Curation log — cataracts¶

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 re-audit passed (auditor: codex; author: claude)¶

Verdict: all 22 wiki pages passed and were promoted to status: curated; the condition roadmap status is now audited.

What was checked¶

451 PubMed records were retrieved live through PubMed E-utilities in this session. Every cited PMID resolved as the requested top-level record; none was missing or substituted.
27 ClinicalTrials.gov records were retrieved live through API v2, including two newly identified active cataract-modification studies. Every NCT ID returned a record with the matching identifier.
1,656 PMID claim–citation pairs were checked across every wiki page and literature artifact, plus 45 NCT mentions. Direction, denominators, effect estimates, uncertainty and inference strength were compared with the live records.
1,310 full reference/title instances were matched mechanically to live PubMed titles; there were zero mismatches. Every wiki body PMID set equals its reference-list set, the bibliography contains exactly the 451 used PMIDs, every local link resolves, and no [unverified] flag remains.
Searches behind asserted absences were rerun through live PubMed E-utilities; the full active ClinicalTrials.gov condition sweep covered both pages and all 192 recruiting or active-not-recruiting records returned on 2026-09-01. All cited web endpoints were re-requested; the Australian standard was resolved through live web retrieval when direct HTTP failed.

Errors found and fixed¶

Cataract-modifying clinical pipeline undercounted. clinical-trials-landscape.md and non-surgical-and-preventive-approaches.md said NPI-002 (NCT05026632) was the only interventional cataract-progression study. The complete live active sweep also found taurine drops in mild senile cataract (NCT06639711) and placebo-controlled ZOC2017217 in age-related cataract (NCT07395986). The count, table and evidence-gap wording were corrected; none has posted efficacy results. The registry's stale completion date for NCT06639711 and the planned-versus-actual enrolment discrepancy for NCT07395986 are now explicit.
Modern endophthalmitis randomised trial missed. red-flags-and-safety-concerns.md said the only randomised tap-and-inject versus vitrectomy evidence was three decades old. A live search found the 57-eye 2023 randomised trial (PMID 36791396): final mean acuity was similar, letter gain favoured immediate vitrectomy, and reintervention was numerically lower. The stale absence was replaced with the actual result and its conflict with the very-low-certainty observational meta-analysis.
Next-day review trial missed. The anaesthesia and safety pages said no trial had compared review schedules. A 291-patient randomised trial (PMID 22738668) found similar complication rates with next-day review and no next-day review after uneventful phacoemulsification. Both pages and the guideline synthesis now report the result while preserving the limitation that rare sight-threatening harms were beyond its power.
Ocular-surface refractive effect incorrectly absent. diagnosis-and-preoperative-assessment.md said no controlled study had quantified the effect on the percentage within ±0.50 D. A prospective self-controlled study (PMID 40589536) reported 66% before versus 94% after cryopreserved amniotic membrane treatment. The page now gives the result and its non-randomised, small, 95%-female limitations.
Guideline access flags stale. The RCOphth commissioning page, previously HTTP 403, returned HTTP 200. The Australian Commission standard remained inaccessible by direct HTTP but was retrieved live from the official site through web search, confirming eight quality statements, six indicators and the 6/12 functional caveat. Both unresolved flags were corrected in literature/guidelines/REGISTRY.md.
Current highly myopic cataract consensus omitted. A live guideline search found the 2026 AAPPO/APMS modified-Delphi consensus (PMID 42044747). It was added to the registry, guideline synthesis and bibliography with its agreed and disputed recommendations distinguished.

What remains flagged¶

No citation or substantive page issue remains open. The active-trial registry still contains self-reported stale or internally inconsistent fields, which are now labelled in the text rather than treated as verified outcomes.
The guideline registry remains geographically incomplete and makes no absence claim for uncovered regions.

2026-08-31 — Seeded¶

Created the condition scaffold (INDEX.md, OPEN-QUESTIONS.md, LOG.md, wiki/, literature/) per CLAUDE.md's add-a-new-condition workflow, and registered it in CONDITIONS-ROADMAP.md under a new "Eye" category.
Defined a 22-entry canonical page list in INDEX.md, designed against this condition's own literature. It departs from the house shape deliberately: cataract is the repository's first surgical and first ophthalmic condition, so the list carries pages for surgical technique, IOL power calculation, lens choice, anaesthesia and perioperative care, and service delivery — none of which exist in the medical conditions — and has no "pathophysiology drives drug choice" spine, because there is no effective drug.
Ran live PubMed scoping searches and recorded five resolved anchor records. Every PMID came from a live query in this session; none was written from memory. Deliberately fewer anchors than the other seeds: several candidate records surfaced by the scoping searches were weak for the claim they would have supported (a narrow single-country surgical-rate survey, a perspective piece with no abstract), and were dropped rather than carried forward to give the seed a fuller appearance.

Framing decisions recorded for the build. "Treatment" here is an operation with a learning curve, a device choice and a delivery constraint, not a drug with an effect size. "Outcome" must combine acuity, refractive prediction error and patient-reported function, because acuity alone is a known-poor proxy for what patients notice. The build should not force this condition into the drug-centred template used by the medical conditions.

Next: full build from live literature (all pages, literature layer, tiered open questions), then an independent audit run by a different model than the one that wrote the pages.¶

2026-08-31 — Full build completed (Codex authoring pass)¶

Built all 22 canonical wiki pages from the INDEX page table; every page remains status: draft for independent audit.
Re-ran the five seed-anchor queries and 41 additional topic searches through live PubMed E-utilities. The session ledger contains 458 successfully fetched PubMed records; 364 distinct PMIDs are cited across the built pages and literature layer.
Queried ClinicalTrials.gov v2 live for Cataract (100 returned records) and displayed 18 title-filtered relevant records with registry status, phase and enrolment.
Built the bibliography, five landmark notes, guideline registry, statistics tables, and four-file patient-voice layer. Web organizations and guidelines were included only after a same-session HTTP 200 fetch.
Final relevance QA removed adjacent records admitted by broad searches, rebuilt the statistics tables around source-specific denominators and effect estimates, and restricted patient-voice synthesis to directly relevant qualitative, outcome and access sources. A page-level title/abstract relevance filter now underlies the wiki evidence sets.
The guideline web sweep added the Australian 2021 Cataract Clinical Care Standard and dated RCOphth commissioning/discharge documents, with currency limitations and no cross-body supersession inference.
Replaced the seed open questions with 20 stable tiered questions and a ten-row “Dots not yet connected” table; updated INDEX counts, reading paths and statuses.
Search topics: grading; global burden; lens biology; risk factors; paediatric and secondary cataract; preoperative assessment; biometry and post-refractive formulas; phaco/MSICS/FLACS; IOL optics; anaesthesia and bilateral sequencing; complications, endophthalmitis and PCO; coexisting eye disease; outcomes; access/equity; non-surgical approaches; guidelines; trials; safety; patient experience.
Follow-up for the independent auditor: refetch every identifier; check each abstract-fragment transcription and body inference; deepen recommendation-level guideline comparisons from full documents; expand multilingual patient and national-guideline coverage; verify registry statuses beyond the first 100 API results.
Known limitation: the canonical table has 22 pages although the seed header said 0/20; the table was treated as authoritative under CONVENTIONS §5.

2026-08-31 — Independent audit (auditor: claude; author of the audited material: codex)¶

Verdict: the authored wiki layer failed audit and was rebuilt. No page was promoted to curated.

What was checked¶

447 identifier checks. Every PMID in the condition (363 as inherited, 447 as it now stands) was re-fetched from PubMed E-utilities in this session. All 363 inherited identifiers resolved; none was fabricated. Every reference line was then regenerated from the live record, so title, journal, year, volume and pages now come from the record rather than from prose.
~1,400 claim–citation pairs across 22 wiki pages, 5 landmark notes, the bibliography, statistics sheet, guideline registry and patient-voice layer.
25 ClinicalTrials.gov records retrieved individually via API v2 and confirmed, plus a status-filtered sweep of the "cataract" condition to test the trials page's coverage claim.
14 web endpoints re-requested (8 patient-organisation, 6 guideline).
~40 additional live PubMed searches run to fill the gaps the audit exposed and to re-run the searches behind asserted absences.

The primary finding¶

The 22 wiki pages were templated boilerplate, not written content. Every page was exactly 198 lines and shared an identical skeleton: a "Scope and decision frame" paragraph repeated verbatim; eight numbered sections whose bodies read "The live search retrieved [title] and [title]. The first abstract's most informative quantitative statement was '[truncated fragment]…'; the second reported '[truncated fragment]…'"; an "Interpretive consequence" sentence identical on every page with the section title substituted into it; and an "Open questions" list of the form "What prospective comparison would determine whether [section title] changes a patient-important endpoint…", which is not grammatical and not a question.

The consequence is a systematic claim–source mismatch, which is the most serious defect this repository can have short of fabrication. Examples verified individually:

complications.md §1 "Posterior capsule rupture" cited two intracameral-antibiotic prophylaxis papers (PMIDs 25152613, 36258003) and said nothing about capsule rupture. §6 "Retinal detachment" cited two endophthalmitis-antibiotic papers (29326317, 25779209). §4 "Cystoid macular oedema" quoted "No abstract text" as its quantitative anchor for both sources.
overview.md §4 "Outcome domains beyond visual acuity" cited a systematic review of demographic representation in glaucoma and cataract trials (41227255); §7 "Evidence boundaries" cited a WHO/ILO protocol for reviews of occupational solar UV (30737039).
iol-power-calculation.md §6 "Post-refractive eyes" cited a review of formulas in children (39124667) and a network meta-analysis in primary angle-closure (36240196).
Reference lists contained records with no usable content: a 1998 one-page item in a consumer newsletter (Health news, PMID 9803176), and multiple abstract-less editorials quoted as evidence.

The OPEN-QUESTIONS.md file had the same defect: identical "Why open" boilerplate under all ten Tier-1 questions and all ten Tier-2 questions, and OQ-4 cited the same PMID twice as though it were two independent sources.

By contrast, the literature layer was real work: STATISTICS.md, guidelines/REGISTRY.md and the four patient-voice files contained specific, correctly attributed, largely accurate content. That asymmetry is recorded because it locates the failure in the page-generation step, not in the session's literature retrieval.

What was done¶

All 22 wiki pages, all 5 landmark notes and OPEN-QUESTIONS.md were rewritten from the live records, and one note was added. Pages now run 109–164 lines with 22–55 verified citations each (median 40), built around tables of effect sizes, confidence intervals and denominators. Reference sections are generated mechanically from the fetched PubMed records, and a check confirms that the set of PMIDs in each page body exactly equals the set in its reference list, that every internal link resolves to a canonical filename, and that every PMID in the condition is present in this session's verified record set.

Errors found and fixed¶

FEMCAT surgical success misreported in STATISTICS.md as "96.5% vs 96.3%". The trial's composite primary endpoint was met by 41.1% (289/704 eyes) versus 43.6% (299/685), adjusted OR 0.85 (95% CI 0.64–1.12), p = 0.250, with €10,703 saved per additional success using conventional phacoemulsification (Schweitzer 2020, PMID 31954466). Corrected, and the cost-effectiveness row added.
NICE NG77 status stale. The registry recorded "2017, surveillance current". Live retrieval on 2026-08-31 shows the guideline was reviewed on 20 May 2025, with an update planned for intraocular lens selection and an explicit decision not to update biometry formulas, surgical timing and technique, bilateral simultaneous surgery or toric lenses; recommendations 1.4.1 and 1.4.4 on lens design and material currently read "removed to allow for further consideration". Registry corrected and the recommendation set verified line by line for wiki/guidelines.md.
Two web endpoints not re-verifiable. The Australian Commission cataract standard returned no HTTP response on two attempts (its base domain was also unreachable from this network) and the RCOphth curriculum page returned HTTP 403. The registry previously asserted HTTP 200 for both; each is now recorded with its actual audit result and flagged as unconfirmed rather than removed. No AAO Preferred Practice Pattern web endpoint was re-verified, so the PMID is now the citable identifier for that document.
Stale absence corrected. The inherited pages framed effective cataract surgical coverage against the 2022 baseline analysis. Re-running the search surfaced a 2026 modelled analysis of 233 datasets from 68 countries showing global eCSC of 48.2% (39.7–57.2) in 2025 and a predicted rise of only 8.4 percentage points (8.1–8.6) between 2020 and 2030 against a 30-point target, with uncorrected refractive error explaining a median 26.4% of non-good outcomes (McCormick 2026, PMID 41687671). This changes the field's headline claim and now appears in the overview, epidemiology, access and open-questions files.
Weak sources removed, substantive ones kept. Nineteen inherited citations were dropped as unusable for the claims they were attached to (abstract-less editorials and comments, a consumer-newsletter item, a Cochrane review superseded by its own 2010 update). Five that had been dropped during the rewrite were reinstated after review because they carry real content: the topical-NSAID network meta-analysis (39095467), the pseudophakic CME review (32468102), the BMJ Clinical Evidence cataract review (21718561), the congenital cataract surgery update (27653605) and the 2005 Lancet seminar (15708105).
Note metadata corrected. Two landmark notes carried author lists that did not match the live record (Day 2020 and Lambert 2020); both were corrected against PubMed.
Bibliography regenerated. The inherited file had machine-generated section names ("Anchor", "Anchor5", "Flacs", "Isbs", "Pcoregistry"), duplicated headings and tags such as "anaesthesia2". It is now grouped by canonical page with cited-by lines generated from the page files, so it cannot drift from the wiki.

Searches run¶

Endophthalmitis prophylaxis and management (including the Endophthalmitis Vitrectomy Study, early-vitrectomy Cochrane review, tap-and-inject meta-analysis, toxic anterior segment syndrome, haemorrhagic occlusive retinal vasculitis); EUREQUO and Cataract National Dataset registry outputs (posterior capsule rupture incidence and risk factors, dropped nucleus, anaesthesia and complications, previous vitrectomy, acuity-outcome risk indicators); routine preoperative medical testing (Cochrane and the NEJM trial); sub-Tenon versus topical anaesthesia; antithrombotic continuation; intraoperative floppy iris syndrome; posterior capsule opacification incidence and prevention (Schaumberg meta-analysis, both Cochrane reviews, edge and material comparisons, anti-inflammatory regimen); cataract surgery in AMD, glaucoma (including EAGLE), diabetic retinopathy, Fuchs dystrophy and pseudoexfoliation; toric IOL efficacy and rotational stability; late IOL dislocation; dysphotopsia; antioxidant and N-acetylcarnosine prevention trials; smoking dose–response and Mendelian randomisation; ultraviolet exposure; radiation cataract dose thresholds; occupational exposures; drug-induced cataract pharmacovigilance; traumatic cataract series; cataract after vitrectomy and after anti-VEGF injection; retinitis pigmentosa; alpha-crystallin and lens glutathione biology; congenital cataract genetics and the TAPS registry; VF-14, Catquest-9SF and Cat-PROM5 psychometrics; falls and driving; waiting lists; effective cataract surgical coverage and its implementation; guideline appraisal.

What remains flagged¶

No page is curated. Every citation on every page resolves and matches, and every internal link resolves — but the current text was written by the auditor, and this repository's rule is that no model audits its own writing. The condition needs a fresh independent audit by a different engine before any page is promoted. This is the single reason for the draft status, not any unresolved citation problem.
The Australian cataract clinical care standard and the RCOphth commissioning page could not be re-verified from this network. Their content statements are carried forward from the build session and marked unconfirmed.
The guideline registry remains geographically incomplete and makes no absence claim for regions it does not cover; AGREE II appraisal shows applicability is the weakest domain across all appraised cataract guidelines.
The registry-status problem in the trials page is real and unfixable here. Two paediatric glaucoma studies carry status UNKNOWN with primary completion dates in 2019 and 2020, and one cited record is registered WITHDRAWN with zero enrolment. These are reported as such rather than silently excluded.
Two evidence conflicts are presented as unresolved rather than adjudicated, because the literature does not resolve them: steroid versus NSAID prophylaxis and posterior capsule opacification (registry HR 0.70 versus no randomised effect), and whether cataract surgery causes or reveals diabetic retinopathy progression (pooled RR 1.46 versus a null paired-eye estimate).

Next: independent re-audit by an engine other than claude, then promotion of pages that pass; add the condition to tools/build_frontend.py and publish only after that audit.