Skip to content

Curation log — colorectal adenocarcinoma

2026-08-28 — Seed

  • Created: four seed files, 18 canonical pages, overview with 25 PMIDs, and 7 seed questions.
  • Searches run (live PubMed E-utilities esearch/esummary): global burden and young-onset epidemiology; adenoma–carcinoma/serrated pathways and CMS; Lynch/FAP; USPSTF, NordICC and FIT screening; MOSAIC; DYNAMIC ctDNA; RAPIDO, PRODIGE 23, STELLAR and OPRA; TRIBE2, cetuximab/bevacizumab, BEACON and KEYNOTE-177; metastatic review.
  • Verification: every cited PMID was returned live by PubMed. One recalled historical rectal-surgery identifier returned no usable metadata and was excluded.
  • Next: full build queued per CONDITIONS-ROADMAP.md.

2026-08-29 — Full build

  • Scope completed: wrote all 18 planned wiki pages. Each page is 182–223 lines and carries 25–49 verified PubMed citations, except the clinical-trials landscape, which combines 19 verified PubMed papers with 31 live-verified ClinicalTrials.gov records.
  • PubMed retrieval: ran 131 live E-utilities searches across epidemiology, carcinogenesis, pathology, molecular biology, hereditary syndromes, screening, localized and metastatic treatment, metastasis-directed therapy, survivorship, safety, guidelines, biomarkers and patient experience. The combined live cache contains 613 retrievable PubMed records; 301 unique papers are cited in the condition and indexed in literature/BIBLIOGRAPHY.md.
  • ClinicalTrials.gov retrieval: ran 15 live API-v2 searches covering adjuvant ctDNA, rectal organ preservation, neoadjuvant immunotherapy, metastatic targeted therapy, KRAS/BRAF/HER2, cellular and vaccine approaches, liver metastasis and young-onset disease. The union cache contained 445 NCT records; the landscape page selected 31 records with status and design context.
  • Literature layer: built the master bibliography; six landmark-paper notes; a 28-document guideline registry with supersession chains and disagreement matrix; a 164-line statistics sourcebook; and a four-file patient-voice layer.
  • Patient voice: fetched ten organizations' own public sites across North America, Europe, Australasia, Malaysia and South Africa; documented visible services and limitations; and synthesized 12 aggregate themes under a minimum two-independent-source rule. No private groups, named private individuals or individual testimonials were used.
  • Research frontier: replaced the seed list with 30 stable-ID questions in two tiers and a 12-row “Dots not yet connected” table.
  • Indexing: updated INDEX.md with reading paths, canonical links, page statuses and literature-layer counts; changed the condition roadmap state from seeded to built.
  • Citation integrity: every PMID written during this build was checked against the session's live PubMed cache; every NCT identifier was checked against the session's live ClinicalTrials.gov cache. Recalled identifiers that were not present in those caches were rejected or replaced with the live-verified record.
  • Deliberate limits: no formal search-protocol registration, dual-review screening or risk-of-bias adjudication was attempted, so this is a deep evidence map rather than a systematic review. Patient-experience coverage remains weighted toward English-language, high-income-country sources. Living commercial guidelines were represented by retrievable peer-reviewed guideline records and the registry flags version volatility.
  • Audit state: full build completed, but no separate audit was performed in this session. All wiki pages therefore remain status: draft, as required by the build playbook.

2026-08-30 — Full independent audit

  • Scope checked: audited all 18 canonical wiki pages and all 13 Markdown artifacts in literature/, plus OPEN-QUESTIONS.md, INDEX.md and the condition-level bibliography relationships. All pages remained within the 150–400-line target.
  • PubMed verification: re-fetched all 301 pre-existing unique PMIDs with live PubMed E-utilities and fetched three newly relevant records exposed by refreshed gap searches, for 304/304 resolving records. Reconciled 983 non-bibliography claim→PMID links across 728 claim lines and 1,773 total PMID occurrences when page references, bibliography entries and this audit log are included. Author, title, journal and year matched the live records; ePub/print offsets were tolerated.
  • ClinicalTrials.gov verification: re-fetched all 31 unique cited NCT records (36 occurrences) through API v2. Every identifier resolved and every displayed recruitment status matched the 2026-08-30 live record; no status correction was required.
  • Non-journal verification: fetched all 15 unique external URLs cited in the condition; 15/15 returned HTTP 200 after redirects. Updated organization and guideline access dates to 2026-08-30.
  • Errors found and fixed: 11 substantive issue clusters affecting 37 claim or table locations. Corrected GBD 2019 CRC burden from 1.2 million cases/572,000 deaths to 2.17 million (95% UI 2.00–2.34)/1.09 million (95% UI 1.00–1.15); removed unsupported rounded GLOBOCAN percentage increases; corrected NordICC usual-care incidence from 1.22% to 1.20%; corrected the Pieniowski cohort from 41% major/35% no LARS to 53.1%/22.6%; aligned RAPIDO recurrence reporting to 44/431 (10%) versus 26/428 (6%); removed MOUNTAINEER duration/PFS numbers absent from the cited PubMed abstract while retaining response 38.1% (95% CI 27.7–49.3); removed an unsupported 30–60% lung-metastasis survival range; replaced unstated transplant-score thresholds with abstract-reported strata and estimates; and softened overbroad CMS utility language.
  • Refreshed evidence-gap searches: re-ran targeted PubMed searches for ablation versus resection, CMS-guided therapy, bevacizumab expression selectors, blood-screening mortality, screening below age 50, field-effect surveillance, pMMR/MSS immunotherapy conversion, post-metastasectomy ctDNA and cascade-testing interventions. The former “no randomized ablation comparison” statement was stale: a 2026 synthesis includes one randomized trial (PMID 42463528). The former “no comparative direct-contact evidence” statement was also stale: a 2025 Swedish randomized trial found 71% versus 67% counselling uptake, adjusted OR 1.24 (95% CI 0.79–1.95; PMID 40745491). PanaMa retrospective CMS analysis within a randomized trial (PMID 37018649) was added while preserving the dated gap in prospective CMS-directed treatment assignment.
  • Bibliography and closure checks: bibliography now contains 304 unique PMIDs, exactly covering the condition's PMID set with no orphan or missing entry. Every wiki body PMID appears in that page's References section and every page reference is used in the body. All 18 page link targets are canonical and present.
  • Remaining flags: zero unresolved verification markers, zero unresolved PMID/NCT identifiers, zero unresolved citation-reference mismatches and zero unresolved substantive audit issues. Methodological limitations remain: this is not a registered systematic review, no dual-review risk-of-bias adjudication was performed, and patient-voice coverage remains English-language/high-income-country weighted.
  • Status changes: promoted all 18 wiki pages from draft to curated, updated INDEX.md to audited, and changed the colorectal adenocarcinoma row in CONDITIONS-ROADMAP.md from built to audited.

2026-10-12 — Periodic literature sweep (pending independent audit)

  • Window and searches: PubMed E-utilities searched publications dated 2026-08-28 through 2026-10-12 across 15 topics: burden/young onset, carcinogenesis/serrated lesions, molecular subtypes, hereditary genetics, screening, pathology/staging, localized colon, localized rectal, metastatic systemic therapy, precision/immunotherapy, liver/lung metastases, ctDNA/biomarkers, survivorship/patient experience, complications/safety, and guidelines. Each topic received a separate "Cochrane Database Syst Rev"[jour] AND <topic> PubMed search with the same publication-date filter. Initial HTTP 429 responses for the serrated Cochrane and ctDNA searches were retried successfully. Broad pathology search returned 1,019 hits but only the first 1,000 IDs were harvested; targeted follow-up searches covered prioritized trial/guideline questions. The Cochrane search surfaced the rectal-surgery transanal-tube review (PMID 42770376); no CENTRAL search was performed.
  • Beyond PubMed: Europe PMC REST search was queried for all 15 topics with FIRST_PDATE:[2026-08-28 TO 2026-10-12], including non-MEDLINE and full-text-availability fields. For large result sets, only the first 1,000 records were retrieved, so this is a sweep rather than exhaustive screening. Europe PMC full-text XML was opened for GLOBOCAN 2024 (PMID 42417444) and LAVA-CRLM (PMID 42829779); their Methods/Results and Table 1/Table 2 respectively supplied the numbers added below. Crossref works/{doi} updated-by was checked for all 11 bibliography additions, alongside PubMed publication types; none had a retraction or expression-of-concern flag. All 315 bibliography PMIDs resolved in a fresh live PubMed E-utilities batch. No record without a PMID was cited, so no DOI-only citation was added. Unpaywall was not queried. Cochrane Library/CENTRAL and SciELO/LILACS were not searched; their HTTP-403 access limitation remains, so no global or regional-literature absence is claimed.
  • Trial registries: ClinicalTrials.gov API v2 was queried live for all 31 NCT IDs cited in the trial-landscape page; only NCT02864485 changed displayed recruitment status, from recruiting to active, not recruiting (registry update posted 2026-10-02). No new results posting was found among the 31 tracked records. The API was also queried for the new EPoS II, ASAC and LAVA-CRLM study registrations before their trial context was considered. The EU Clinical Trials Register was queried for the Scandinavian ASAC study (EudraCT 2014-003601-15; completed Danish protocol, no results posted there) and by EPoS II and LAVA-CRLM names; the EPoS term returned an unrelated trial, so it was not treated as confirmation. WHO ICTRP portal query returned HTTP 403. Trial status in this condition is therefore grounded in the accessible named/API records, not an inference from the unavailable registry.
  • Guideline, burden and patient sources: Checked the American Cancer Society's 2026 screening recommendation page, ACG guideline index, ESMO guideline news/living-guideline listing, IARC's GLOBOCAN 2024 article and data portal, Fight Colorectal Cancer's September 2026 coalition update, and Colostomy UK's 2026 Stomas Unlocked campaign. These are society/organization primary pages fetched this session. The 2026 ACS screening update (PMID 42200680), ACG adenomatous-polyposis guideline (PMID 42683623), US Multi-Society Task Force serrated-polyposis consensus (PMID 42814045), ESMO metastatic guideline (PMID 41990853) and focused exercise update (PMID 41741110) were added to the guideline registry, with supersession limited to their scopes. GLOBOCAN 2024 reports 2,041,007 incident cases and 917,895 deaths in a colorectum grouping that includes anal cancer; it is not an adenocarcinoma-only count (Sung 2026, PMID 42417444, Table 1 and Methods).
  • Triage and changes: Added 11 records to literature/BIBLIOGRAPHY.md, reaching 315 unique PMIDs. Updated wiki/epidemiology-and-burden.md and literature/statistics/STATISTICS.md with the qualified 2024 burden; wiki/guidelines.md, wiki/screening-and-early-detection.md and literature/guidelines/REGISTRY.md with current guideline lineage and the ACS blood-test hierarchy; wiki/liver-and-lung-metastasis.md and statistics with the full-text LAVA-CRLM phase II comparison; wiki/clinical-trials-landscape.md with the recruitment change; OPEN-QUESTIONS.md (OQ-15 and OQ-25); and literature/patient-voice/sources.md with two public campaign pages. INDEX.md reflects 13 curated and 5 draft wiki pages, 33 guideline documents and 115 statistics rows. No deep note was warranted. EPoS II (PMID 42748427), ASAC (PMID 42777749), the Cochrane review (PMID 42770376) and a ctDNA synthesis (PMID 42745098) were catalogued but did not change a wiki recommendation because their full reports could not be checked through the available full-text route.
  • Verification and next sweep: New quantitative claims were checked against accessible original full text and tables; new guideline positioning was checked against the societies' own pages. All 31 tracked NCT IDs resolved, all 315 bibliography PMIDs resolved, and no new paper was retraction-flagged in PubMed or Crossref. The edited five wiki pages remain status: draft for a different engine's audit. Next: obtain the full EPoS II and ASAC reports and examine their outcome tables and applicability before changing surveillance or post-metastasis-treatment conclusions; read the full 2026 ACG/USMSTF and ESMO texts for exact recommendation wording and version changes, and verify current NCCN colon/rectal versions from the society portal (the public search yielded no version record this session); review the Cochrane transanal-tube full text if access opens; recheck NCT02864485 recruitment and any results postings; and reassess regional transport only when SciELO/LILACS or another suitable regional source is reachable.

2026-10-19 — Periodic literature sweep (author: Grok, grok-4.6)

  • Window and searches: PubMed E-utilities esearch/esummary/efetch for publications dated 2026-08-28 onward across 15 topics (burden/young-onset, carcinogenesis/serrated, molecular subtypes, hereditary, screening, pathology/staging, localized colon, localized rectal, metastatic systemic, precision/immuno, liver/lung mets, ctDNA/biomarkers, survivorship, safety, guidelines), plus Entrez-date catch from 2026-10-12, RCT/systematic-review/guideline/high-impact filters, and a retraction/EoC window search. Each topic also received "Cochrane Database Syst Rev"[jour] AND <topic> with the same publication-date filter. Query strings, translations, counts and harvested IDs are in literature/sweeps/2026-10-19.json. PubMed MCP tools were not available; E-utilities were the live equivalent.
  • Beyond PubMed: Europe PMC REST search was queried (Cochrane journal slice, preprint SRC:PPR, high-signal RCT/guideline slice, and a non-PMID SRC:CBA/AGR/HIR slice). Preprints without PMIDs were not cited. Crossref works/{doi} updated-by was checked for every record added to the bibliography. Unpaywall was skipped (LIFEBANK_CONTACT_EMAIL unset). Cochrane Library/CENTRAL and SciELO/LILACS remain HTTP 403; no global or regional-literature absence is claimed. WHO ICTRP portal returned HTTP 200 but is JS-driven, so it was not counted as searched. EU CTR was queried for EudraCT 2014-003601-15 (ASAC). ISRCTN HTML for STAR-TREC did not expose a structured status field. ChiCTR search HTML was JS. ASCO guideline listing and Bowel Cancer Australia returned HTTP 403; SEER colorectal statfacts timed out.
  • Cochrane: The only CRC-titled CDSR hit in the window was the transanal-tube review already catalogued (PMID 42770376). Other CDSR IDs matching topic extra-terms were off-topic (sleep apnoea, dental implants, preterm care, etc.). Europe PMC Cochrane journal slice for colorectal/colon/rectal in the window: 0 hits. CENTRAL was not searched.
  • Retraction: Three CRC-window records were Expressions of Concern (PMIDs 42825377, 42844080, 42865666); none is in this bibliography. STELLAR 5-year update (PMID 42507974) has a Crossref correction (DOI 10.1200/jco-26-02343, PMID 42789827); wiki records the update’s existence and does not restate the 5-year OS figures. CADLY2 (PMID 42462747) has a Crossref erratum and was not added to wiki pages.
  • Trial registries: All 31 previously tracked NCT IDs resolved on ClinicalTrials.gov v2; none changed displayed recruitment status after 2026-10-12. NCT02864485 remains active, not recruiting (last update 2026-10-02). Newly used identifiers: NCT03177382 (TNTCRT, unknown), NCT02997228 (COMMIT, active not recruiting), NCT04094688 (SOLARIS, active not recruiting, results posted 2026-10-14), NCT03326791 (ASAC, completed), NCT02319928 (EPoS II, active not recruiting), ISRCTN14240288 (STAR-TREC).
  • Guidelines, burden, patient sources: ACS screening page, USPSTF 2021 statement, ACG guideline index, ESMO news/mCRC pages, NCCN colon and rectal public pages, NICE NG151/history, IARC GCO, Fight CRC, Colorectal Cancer Alliance, Bowel Cancer UK and Colostomy UK were fetched. NCCN Colon Cancer and Rectal Cancer public pages displayed Version 2.2026 (the 2026-10-12 sweep had no version record). USPSTF remains 2021. NICE NG151 history last lists 15 December 2021. Colostomy UK still lists the 2026 Stomas Unlocked campaign already catalogued; no new patient-voice theme was added. GLOBOCAN 2024 figures were not replaced.
  • Triage and changes: Added 16 records to literature/BIBLIOGRAPHY.md (315 → 331 unique PMIDs). Wiki updates: STAR-TREC 12-month organ-preservation and TNTCRT DFS (localized-rectal-cancer.md); SOLARIS negative vitamin D3 and COMMIT dMMR combination (metastatic-systemic-therapy.md, precision-oncology.md); CaPP3 aspirin dose (hereditary-syndromes-and-genetics.md); EPoS II interim, Ramai uptake NMA and DENEB accuracy (screening-and-early-detection.md); ASAC negative aspirin and GALAXY observational ctDNA–ACT association (liver-and-lung-metastasis.md, biomarkers-and-liquid-biopsy.md); UK UGT1A1 testing guideline (red-flags-and-safety-concerns.md); NCCN Version 2.2026 (guidelines.md, literature/guidelines/REGISTRY.md now 38 documents); trial landscape date-stamp and published studies. OPEN-QUESTIONS.md sharpened OQ-3, OQ-4, OQ-10, OQ-14, OQ-15, OQ-17, OQ-20, OQ-25. literature/statistics/STATISTICS.md now 131 rows. No deep note: STAR-TREC is a 12-month interim, TNTCRT is one more TNT comparison, CaPP3 did not formally conclude non-inferiority. Catalogued without wiki recommendation change: Mannucci germline panel yield (PMID 42692037), Soriano MSH3/MLH3 heterozygotes (PMID 42728028), Kim NAA surveillance (PMID 42142544), MASLD–CRC Delphi and Chinese ICI-LARC consensus (registry only), Cochrane transanal-tube (still no OA full text).
  • Verification depth: SOLARIS numbers were checked against the PMC author manuscript (PMC13434973), including the HR 0.92 and one-sided P = 0.25. STAR-TREC, TNTCRT, COMMIT, CaPP3, EPoS II, ASAC, Ramai, DENEB and Kataoka numbers come from structured abstracts retrieved live this session; their paywalled bodies were not opened (COMMIT Europe PMC XML returned HTTP 500). NCCN version identity was read from the society HTML. Edited wiki pages are status: draft.
  • Next sweep should chase: STELLAR 5-year erratum tables (PMID 42789827) before restating OS; CADLY2 erratum (PMID 42462747); STAR-TREC 30-month primary; EPoS II 10-year primary and ASAC/COMMIT/TNTCRT full-text tables; Cochrane transanal-tube full text if Wiley access opens; NCCN 2.2026 recommendation text if a public Insights article appears; ACG/USMSTF/ESMO full texts still paywalled; NCT02864485 results; SciELO/LILACS if the 403 lifts. Auditor must be a different engine.

2026-10-26 — Periodic literature sweep (author: Claude, claude-opus-5-5; pending audit by a different engine)

  • Window and searches: PubMed E-utilities (the PubMed MCP tools were disconnected this session; E-utilities used as the live equivalent) re-ran the 15 topic queries from the 2026-10-19 manifest with publication date ≥ 2026-08-28, plus RCT, systematic-review, guideline, high-impact-journal and retraction/EoC slices, plus an Entrez-date catch from 2026-10-19 (279 hits). 8,156 IDs were harvested in total. Triage was restricted to records neither harvested by the 2026-10-19 sweep nor already in the bibliography: 40 in the high-signal slices, read by title, type and abstract, and 1,403 from the topic searches, screened by title for trial, guideline, registry or burden terms. Each of the 15 topics also got a "Cochrane Database Syst Rev"[jour] AND <topic> search over the same window, plus one broad CRC search. The only CDSR hit was the transanal-tube review that is already catalogued (PMID 42770376); no new or updated Cochrane review covers any topic of this condition. Queries, translations and IDs are in literature/sweeps/2026-10-26.json.
  • Beyond PubMed: Europe PMC REST was queried for full-text availability for all 16 candidates, for the Cochrane journal (0 hits), for non-MEDLINE sources SRC:CBA/AGR/HIR/CTX/ETH (0), for preprints (189, all without PMIDs, none cited), for non-MEDLINE trial records (62, none practice-relevant enough to cite by DOI) and for a high-signal MEDLINE slice. For each of the 14 additions, Crossref updated-by returned no update (HTTP 200, no mailto sent) and no PubMed retraction or EoC pubtype was found. All 331 existing bibliography PMIDs were re-fetched and none is now retraction- or EoC-flagged. The window's three EoC records (PMIDs 42825377, 42844080, 42865666) are the same as on 2026-10-19 and none is in the bibliography. Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL is unset. Cochrane Library/CENTRAL and SciELO/LILACS were not searched (HTTP 403 per SOURCES.md), so no absence claim beyond PubMed and Europe PMC is made. WHO ICTRP was not used. The Japanese registry record for PRABITAS (jRCTs041230120) was fetched directly: Not Recruiting, last modified 2026-10-01.
  • Trial registries: ClinicalTrials.gov v2 was queried live for all 36 NCT IDs the condition cited, plus 3 newly cited (NCT02967289 IROCAS, completed; NCT04928807 UNION, active not recruiting; NCT03026140 NICHE, recruiting). One status changed: COLT (NCT03803436) went from unknown to completed (primary completion 2025-10-31, registry update 2026-10-21, no results posted). No other tracked record changed status or posted new results. NCT04008030 (CheckMate 8HW) already had results posted (2025-09-22).
  • Guidelines, statistics, patient sources: The NCCN colon and rectal pages still show Version 2.2026. USPSTF is still the 18 May 2021 statement. The German S3 colorectal guideline was published in Z Gastroenterol (Pox 2026, PMID 42906536; methods report Klug 2026, PMID 42906535). The Leitlinienprogramm Onkologie portal lists Version 3.2 (2026) as current (3.0 Sept 2025, 3.1 Feb 2026). It was added as registry row 39 with no recommendation content, because neither record has an abstract and the full text was not opened. The ESMO guideline index URL returned 404, the ACS page 406 and Fight CRC 403. The Colorectal Cancer Alliance's 1 September 2026 hub on unrequested mailed stool-test kits (built with the test's manufacturer, conflict noted) was added to patient-voice/sources.md. No new patient-voice theme was added.
  • Triage and changes: 14 records were added to literature/BIBLIOGRAPHY.md (331 → 345). Wiki edits:
  • localized-colon-cancer.md (demoted curated → draft): IROCAS was negative for adjuvant mFOLFIRINOX intensification (PMID 42927853). NICHE-2 is now covered: 2024 pathological response (PMID 38838311) and 2026 3-year DFS 100% (PMID 42906564), with a caveat table. Before this sweep the page did not mention neoadjuvant immunotherapy for dMMR colon cancer at all, which was a real gap.
  • localized-rectal-cancer.md: UNION interim EFS (PMID 42927765) with a limitations table; AZUR-1 interim cCR12 68% (PMID 42927858), stated explicitly as lower than the 12/12 pilot.
  • metastatic-systemic-therapy.md: CheckMate 8HW final OS (PMID 42926106); PRABITAS 14-day TAS-102–bevacizumab non-inferiority (PMID 42925880).
  • precision-oncology.md: CheckMate 8HW final OS.
  • clinical-trials-landscape.md: new localized-colon trial table, UNION row, AZUR-1/8HW annotations, COLT status.

Other files: literature/statistics/STATISTICS.md gained 12 rows (131 → 143); literature/guidelines/REGISTRY.md gained row 39, an annotated disagreement #11, an NCCN re-check and the watch list; OPEN-QUESTIONS.md sharpened OQ-4 (UNION as the first phase III pMMR-inclusive PD-1 signal, MMR-stratified data not seen), OQ-7 (mailed-FIT effect decaying by year 3, PMID 42905707), OQ-9 (first-line 8HW OS vs nivolumab missed threshold), OQ-19 (IROCAS) and OQ-20 (UNION), and opened OQ-31 (checkpoint blockade versus standard care in localized dMMR disease; duration and surveillance; Gao 2026 observational maintenance-vs-observation cohort, PMID 42901720). The following were catalogued without a wiki change: Lynch detection-rate meta-analysis (Poo 2026, PMID 42442403; ADR 23.3%, CRCDR 2.5%, higher in MLH1/MSH2), VA cluster-randomized CADe availability (Dominitz 2026, PMID 42250891; ADR +4.2 points, OR 1.22) and SEER early-onset trends through 2023 (Al Khaldi 2026, PMID 42691927; AAPC 8.3% for 2019–2023, now a statistics row). No deep note: CheckMate 8HW final is an update to an already-cited trial, and NICHE-2 and AZUR-1 are single-arm. If a randomized dMMR neoadjuvant trial reports, NICHE-2 should be revisited as a note candidate. - Verification depth, stated plainly: Europe PMC reports no free full text for CheckMate 8HW final, IROCAS, AZUR-1, NICHE-2 2026, UNION, PRABITAS or the German S3, so every new number comes from the structured PubMed abstract, not from tables. Each such claim says so inline. Under CLAUDE.md rule 5, none of these figures has been checked against the paper's own table. The auditor should do that before any of these pages is promoted. Closure checks: every PMID in the condition is in the bibliography, every body PMID on the edited pages is in that page's References, and every NCT ID in the condition was verified live this session. One problem predates this sweep and was not touched: wiki/guidelines.md lists PMID 29846947 in References without citing it in the body. - Next sweep should chase: full texts or tables for UNION (MMR-stratified EFS), AZUR-1 (central-review cCR12), CheckMate 8HW final (first-line subgroup OS) and IROCAS (OS maturity); recommendation content of German S3 v3.2; any guideline adopting neoadjuvant immunotherapy for dMMR colon cancer; COLT results; STELLAR erratum tables (PMID 42789827) and CADLY2 erratum, still unread; STAR-TREC 30-month primary; NCT02864485 results; ESMO's new guideline index URL (old one 404); SciELO/LILACS if the 403 lifts. The auditor must be a different engine from Claude.