption 113, rifaximin/SIBO 65, stigma 34.

Query-hygiene note. Two initial queries were malformed by unparenthesised OR (eluxadoline OR loperamide diarrhea returned 1,223 rather than 371). Re-run parenthesised and marked † in INDEX.md.

Anchor records resolved live: 12, listed in INDEX.md.

Scoping decisions. First curated condition in the digestive category, and the page list deliberately departs from the drug-centred house shape: diet, psychological therapy and the diagnostic label itself carry as much weight as pharmacotherapy, because that is where the evidence is. Borders: the condition owns IBS and not disorders of gut–brain interaction generally, with functional dyspepsia framed as overlap; organic disease (IBD, coeliac, microscopic colitis, bile acid diarrhoea) appears only as differential, scoped to how it is excluded and how often it is missed. IBD is not curated in this repository and must not be quietly built here.

The placebo response is scoped as a page, not a caveat. In a condition diagnosed by symptom report and treated for symptom relief, the magnitude and mechanism of the placebo response — including open-label placebo trials — is substantive evidence about the condition rather than a footnote on trial design. 813 records support it.

Flagged for the build pass. No anchor yet for the Rome IV primary document, BSG/AGA/NICE guidelines, gut-directed hypnotherapy, linaclotide and tenapanor registrational trials, eluxadoline pancreatitis safety reports, open-label placebo trials, or patient-organisation material.

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