Not yet independently audited

This page was written from verified sources but has not yet passed the independent citation audit. Treat its claims as provisional.

Curation log — vascular dementia¶

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

2026-10-25 — Periodic literature sweep (Grok author; independent different-engine audit pending)¶

Window and method. Inclusive 2026-10-18 → 2026-10-25 publication-date searches, plus an Entrez-entry-date supplement. Read CONVENTIONS, CLAUDE, SOURCES, INDEX, the prior log and open questions before editing. Search manifest: literature/sweeps/2026-10-25.json. No Git, publication, deployment or other-condition changes were performed.

Sources actually queried. PubMed E-utilities: 14 topics covering VCI/diagnosis, SVD, CAA, inherited disease, mixed pathology, biomarkers, prevention, treatment, imaging, guidelines, burden, patient experience, post-stroke assessment and mechanisms, each by publication and entry date (102 unique PMIDs combined). Per-topic CDSR queries returned 0. A journal-only CDSR window query returned 5 records, all off-topic (ADHD/ASD smoking cessation, ERAS colorectal, adrenaline storage, pelvic inflammatory disease, T-cell lymphoma transplant). Europe PMC REST: the same 14 topics using FIRST_PDATE (199 unique source/ID pairs; 13 unique preprints). ClinicalTrials.gov v2: all 53 previously present NCT IDs, plus eight last-update topic queries. ISRCTN documented XML API: dated VCI/SVD/CAA/CADASIL lastEdited query, plus individual LACI-3 and LACI-Cog1 records. Crossref updated-by on the article DOI of every added journal record, without mailto. Official web checks: NICE NG97, Canadian VCI, ESO directory, WHO fact sheet/second-edition executive summary/GDO; CureCADASIL, Alzheimer's Society news, Alzheimer's Association news, ADI news, Heart & Stroke news, American Stroke Association, Stroke Association UK, International CAA Association.

Cochrane result. No completed new VCI, SVD, CAA, CADASIL, post-stroke cognition or blood-pressure CDSR review in this window. This is a PubMed CDSR finding; CENTRAL was not searched (cochranelibrary.com HTTP 403).

What survived triage. Seven journal records and one newly cited NCT. No landmark deep note.

Record	Decision
Markus 2026, PMID 42894468	IJS treatment review; restates Western-null for pure VaD from full text (PMC13680733). Treatment page
van der Lelij 2026, PMID 42895925	Meta-VCI-Map n=3,481; tract-specific WMH > total volume. Neuroimaging, subtypes, OQ-7/D25. Abstract
van Dijk 2026, PMID 42898208	Fazekas vs volumetry, n=246 plus SR. Neuroimaging, OQ-7. Full text (PMC13677418)
Kuo 2026, PMID 42907303	CADASIL vs sporadic SVD NPS, n=245. Cognitive profile, monogenic. Abstract
Yang 2026, PMID 42912759	Post-stroke exercise 19 RCTs, SMD 0.25. Treatment, OQ-8. Abstract; not VCI-specific
Liaw 2026, PMID 42910061	NOTCH3 p.R544C Taiwan Biobank, stroke OR 1.54. Monogenic. Stroke association only
Cao 2026, PMID 42858034	OCTA in carotid stenosis, n=860. Bibliography only; not a D9 trial-mediation test
NCT06393712	ALN-APP/mivelsiran phase 2 in CAA, n=210 actual, active not recruiting. Trials page

Dropped/deferred. Animal/TCM mechanisms, acupuncture ML, case reports, ARIA 3-patient PET series, memantine nationwide analysis (not VaD-stratified), choline alfoscerate 3-trial SR in mixed MCI, generic dementia caregiver guideline, Kano Mini-Cog outpatient screen (n=75), air-pollution/walkability CSVD (weak), genomic-SEM SVD factor, ALPS/heart-dysfunction cross-sectional mediation, simplified Fazekas methods paper. Preprints were not cited. HOPE (PMID 42887919) and Vella tables remain unretrieved from last week.

Trials. All 53 cited NCT records resolved. No status, last-update-date or results-availability change inside this window. GBE50 still active, not recruiting, estimated primary completion 2026-10-30, no results. COMPEX still active, not recruiting, primary completion 2026-11-05, no results. LACI-3 and LACI-Cog1 still recruiting. Dated ISRCTN lastEdited query returned one off-topic KarX formulation study. Added NCT06393712 (ALN-APP in CAA), last updated 2026-10-20, primary outcome annualized new lobar microbleeds, not cognition.

Guidelines, burden and patient material. NICE NG97 still records the 24 October 2025 decision not to update. WHO second-edition executive summary page remains current; fact-sheet/GDO checks yielded no new VaD-specific burden series. No newly dated condition-specific campaign survived the checked patient/news listings. Stroke Association UK remains 403. The previously listed MGH CAA treatment URL redirected to the hospital homepage; that is URL rot, not a new campaign, and the curated patient-voice files were left untouched. ADI and ASA returned 200.

Retractions and evidence boundary. All seven added journal records have no PubMed Retracted Publication/Expression of Concern flag and empty Crossref updated-by on the article DOI. Markus and van Dijk numbers came from retrieved full text. Kuo, van der Lelij, Yang and Liaw rest on abstracts. Unpaywall skipped because LIFEBANK_CONTACT_EMAIL is unset.

Not searched / access limits. Cochrane Library/CENTRAL and SciELO/LILACS remain unavailable (HTTP 403); they were not claimed as searched. WHO ICTRP and ChiCTR homepages returned HTTP 200 but did not yield extractable search results (ICTRP is JS-driven; ChiCTR served compressed non-text). These are failed or non-interpretable checks, not completed negative registry searches. OpenAlex/DOAJ and direct preprint-server searches were not run. No exhaustive regional or worldwide trial absence is asserted.

Files and status. Edited six wiki pages (treatment, neuroimaging, cognitive-profile-and-assessment, monogenic-and-inherited-forms, subtypes-and-clinical-syndromes, clinical-trials-landscape), BIBLIOGRAPHY, STATISTICS, OPEN-QUESTIONS, INDEX NCT count, this log and the dated manifest. Edited content pages have last-curated 2026-10-25 and draft status. Pages that were curated (cognitive-profile-and-assessment, monogenic-and-inherited-forms, subtypes-and-clinical-syndromes) were demoted because this author does not audit its own work. No independent audit or curated promotion was performed.

Next sweep. Chase GBE50 results after 2026-10-30 and COMPEX after 2026-11. Monitor LACI-3, LACI-Cog1, GiCIIS and ALN-APP (NCT06393712). Retrieve HOPE's report, Vella's tables, Alten full text, and WHO second-edition stroke/hypertension/multidomain tables. Recheck ICTRP/ChiCTR for usable APIs. Joint A4 PVS/ARIA/spontaneous-ICH model and D11–D27 still need dedicated follow-up. Repair the redirected MGH CAA URL at the next patient-voice re-fetch. Independent audit must be by a different engine.

2026-10-18 — Periodic literature sweep (Codex author; independent different-engine audit pending)¶

Window and method. Inclusive 2026-10-11 → 2026-10-18 publication-date searches, plus an Entrez-entry-date supplement to catch late indexing. Read CONVENTIONS, CLAUDE, SOURCES, INDEX, the prior log and open questions before editing. Search manifest: literature/sweeps/2026-10-18.json (the canonical published path). No Git, publication, deployment or other-condition changes were performed.

Sources actually queried. PubMed E-utilities: 14 topics covering VCI/diagnosis, SVD, CAA, inherited disease, mixed pathology, biomarkers, prevention, treatment, imaging, guidelines, burden, patient experience, post-stroke assessment and mechanisms. Each topic was searched by publication and entry date: 83 and 146 unique records respectively, 164 combined. Each topic also had a dated CDSR query; broader CDSR searches prevented cognitive-outcome filters from hiding protocols. Europe PMC REST: the same 14 topics using FIRST_PDATE, 255 unique source/ID pairs, with full-text availability fields retained. PubMed efetch verified the selected papers and additional identifiers retained on edited lines. Crossref title/DOI metadata and updated-by checked for every added journal record and the WHO executive summary, without mailto.

ClinicalTrials.gov v2: all 52 NCT IDs already present in condition Markdown (including log-only IDs), plus eight dated last-update topic queries returning 71 distinct IDs. ISRCTN's documented XML API: dated VCI/SVD/CAA/CADASIL search, then undated “vascular cognitive” and “small vessel” searches, with individual LACI-3 and LACI-Cog1 records checked. WHO ICTRP attempted (403); ChiCTR GBE50/butylphthalide and GiCIIS cross-registration searches attempted (405). These are failed access checks, not completed negative registry searches. Official web checks: NICE NG97, Canadian VCI, ESO directory, International CAA Association, WHO second edition/executive summary/fact sheet/GDO; CureCADASIL, Alzheimer's Society news, Alzheimer's Association news, ADI news, Heart & Stroke news, American Stroke Association, the previously listed CAA URL and MGH CAA Research Program.

Cochrane result. The broad dated publication query returned three records: a BP-target protocol (PMID 42861682), an anti-amyloid editorial letter (PMID 42878653), and an off-topic neonatal home-visiting review (PMID 42861678). None is a completed new VCI intervention review. No reviewed recommendation was changed on their basis. This is a PubMed CDSR finding; CENTRAL was not searched.

What survived triage. Seven journal records and one WHO executive summary were catalogued; one journal record and the WHO document are older follow-ups/coverage corrections as indicated. No landmark deep note.

Record	Decision
Tang 2026, PMID 42886840	Post-stroke synthesis; treatment page, statistics and OQ-8. Publisher full text/Table 2 retrieved through live web search; direct fetch remained blocked
Luzum 2026, PMID 42865713	Nor-COAST within-person/between-person distinction; treatment page, full text checked
Chagnot 2026, PMID 42867637	Human vascular phenotypes; mechanism page and OQ-12, full text checked
Rosenau 2026, PMID 42863385	Prevention review, bibliography only; full text not reached
Rivier 2026, PMID 42885481	CTSB target nomination, bibliography only; abstract, no treatment inference
Vella 2026, PMID 42862582	All-cause dementia incidence, bibliography only; full-text tables unavailable, no rates imported
Jaime Garcia 2025, PMID 40275856	Older D4 follow-up: full text reports baseline BBB associations; baseline-only serology does not validate a longitudinal signature. Mechanism page/OQ-12
WHO 2026, doi:10.2471/b09867	Executive-summary DOI resolved; second edition replaces 2019 in the guideline catalog. Pre-window correction, not this week's new guidance

Dropped/deferred. Animal/TCM mechanisms, network pharmacology, small cross-sectional biomarker/ML pilots, case reports, non-VCI material and preprints did not justify additions. HOPE (PMID 42887919) has no PubMed abstract and its publisher returned 403; no result or efficacy claim was extracted. Other updated registry hits mostly concerned motor rehabilitation, non-vascular dementia or unrelated disease; none supplied a retained VCI result.

Trials and coverage corrections. All 52 cited NCT records resolved. The 45 with saved previous-sweep baselines showed no status, update-date or results-availability change. GBE50 (NCT06495476), NHLBI CADASIL (NCT05072483) and COMPEX (NCT04229056) also retain the statuses tabulated last week and have no posted results. Two older ISRCTN studies were missing: LACI-3, ISRCTN44436843, recruiting, phase III, planned n=1,300, ISMN/cilostazol with cognition primary at 18 months; and LACI-Cog1, ISRCTN84539143, recruiting, phase II, planned n=60, feasibility/tolerance primary. The 2026-10-11 → 2026-10-18 ISRCTN update query itself returned no matches. Added GiCIIS, NCT07220538, updated 2026-10-13: phase IV, planned n=400, recruiting, open-label ginkgo add-on after acute ischaemic stroke, multiple cognitive and clinical primary outcomes, no results. “Sham comparator” in its registry arm label does not mean placebo masking; the described control is standard care. Trial page, treatment and OQ-3 now distinguish LACI-3's cognitive primary from IMPACT's disability primary.

Guidelines, burden and patient material. WHO's official second-edition page is dated 2026-07-15; its executive-summary Crossref record is dated 2026-08-24. The registry's older “current WHO 2019” designation and “await update” watch item were corrected. Recommendation grades and the stroke/hypertension evidence tables remain unextracted; the old blanket claim that no catalogued guideline incorporates newer BP evidence was removed. ESO lists cognitive rehabilitation after stroke as in development. NICE still records its October 2025 decision not to update. WHO's fact-sheet/GDO checks yielded no new VaD-specific burden series; they are not a global registry census. No newly dated condition-specific campaign survived the checked patient/news listings. ADI news is accessible again (200); ASA remains 403. The listed thecaa.org is an actors' club, not a verified CAA alliance: quarantined it, added the fetched institutional MGH CAA resource, and installed an explicit title/identity/disease-scope check in organizations.md.

Retractions and evidence boundary. All seven added journal records have no PubMed Retracted Publication/Expression of Concern flag and empty Crossref updated-by; WHO's monograph metadata also has no update flag. Numerical synthesis data came from the retrieved table, not an abstract. The Chagnot, Luzum and older Jaime Garcia full texts were opened; the latter resolves the earlier abstract-only uncertainty without restoring D4 as an absence. A title-only Jaime Garcia PubMed search returned zero despite the exact-ID efetch resolving the paper; it was not interpreted as absence.

Not searched / access limits. Cochrane Library/CENTRAL and SciELO/LILACS remain unavailable (HTTP 403 per SOURCES); they were not claimed as searched. Regional questions OQ-21/OQ-34 now carry this explicit boundary. Unpaywall skipped because LIFEBANK_CONTACT_EMAIL is unset. No owner contact address was used. OpenAlex/DOAJ and direct preprint-server searches were not run; preprints were encountered only in Europe PMC. Subscription databases were not accessed. No exhaustive regional or worldwide trial absence is asserted.

Files and status. Edited four wiki pages (clinical-trials-landscape, treatment, pathophysiology, guidelines), BIBLIOGRAPHY, guideline REGISTRY, STATISTICS, OPEN-QUESTIONS, patient-voice organizations/sources, this log and the dated manifest. All edited content pages have last-curated 2026-10-18 and draft status. No independent audit or curated promotion was performed.

Verification. Targeted local checks passed: all newly written citation lines resolve against the live E-utilities/ClinicalTrials.gov retrievals, seven journal additions have both retraction checks, wiki references are sequential, new local links resolve, and all eleven edited Markdown files carry draft status and last-curated 2026-10-18. The session fetched 177 distinct PubMed records in total, including follow-ups and citation-line checks; this is not the sweep yield. No repository-wide site generation or strict build was run, because publishing is handled downstream.

Next sweep. Chase GBE50 results after 2026-10-30 and COMPEX after 2026-11; monitor LACI-3, LACI-Cog1 and GiCIIS through both registries. Retrieve HOPE's report before judging its intervention, Vella's tables before importing burden estimates, and the WHO second-edition stroke/hypertension/multidomain tables before synthesizing grades. Alten full text remains unavailable: Europe PMC reports no full text and the publisher returned 403. The joint A4 PVS/ARIA/spontaneous-ICH model and D11–D27 still need dedicated follow-up; no new absence claim was substituted for that work. Recheck ICTRP/ChiCTR access and retain organization identity checks. Independent audit must be by a different engine.

2026-10-11 — Periodic literature sweep (Claude author, claude-opus-5-5; independent audit pending)¶

Window. Records entered in PubMed from 2026-10-07 to 2026-10-11 (Entrez date). The broad query was also run by publication date. This is the condition's first periodic sweep; the 2026-10-07 entries below were grounding and audit passes, not sweeps. All PubMed retrieval was live E-utilities. Query manifest: literature/sweeps/2026-10-11.json.

Sources actually queried. PubMed: 13 topic queries (broad VaD/VCI, SVD, CAA, monogenic, mixed pathology, biomarkers, BP/AF prevention, treatment, imaging, guidelines, epidemiology/burden, patient experience, CDSR), giving 61 unique records by Entrez date. Europe PMC REST: FIRST_PDATE:[2026-10-07 TO 2026-10-11], 33 hits. The lite results omit PMIDs for PMC-source records even where a PMID exists, so the count is not an absence denominator. Europe PMC full-text XML for two papers. ClinicalTrials.gov v2: all 45 previously cited NCT IDs re-fetched, plus seven condition queries filtered to first-posted or last-updated on or after 2026-10-07. Crossref updated-by for every added DOI, queried without mailto. NICE NG97 page, WHO dementia fact sheet, and 8 patient-organisation homepages.

Not queried. Cochrane Library/CENTRAL and SciELO/LILACS (HTTP 403, SOURCES.md). Unpaywall (LIFEBANK_CONTACT_EMAIL unset). WHO ICTRP and the named non-US registries were not queried this sweep, so the registry absence statements below cover ClinicalTrials.gov only. The bioRxiv/medRxiv servers were not queried directly; preprints were seen only as Europe PMC PPR records. No regional-literature claim is made.

Cochrane. "Cochrane Database Syst Rev"[jour] × dementia/cognition/stroke/SVD/CAA/BP/AF terms in the window: 0 records. The same query widened to 2026-09-02 → 2026-10-11 returned 7. Two were plausible: cardiorespiratory training after stroke (PMID 42742033) and BP targets in diabetes (PMID 42788405). Neither abstract mentions a cognitive or dementia outcome. No CDSR record supersedes or contradicts anything this condition asserts.

What survived triage. Seven records were added to the bibliography; four drove wiki edits. There is no landmark note, because nothing is practice-changing.

Record	Why it survived	Where
Rosbergen 2026, PMID 42846338	Rotterdam head-to-head of DTI markers, n=5,230, 286 dementias. PSMD/MD best; WMH C-statistic 0.876 not higher. Verified against PMC full text	neuroimaging.md, OQ-7, STATISTICS
Zhou 2026, PMID 42840269	DTI-ALPS lower in arteriolosclerosis than CAA. Tracks cognition only in arteriolosclerosis and does not track PVS burden in either. Verified against PMC full text	cerebral-small-vessel-disease.md, OQ-33
Alten 2026, PMID 42855813	A4: severe CSO-PVS → first lobar microbleed/siderosis, aHR 2.85 (1.91–4.25). Abstract only	cerebral-amyloid-angiopathy.md, OQ-33, D2, STATISTICS
Bigjahan 2026, PMID 42040755	A4: PVS volume fraction → ARIA, aHR 1.32–1.55. Published April 2026, outside the window; found while re-testing D2's absence	cerebral-amyloid-angiopathy.md, D2, STATISTICS
Liu & Raz 2026, PMID 42843461	WMH progression meta-analysis, 54 longitudinal studies, no pooled rate in the abstract	bibliography only
Khand 2026, PMID 42854618	CAA epilepsy → cognitive dysfunction HR 1.34, EHR/NLP cohort	bibliography, STATISTICS
Mah 2026, PMID 42842469	WMH moderates NAC+exercise apathy response, n=51 secondary analysis	bibliography only

Dropped. Jellinger narrative review of depression/anxiety in VaD (restates ranges). Gong review of VCI in China (narrative). Aryal SR (narrative; 15 studies). Tee 2026 stenosis/ASL (cross-sectional). Jansen CSF-NfL CAA-vs-AD pilot (n=30; LOOCV AUC 0.61). Duijs CAA microRNAs (discovery n=6). Han CAA proteomics (no ARIA outcome). Animal and TCM mechanism papers. Off-topic hits (ALS, glioma, moyamoya, DLB, PD, coronary consensus). Research Square preprints, including a post-stroke GFAP/IFN-α cohort: not peer reviewed, single cohort.

D2 absence re-tested and holds, with new neighbours. The 2026-10-07 claim was that no prospective model predicts both ICH and ARIA from Boston v2.0 burden. Re-query: ("amyloid-related imaging abnormalit*"[tiab] OR ARIA[tiab]) AND ("perivascular space*"[tiab] OR EPVS[tiab]) AND (microbleed*[tiab] OR siderosis[tiab] OR hemorrhag*[tiab] OR haemorrhag*[tiab]), no date limit, returned 6 records. (A4[tiab] OR "Anti-Amyloid Treatment in Asymptomatic"[tiab]) AND ("perivascular space*"[tiab]) returned 3. Neither query found a joint model. The two halves now exist separately in A4, using different PVS measures. D2's junction text was sharpened accordingly. These are PubMed-only absences.

ClinicalTrials.gov. All 45 cited NCT IDs resolved. One status change: NCT04229056 (COMPEX) RECRUITING → ACTIVE_NOT_RECRUITING, registry update 2026-10-02, primary completion now 2026-11-05. No cited record posted new results. Two records were added. NCT06495476 is GBE50 ginkgo versus placebo in SVD-associated MCI: phase 2/3, n=144, MoCA at 12 months, primary completion 2026-10-30. It is a cognitive-primary drug trial in an SVD population that both earlier sweeps missed, because its condition field says only "Mild Cognitive Impairment". NCT05072483 is the NHLBI CADASIL natural-history study (n=155, to 2034). Not tabulated: PhiCube upper-limb rehabilitation (NCT07865260), VR music therapy (NCT07010536), aquatic therapy (NCT07861009), Fabry agalsidase (NCT06052800) and aneurysm coiling (NCT06064435). None is VCI-specific.

Guidelines, statistics and patient voice. PubMed guideline query in the window: 3 records, none VaD/VCI (Italian anti-amyloid service consensus, a coronary consensus, dementia-living-alone recommendations). NICE NG97 page (HTTP 200): "Last reviewed: 24 October 2025 … will not update the recommendations". This is recorded in REGISTRY and resolves the watch-list item on the next NG97 revision. The WHO dementia fact sheet (published 2026-07-03) still gives 57 million (2021), so there is no new burden figure. Patient organisations: CureCADASIL, CAA Alliance, Alzheimer's Association, Alzheimer's Society and Heart & Stroke Canada returned 200. Alzheimer's Disease International and the American Stroke Association, both 200 at build, now return 403. Stroke Association UK is still 403. A 403 from a bot-blocked site is not evidence of closure, so organizations.md (curated) was not edited.

Retraction checks. All 7 added PMIDs: PubMed publication types show no Retracted Publication or Expression of Concern, and Crossref updated-by is empty for every DOI.

Pages edited and demoted curated → draft (last-curated 2026-10-11): neuroimaging.md, cerebral-small-vessel-disease.md, cerebral-amyloid-angiopathy.md, clinical-trials-landscape.md. Rosbergen and Zhou numbers were checked against full text, and Alten and Bigjahan against abstracts only. All four pages were demoted anyway, because the author does not audit its own work. Also edited: OPEN-QUESTIONS (OQ-7 and OQ-33 sharpened, D2 sharpened), BIBLIOGRAPHY (+7), STATISTICS (+1 section, 5 rows), guidelines REGISTRY (NG97 surveillance status), INDEX (junction count corrected 27 → 23, closing the 2026-10-07 audit's follow-up; NCT count updated).

Next sweep should chase. NCT06495476 results (primary completion 2026-10-30). COMPEX NCT04229056 results after 2026-11. Alten 2026 full text, to establish whether the solanezumab arm was pooled and how many participants were on drug. A joint A4 model of PVS → ARIA and → lobar haemorrhagic lesions. Whether WHO ICTRP/ChiCTR carries further GBE50 or butylphthalide SVD trials. ADI and ASA access (403). The D4 leakage-signature question against the Jaime Garcia full text, still open from 2026-10-07. D11–D27 remain unaudited. The auditor must be a different engine.

2026-10-07 — Independent audit of the dot-row grounding pass (Claude auditor; Grok author)¶

Scope. Grok's grounding of D1–D10 (entry below): 6 rewritten rows (D1, D2, D5, D6, D8, D9) and 4 deletions (D3, D4, D7, D10). D11–D27 already carried citations and were not audited. The table holds 23 rows after this audit. No row was deleted or restored by the auditor. The seeded wording of the deleted rows was read from logs/grounding-vascular-dementia.md.

Citation integrity¶

Every record was re-fetched live. All 29 PMIDs cited in the rewritten rows or named in the deletion reasons were fetched from PubMed E-utilities (esummary and efetch abstract) on 2026-10-07: 19139365, 31408137, 37840499, 41614465, 33716917, 35841910, 38710005, 36449413, 37459141, 42252508, 33704781, 42635820, 16700849, 33901427, 35349186, 33728742, 40682194, 38508839, 10097779, 19187414, 36300327, 42743445, 40966613, 35421194, 28095412, 12742805, 40275856, 39967308 and 31983312. Two more that the grounder named in its search notes were also fetched: 41535445 and 36324417. All exist and all are the paper the row implies.
Retraction check. PubMed publication types: no Retracted Publication and no Expression of Concern. Crossref updated-by was checked on every DOI. The only hit is the erratum on Weaver 2021 (DOI 10.1016/s1474-4422(24)00094-2, PMID 38508839, a Published Erratum). Grok reported it correctly, and the location-impact figures in D6 are not the corrected panel.
Citations that failed verification: 0. No PMID was wrong or nonexistent, and none was attached to a different paper.
Number drift: 2 rows repaired.
D2: the row read "205 of 860 (24.0%)" and "18 of 876 (2.1%)". Sims 2023 gives "205 participants (24.0%)" and "18 (2.1%)". The randomized denominators of 860 and 876 do not produce 24.0% (205/860 = 23.8%), because the abstract's percentages come from a different denominator. The denominators were removed and the abstract's wording restored.
D5: the row attributed the donepezil 5 mg estimate (−0.92, −1.44 to −0.40) to "8 trials and 4373 participants". Those are the whole review's totals; the donepezil 5 mg estimate came from a subset of trials. The row now says "in a Cochrane review of 8 trials (4373 participants)". It also adds galantamine's −2.01 so that the treatment side shows the range of effects.
D5 also described Cheng 2026 as "16 placebo-controlled vascular-dementia trials". The abstract says 16 parallel-group randomized trials of pharmacological interventions, with placebo-controlled outcomes pooled where comparable. The row now says "16 randomized trials" of drug therapy.
Smaller precision edit. D1 now names Nasrallah 2019's population: the SPRINT MRI substudy of hypertensive adults without diabetes or stroke, 449 with follow-up MRI.
Spot-checked and confirmed correct against the abstracts. Liem (25 carriers, CAMCOG p<0.01, lacunes, microbleeds and ventricles but not WMH); Bahrani (p<0.028, growth not regression); Neilson (0.26%, 0.14%, 0.05%; n=129 and n=460); Yoon (9/63, 14.3%); Charidimou (74.5% and 95.0% at autopsy); Switzer (28.6% and 65.3%, n=54); van Dyck (12.6%); Weaver (1286/2950, OR >20, location-only model performing like the full model); Gibson (2.26 FIM, 0.17–4.22, 6 trials, 336 participants, 22-point MCID, 20 of 24 trials remediation); Skidmore (χ² 4.60, P=.03, n=71 versus n=85); Vetter (stage reversal, no means in the abstract); Molnar (57 trials, 46%, 4-point ADAS-Cog, one patient-perspective study); Wiseman (115 of 123; β −0.16, 0.17 and 0.19); Gibbon (189 and 154; −0.045, −0.050 and 0.136); Robb (3540 and 489, 0.40 points/year). For the deletions: Bath (ACE-R −3.6, −9.7 to 2.4), Tzourio (19% and 34%), Jamshidian (4 and 1 percentage points; 100% and 85%) and Juraschek (5.7% versus 5.0%; HR 1.06 and 1.77).

Dot A versus Dot B¶

All six rewritten rows have distinct sides. D9 is the closest call: both sides are retina–brain associations. Dot A is cross-sectional OCT angiography against diffusion and reactivity; Dot B is longitudinal co-progression plus a null for fundus calibre. These are different findings from different designs, and the junction (treatment-effect mediation) is the step neither takes, so the row stays. D9 does not duplicate D13: the two concern different candidate surrogates.

Absences re-run (PubMed E-utilities, 2026-10-07, plus Europe PMC where named)¶

D1: unsupported absence, repaired. The grounder's 4-record query required "endpoint/surrogate/mediat*" and did not surface the directly relevant sporadic study. Auditor query: (CADASIL OR NOTCH3) AND sporadic AND (lacun* OR microbleed*) AND cognit* AND (longitudinal OR progression OR follow-up) returned 2 records, neither relevant. A second query, "small vessel disease" AND ("incident lacune*" OR "new lacune*" OR "lacune count") AND ("trial endpoint" OR "outcome measure" OR "sample size" OR surrogate) (all [Title/Abstract]), returned 4 records. Among them were Zeestraten 2017 (PMID 28978655) and a 2026 incident-lacune meta-analysis (PMID 41605866); both were fetched and both are clean on PubMed types and Crossref. Zeestraten shows that in sporadic SVD (SCANS, n=99), new lacunes and DTI change tracked executive and global decline, but it was WMH growth (together with DTI) that predicted dementia. Liem's dissociation therefore does not cleanly transfer, which contradicts the framing that only the transfer test was missing. The junction now names both papers and narrows to the same-method cross-etiology comparison.
D2: holds. The grounder's shared-model query re-run returned 0. "amyloid-related imaging abnormalities" AND "cerebral amyloid angiopathy" AND (predict* OR risk score OR risk model) AND (microbleed* OR siderosis) returned 2 reviews. "amyloid-related imaging abnormalities" AND "Boston criteria" returned 7; the only primary study among them is a Boston-version prevalence comparison in 75 Alzheimer patients (PMID 39836667) with no outcome model. An anti-amyloid ARIA-plus-ICH prediction query restricted to cohorts returned 0. Europe PMC ("amyloid-related imaging abnormalities" AND "Boston criteria" AND ICH AND ("prediction model" OR "risk score")) returned 10 hits: reviews (including PMID 41559482, fetched), CAA-severity and polygenic models with no ARIA outcome, and conference-abstract compilations. None is a shared model.
D5: holds. Cholinesterase inhibitor × (subcortical vascular OR vascular dementia OR vascular cognitive OR small vessel) × (amyloid OR PiB OR p-tau OR florbetaben OR florbetapir) returned 31 records. Titles were read: reviews, guidelines, animal work and STREAM-VCI (methylphenidate). None is a biomarker-stratified cholinesterase analysis. A narrower subcortical-VCI × amyloid × response query returned 0.
D6: holds. (lesion location OR infarct location OR lesion-symptom OR lesion network OR strategic infarct) AND rehabilitation AND cognit* AND stroke AND (randomized OR stratif* OR tailored OR personali*) returned 13 records. The only trial is the virtual-reality perspective trial the row already names. Nothing assigns rehabilitation strategy from location.
D8: holds, but narrowly; context added. The grounder's 0 depends on requiring "caregiver". A wider query, (vascular dementia OR vascular cognitive OR post-stroke cognitive) AND (minimal clinically important OR minimally important OR meaningful change OR clinically meaningful OR anchor-based OR responder), returned 29 records. It surfaced Wu 2019 (PMID 31097928, fetched, clean): an anchor-based MoCA MCID in 65 stroke-rehabilitation patients, which Gibson 2022 cites as 1.22 points. That threshold is patient-side, not caregiver-anchored and not a vascular-dementia trial responder definition, so the junction stands. The row now names Wu 2019 as the nearest neighbour.
D9: holds, trial context added. retina* AND (small vessel disease OR WMH OR lacunar) AND (randomized OR trial) AND (surrogate OR mediat* OR treatment effect OR outcome measure) returned 2 records. One was the SNORE-ASA protocol (PMID 29097299); its results paper, Ward 2025 (PMID 39301859, RCT, fetched, clean on both checks), has 277 participants with repeated retinal calibre and WMH at 3 years inside ASPREE's aspirin-versus-placebo randomization. It tested OSA and the OSA × aspirin interaction, not retinal mediation. The row now names it as the one randomized setting where the test could be run.

Deletions checked¶

D3 deletion upheld. Seeded junction: "cognitive endpoint embedded in stroke prevention." PROGRESS (Tzourio 2003; 6105 patients with prior stroke or TIA; dementia and MMSE decline as outcomes) and PODCAST (Bath 2017; ACE-R as primary outcome) are exactly that, and their figures were verified. OQ-3 keeps the open part.
D4 deletion upheld, with a caveat. Jaime Garcia 2025 (181 mild-stroke patients enriched for SVD; 13 blood markers; BBB permeability and CVR at baseline and 1 year) is the longitudinal multimodal pairing the seeded row described. However, its abstract reports marker associations with DWI lesions, CVR and MCI and no marker–leakage association. The row was groundable in narrower form and should then have said: "Jaime Garcia measured leakage alongside 13 endothelial and pericyte markers over a year; the abstract reports no marker that tracks leakage, so a leakage-specific blood signature is unshown." The auditor did not restore it, because that claim rests on an abstract's silence rather than a reported null. It is recorded here as a candidate for the next sweep, with full text.
D7 deletion upheld. Jamshidian 2025 computed individualized net benefit inside SPRINT across death, cardiovascular, cognitive and serious-adverse-event outcomes. Juraschek 2020 found orthostatic hypotension unassociated with CV events, syncope or injurious falls, and its association with hypotension-related visits did not differ by BP goal. Together they remove the premise that orthostatic harm offsets the cognitive benefit. The "dynamic, visit-by-visit" target remains unpublished, as Grok noted; that is a design idea rather than a junction between two evidenced dots.
D10 deletion upheld. The seeded wording (mixed-pathology autopsy → living multimodal biomarkers → calibrated attributable fractions) is D12's junction (autopsy attributable fractions 22–100% → plasma NfL/p-tau217/GFAP dissociation → in-vivo individual attributable fractions). The deletion is a genuine duplicate, not a deletion made because the row was hard to ground.

Motivated anchoring¶

Count: 0 rows where a citation was attached to a claim it does not support. The two number drifts (D2, D5) were imprecise, not reached for. D6's Gibson 2022 is remediation-dominated occupational therapy, a looser fit to the seeded "individualized rehabilitation goals" than Skidmore. The row states what Gibson found and leans on Skidmore for the junction, so it was left. The grounder disclosed where abstracts lacked numbers (Dewenter, Vetter) and did not invent any.

Judgement¶

The six grounded rows can now be trusted by the novelty gate: citations are verified, the numbers match the sources, and each absence carries a re-run query. One absence (D1) was genuinely unsupported and has been corrected; it is the failure mode this pass was designed to catch, and it slipped through the grounder's query design. The four deletions are justified. Follow-ups: INDEX.md still states 27 junctions and should say 23 (outside this audit's write scope). Revisit D4's narrower leakage-signature question against the Jaime Garcia full text. D11–D27 were not audited here.

2026-10-07 — Grounding of uncited dot rows D1–D10 (Grok, grounder)¶

Scope. Ten uncited rows (D1–D10). Rewrote 6 (D1, D2, D5, D6, D8, D9). Deleted 4 (D3, D4, D7, D10). D11–D27 were left as written. IDs were not renumbered, so the table now runs D1, D2, D5, D6, D8, D9, then D11–D27. INDEX.md still says 27 junctions; it was outside the write scope for this pass.

Every PMID below was fetched from PubMed E-utilities in this session and read before it was written. Publication types were checked for Retracted Publication and Expression of Concern. Crossref updated-by was checked on the DOI of every record cited in the rewritten rows or named here as a reason for deletion. No cited record was retracted or under an expression of concern. The one updated-by hit is an erratum, described under D6.

Rows rewritten¶

D1. CADASIL side is Liem 2009 (PMID 19139365): in 25 NOTCH3 carriers over 7 years, cognitive decline tracked lacunes, microbleeds and ventricular volume, and white-matter hyperintensity change did not. Sporadic side is Nasrallah 2019 (PMID 31408137), white-matter hyperintensity growth lower by 0.54 cm³ under intensive systolic control, and Bahrani 2023 (PMID 37840499), delayed recall associated with growth (p<0.028) and not with regression. The junction is not "nobody has compared the etiologies." Neilson 2026 (PMID 41614465) reports white-matter hyperintensity progression of 0.26%, 0.14% and 0.05% across proximal CADASIL, distal CADASIL and sporadic disease, and Yoon 2021 (PMID 33716917) found greater lacune and microbleed accrual in 9/63 NOTCH3-variant subcortical-MCI patients without a cognitive difference. Neither tests Liem's dissociation as a sporadic trial endpoint.
D2. Boston v2.0 probable-CAA sensitivity 74.5% and specificity 95.0% at autopsy (Charidimou 2022, PMID 35841910), falling to 28.6% and 65.3% without symptomatic haemorrhage (Switzer 2024, PMID 38710005). ARIA-E in 12.6% on lecanemab (van Dyck 2023, PMID 36449413) and 205/860 (24.0%) on donanemab (Sims 2023, PMID 37459141). The shared prospective model is absent on the query below.
D5. Dewenter 2026 (PMID 42252508) dissociates plasma NfL, p-tau217 and GFAP in 76 memory-clinic patients and 41 ADNI participants. The abstract gives no coefficients, and the row says so. Cholinesterase side is Battle 2021 (PMID 33704781), donepezil 5 mg ADAS-Cog −0.92, and Cheng 2026 (PMID 42635820), 16 placebo-controlled vascular-dementia trials. The biomarker-by-treatment interaction was not in the searches below. The nearest cholinesterase hit, Thavichachart 2006 (PMID 16700849), is an open-label uncontrolled series of 75 patients pooled by clinical diagnosis.
D6. Weaver 2021 (PMID 33901427): 1286/2950 (43.6%) with post-stroke cognitive impairment, voxel-wise odds ratios above 20 in three regions, location impact score alone similar to the full clinical model. Gibson 2022 (PMID 35349186): basic ADL +2.26 FIM points, below the 22-point stroke minimal important difference, 20 of 24 trials remediation. Skidmore 2022 (PMID 33728742) is a secondary analysis in which a ventromedial lesion attenuated strategy-training response (χ² 4.60, P=0.03, n=71). That is moderation, not assignment of strategy from the map. Crossref records an erratum to Weaver (DOI 10.1016/s1474-4422(24)00094-2, PMID 38508839, fetched from PMC10969985): the image labelled "Language 222/1152 (19.3%) impaired" in figure 3A was wrong. The location-impact figures used in the row are not that panel.
D8. Vetter 1999 (PMID 10097779) compares caregiver burden and everyday-task failure in 36 vascular and 36 Alzheimer homecare patients and reports a stage reversal. The abstract gives no mean scores, and the row says so. Trial side is Battle's −0.92 and Molnar 2009 (PMID 19187414): of 57 dementia-drug trials spanning several diagnoses, 46% discussed clinical significance, usually as an opinion-based 4-point ADAS-Cog change, and one trial measured the patient's perspective.
D9. Wiseman 2023 (PMID 36300327) is the cross-sectional OCT-angiography association (right-sided mean-diffusivity β −0.16). Gibbon 2026 (PMID 42743445) is the longitudinal one: in 189 mild-stroke patients (154 at 1 year), falling retinal vessel density tracked white-matter hyperintensity increase and a lower MoCA. Robb 2025 (PMID 40966613) found fundus calibre null against cognition (n=3540) and against white-matter hyperintensity volume (n=489). The junction that remains is surrogate validation, a treatment-effect mediation, which the trial query below did not find. Biffi 2022 (PMID 35421194) said no longitudinal retinal study in small-vessel disease existed; that absence is out of date for OCT angiography.

Rows deleted, and why¶

D3 deleted. The seeded junction was a cognitive endpoint embedded in stroke secondary prevention. That trial has been published. PODCAST randomized 83 people with recent stroke to intensive versus guideline blood pressure and lipid targets with Addenbrooke's Cognitive Examination-Revised as the primary outcome; the blood-pressure contrast on that score was −3.6 (95% CI −9.7 to 2.4) over a median 24 months (Bath 2017, PMID 28095412). PROGRESS, in 6105 people with prior stroke or TIA, reduced cognitive decline by 19% (4% to 32%) and dementia-with-recurrent-stroke by 34% (3% to 55%), with no clear effect on dementia in the absence of recurrent stroke (Tzourio 2003, PMID 12742805). OQ-3, which asks whether optimized prevention reduces delayed post-stroke dementia, remains a different question. The dot claimed the embedding was missing.
D4 deleted. The seeded junction was a longitudinal pairing of blood–brain-barrier imaging and circulating endothelial markers in small-vessel disease. A strict PubMed phrase query returned 0, because it required the contiguous phrases "blood-brain barrier permeability" and "ICAM" or "endothelial dysfunction." A looser Europe PMC search surfaced Jaime Garcia 2025 (PMID 40275856), which was then fetched. In 181 people with mild ischaemic stroke enriched for small-vessel disease, a 13-marker blood panel and MRI that included barrier permeability and cerebrovascular reactivity were analysed cross-sectionally and at 1 year. Vascular endothelial growth factor was associated with incident diffusion-positive lesions, intercellular adhesion molecule-1 with lower reactivity, and platelet-selectin with mild cognitive impairment at 1 year. That is the multimodal longitudinal study the row was pointing at. Residual, not rewritten into a new row: the abstract lists barrier permeability among the imaging measures examined and does not report a coefficient for change in leakage.
D7 deleted. The seeded junction was a net-benefit target balancing orthostatic harm against the cognitive effect of intensive blood-pressure control. Jamshidian 2025 (PMID 39967308) estimated that net benefit inside SPRINT for 5143 participants aged 65 or older, combining death, cardiovascular outcomes, cognitive outcomes and serious adverse events including syncope under simulated preference weights. When benefits were weighted much more heavily than harms, the median net benefit was 4 percentage points and 100% favoured systolic <120 mm Hg; when benefits and harms were weighted similarly, the median was 1 percentage point and 85% favoured intensive treatment. Juraschek 2020 (PMID 31983312) found orthostatic hypotension at 5.7% of follow-up assessments on the standard goal and 5.0% on the intensive goal; it was not associated with cardiovascular events (hazard ratio 1.06, 95% CI 0.78–1.44), syncope, injurious falls or acute kidney injury, and was associated with hypotension-related hospital or emergency visits (1.77, 1.11–2.82). Residual, not rewritten into a new row: Jamshidian is a post-hoc weighting of a fixed target in community-dwelling SPRINT participants, not a prospectively adaptive target in an MRI-defined small-vessel population.
D10 deleted. Grounded, the row is the same junction D12 already states: autopsy attributable fractions on one side, living plasma markers on the other, and in-vivo calibration as the missing step. A second copy would not be auditable as a separate claim. The calibration search below returned 0 PubMed records, which is consistent with D12 remaining open. D12 was not edited.

Searches behind absence claims¶

PubMed E-utilities, 2026-10-07, unless noted. Counts are PubMed esearch counts. Titles of the returned records were read; abstracts were fetched when a title could have been the junction.

D1 endpoint transfer. (CADASIL[Title/Abstract] OR NOTCH3[Title/Abstract]) AND (sporadic[Title/Abstract] OR "small vessel disease"[Title/Abstract]) AND (lacune[Title/Abstract] OR lacunar[Title/Abstract] OR microbleed[Title/Abstract]) AND ("white matter"[Title/Abstract]) AND (cognit*[Title/Abstract]) AND (endpoint[Title/Abstract] OR surrogate[Title/Abstract] OR mediat*[Title/Abstract]) — 4 records (PMIDs 34950895, 36564356, 38162905, 37021189). From the titles: a pericyte review, an arteriolar-neuropathology review, an interstitial-fluid study in CADASIL, and the TREAT-SVDs protocol. None is a test of the lacune-versus-white-matter cognitive hierarchy as a sporadic endpoint. A wider cross-etiology query returned 12, and that is where Yoon 2021 and Neilson 2026 were found and then fetched.
D2 shared model. ("amyloid-related imaging abnormalities"[Title/Abstract] OR ARIA[Title/Abstract]) AND ("intracerebral haemorrhage"[Title/Abstract] OR "intracerebral hemorrhage"[Title/Abstract]) AND ("Boston criteria"[Title/Abstract] OR "probable CAA"[Title/Abstract]) AND (model[Title/Abstract] OR score[Title/Abstract]) — 0 records. Broader: "amyloid-related imaging abnormalities"[Title/Abstract] AND (microbleed*[Title/Abstract] OR siderosis[Title/Abstract] OR "Boston criteria"[Title/Abstract]) AND (predict*[Title] OR "risk score"[Title/Abstract] OR "odds ratio"[Title/Abstract]) AND (lecanemab[Title/Abstract] OR donanemab[Title/Abstract] OR aducanumab[Title/Abstract] OR "anti-amyloid"[Title/Abstract]) — 5 records (PMIDs 38017568, 36281665, 40112274, 42657501, 42040755). Titles are a meta-analysis of antibody trials, an ARIA description in unimpaired people, an incidence meta-analysis, incident microbleeds in A4, and perivascular spaces predicting ARIA in preclinical Alzheimer disease. None is a Boston v2.0 model of both haemorrhage and ARIA.
D3, the search that closed the row. (perindopril[Title/Abstract] OR "blood pressure"[Title/Abstract]) AND (stroke[Title/Abstract]) AND (dementia[Title/Abstract]) AND (randomized[Title/Abstract] OR randomised[Title/Abstract]) AND (cognitive[Title/Abstract] OR cognition[Title/Abstract]) AND ("secondary prevention"[Title/Abstract] OR "recurrent stroke"[Title/Abstract] OR "previous stroke"[Title/Abstract]) — 13 records, including Bath 2017 and Tzourio 2003, both fetched.
D4, the search that closed the row. Strict PubMed query ("dynamic contrast"[Title/Abstract] OR "DCE-MRI"[Title/Abstract] OR "blood-brain barrier permeability"[Title/Abstract]) AND (ICAM[Title/Abstract] OR VCAM[Title/Abstract] OR "E-selectin"[Title/Abstract] OR "endothelial marker"[Title/Abstract] OR "endothelial dysfunction"[Title/Abstract]) AND (longitudinal[Title/Abstract] OR serial[Title/Abstract] OR "follow-up"[Title/Abstract] OR prospective[Title/Abstract]) AND ("small vessel"[Title/Abstract] OR "white matter"[Title/Abstract] OR CADASIL[Title/Abstract] OR SVD[Title/Abstract]) — 0 records. A second PubMed query requiring soluble ICAM or VCAM plus dynamic-contrast imaging in small-vessel disease or white-matter hyperintensities returned 2, a mouse model (PMID 40724955) and a cerebellar-disease paper (PMID 39564172). Europe PMC, queried more loosely as ("DCE-MRI" OR "dynamic contrast" OR "blood-brain barrier permeability") AND (ICAM OR VCAM OR "E-selectin") AND (longitudinal OR serial) AND ("small vessel" OR "white matter"), returned a hit count of 323; the first hit was Jaime Garcia 2025 (PMID 40275856), which was fetched. That loose hit count is not an absence denominator.
D5 interaction. (donepezil[Title/Abstract] OR galantamine[Title/Abstract] OR rivastigmine[Title/Abstract] OR "cholinesterase inhibitor"[Title/Abstract]) AND ("vascular dementia"[Title/Abstract] OR "vascular cognitive impairment"[Title/Abstract] OR "vascular cognitive"[Title/Abstract]) AND (amyloid[Title/Abstract] OR "p-tau"[Title/Abstract] OR ptau217[Title/Abstract] OR "tau"[Title/Abstract] OR "cerebrospinal"[Title/Abstract] OR "PIB"[Title/Abstract] OR "florbetapir"[Title/Abstract]) AND (interaction[Title/Abstract] OR subgroup[Title/Abstract] OR stratified[Title/Abstract] OR moderator[Title/Abstract] OR response[Title/Abstract]) — 2 records (PMID 41535445, a polypharmacy paper; PMID 36324417, methylphenidate in STREAM-VCI). Broader query dropping the cerebrospinal-fluid and interaction wording and requiring amyloid or p-tau plus response, efficacy, subgroup or interaction — 11 records. Titles were read. The only cholinesterase trial among them was Thavichachart 2006, fetched; it is not biomarker-stratified.
D6 assignment. (infarct location[Title/Abstract] OR "lesion location"[Title/Abstract]) AND ("cognitive rehabilitation"[Title/Abstract] OR "strategy training"[Title/Abstract]) AND (randomised[Title/Abstract] OR randomized[Title/Abstract]) AND stroke[Title/Abstract] — 2 records, Skidmore 2022 and Janavičiūtė-Pužauskė 2025 (PMID 40682194), both fetched. The second randomizes virtual-reality perspective, not lesion anatomy. A wider "lesion network" query returned 81 records; the top titles were lesion-symptom maps of gait, speech and swallowing, not strategy assignment. Crossref updated-by on PMID 40682194 was empty.
D7, the search that closed the row. SPRINT[Title] AND (orthostatic[Title/Abstract] OR syncope[Title/Abstract] OR hypotension[Title/Abstract]) — 40 records, including Jamshidian 2025 and Juraschek 2020, both fetched.
D8 anchor. ("vascular dementia"[Title/Abstract] OR "vascular cognitive impairment"[Title/Abstract]) AND (caregiver[Title/Abstract]) AND ("minimal clinically important"[Title/Abstract] OR "minimally important"[Title/Abstract] OR "responder definition"[Title/Abstract] OR "anchor-based"[Title/Abstract]) — 0 records. A wider query that did not require the caregiver term returned 3 (PMID 30891742, the memantine Cochrane review; PMID 19187414, Molnar, fetched and cited; PMID 40749241, moyamoya revascularization). None is a caregiver-anchored vascular-cognitive responder threshold.
D9 surrogate trial. ("OCT angiography"[Title/Abstract] OR OCTA[Title/Abstract] OR "optical coherence tomography angiography"[Title/Abstract]) AND ("small vessel"[Title/Abstract] OR "white matter hyperintensit*"[Title/Abstract]) AND (surrogate[Title/Abstract] OR mediation[Title/Abstract] OR mediates[Title/Abstract]) AND (randomized[Title/Abstract] OR randomised[Title/Abstract] OR trial[Title/Abstract]) — 0 records. A longitudinal OCT-angiography query without the trial restriction returned 15. Titles were read. Full abstracts were fetched for Gibbon 2026 (the mild-stroke progression study, now Dot B), Robb was found on a separate retinal-calibre query and fetched, and for four others that might have been longitudinal progression studies: PMID 42307684 (cross-sectional choroidal OCT angiography in CAA, n=23), PMID 42064977 (longitudinal, but in neuromyelitis optica, not sporadic small-vessel disease), PMID 40799661 (choroid-plexus asymmetry predicting fundus flow, the mediation lost after age adjustment), PMID 35168582 and PMID 32527301 and PMID 36090874 (cross-sectional). Biffi 2022 was already in the candidate set and was fetched.
D10 calibration. ("attributable"[Title/Abstract]) AND (neuropatholog*[Title/Abstract] OR autopsy[Title/Abstract]) AND (plasma[Title/Abstract] OR NfL[Title/Abstract] OR "neurofilament"[Title/Abstract] OR "p-tau"[Title/Abstract]) AND (vascular[Title/Abstract] OR cerebrovascular[Title/Abstract] OR "small vessel"[Title/Abstract]) AND (cognit*[Title/Abstract] OR dementia[Title/Abstract]) — 0 records.

Cochrane CENTRAL and SciELO were not searched; both are unreachable from this environment (SOURCES.md). Absence claims above are PubMed counts, with the one Europe PMC recovery named under D4. No NCT identifier was added. PODCAST's registry number is printed in the Bath abstract as ISRCTN85562386; it was not re-checked against a registry API.

What this pass could not make more precise¶

Dewenter 2026 and Vetter 1999 do not put a mean difference in the abstract. The rows use the sample sizes and the direction the abstracts do state, and they say when a mean is absent.
Jaime Garcia 2025 does not, in the abstract, give the dynamic-contrast leakage coefficient. Deleting D4 treats the paper as the seeded junction. A reader who wants the narrower claim "endothelial-marker change has not been shown to track leakage change" will not find it written as a row.
Jamshidian 2025 does not enrol on MRI small-vessel burden and does not change the target at each visit from the standing blood pressure. Deleting D7 treats the individualized net-benefit estimate as the seeded junction.

No row was left in the uncited keyword form. No row was kept by inventing a number the fetched abstract did not contain.

2026-09-02 — Independent re-audit of Claude's second deepening pass (Codex auditor; Claude additions author)¶

Outcome. Re-audited the substantial additions recorded in the immediately preceding Claude-authored pass. All 19 wiki pages are now status: curated. The audit used fresh live PubMed E-utilities and ClinicalTrials.gov API responses in this session; no identifier was accepted from memory.

Citation and registry integrity¶

Extracted and live-resolved every wiki PMID. The pre-audit wiki contained 416 distinct PMIDs, all 416 of which resolved. After corrections it contains 417 distinct PMIDs: one unique orphan was removed and two live-resolved corrective records were added (PMID 36749936 and PMID 38583421).
Extracted and live-resolved every wiki NCT identifier. The final wiki contains 44 distinct NCT IDs. The 42 already present all resolved; the two corrective additions, NCT03863665 and NCT03785067, were then fetched individually from the live API and resolved to PROHIBIT-ICH and the TRIDENT Cognitive Sub-Study.
PubMed metadata comparison found no substantive title, author, journal or year mismatch among the 416 pre-audit records. Numerical spot-checks emphasized the high-risk passages named in Claude's handoff and supported their reported sample sizes and estimates; interpretive overreach was corrected separately below.
Four registry completion dates had changed and were refreshed from the live records: NCT07252544 to 2028-02, NCT07012629 to 2040-01, NCT04700826 to 2031-01, and NCT06933212 to 2027-07.

Claim-level corrections¶

Removed or softened unsupported causal, temporal, comparative and prescriptive language: GFAP “preceding” injury without baseline MRI; NfL change “outperforming” level; albumin quotient as a positive vascular-copathology indicator; observational DTI-ALPS ordering as proof; ASL heterogeneity as superior to mean perfusion; lenticulostriate pulsatility as proven upstream; cerebrovascular variance explained as a treatment-effect ceiling; cross-trial exercise-versus-drug equivalence; active-control causation; carotid haemodynamic causation; anticoagulation “failure”; CAA recurrence as a rule that should govern restart; and observational deprescribing findings as proof of benefit or safety.
Corrected the unsupported claim that no post-ICH prevention trial has a cognitive endpoint. The iDEF trial cohort has a published MoCA trajectory analysis (PMID 38583421), while PROHIBIT-ICH (NCT03863665) and the TRIDENT Cognitive Sub-Study (NCT03785067) include cognitive outcomes. The remaining gap is narrower: cognition is secondary or a substudy outcome rather than a primary clinical dementia-prevention endpoint.
Corrected the burden narrative after a targeted search found contemporary VaD-specific US Medicare costs (PMID 36749936). The older Swedish study remains the located VaD-specific societal estimate; the modern record measures direct payer cost and does not close that societal-cost gap.
Reframed qualitative caregiver, successful-reperfusion, financial-capacity and cross-country cost findings to match their designs and avoid unsupported generalization.

Absence searches rerun¶

Exercise versus cholinesterase inhibitor: targeted PubMed search returned reviews/management literature but no head-to-head randomized trial; claim retained with cross-trial-comparison caveat.
Fabry randomized cognitive endpoint: targeted PubMed search returned no qualifying randomized cognitive-endpoint trial; claim retained.
Contemporary VaD-specific societal cost: targeted search found the 2023 Medicare subtype study (PMID 36749936), which is direct health-care cost, but no contemporary societal total; wording narrowed accordingly.
Validated sporadic-SVD patient-reported outcome: targeted PubMed search found the CADASIL-specific CADA-PRO validation (PMID 39234671) but no validated sporadic-SVD-specific instrument; claim retained and scoped.
Post-ICH cognitive-endpoint trial: claim failed, was removed, and was replaced with the live evidence above.

Padding and breadth¶

Removed five orphan reference entries from overview.md, one from mixed-pathology-and-alzheimer-overlap.md, and one from monogenic-and-inherited-forms.md. Six of those PMIDs remain legitimately body-cited elsewhere in the wiki; PMID 20851424 was the one unique record removed.
The condition's distinct-records-per-page figure was 416 / 19 = 21.89 before this audit and is 417 / 19 = 21.95 after it. The Claude deepening therefore survives the breadth check: the audit removed padding while adding only evidence needed to correct two incomplete or false claims.
Final wiki size: 4,006 lines across 19 pages. No unresolved claim or identifier defect remains from the audited additions.

2026-09-02 — Second deepening pass: widening the evidence base (Claude author; independent re-audit pending)¶

Why this pass ran. After the first deepening pass the condition was structurally complete and page-level citation density looked adequate, but the measure that matters was low: 17.6 distinct PubMed records per page across the whole condition (335 distinct PMIDs / 19 pages), against 15.4–21.1 for the reference conditions. A condition can look well cited page by page while every page recycles one shared pool. The brief for this pass was therefore to widen the source base — bring in records the condition did not yet cite, sourced to each page's own topic — not to lengthen pages with the references they already had.

Result. 335 → 416 distinct PMIDs across the wiki (21.9 distinct records per page), with 3,578 → 4,011 lines. Every one of the 416 was re-resolved through live PubMed E-utilities on 2026-09-02; all 416 resolved. All 19 pages set to status: draft for independent re-audit.

What was deepened, page by page¶

Page	Records added	What they added
`nosology-and-diagnostic-criteria.md`	5	Knopman's population-based validation (sensitivity 0.75 vs the hospital-series 0.20–0.25 — a criteria-architecture rather than criteria-quality difference); Cerullo's pooled κ 0.79 for ICD-10 VaD against Chui's 0.30–0.61; Erkinjuntti's subcortical-VaD research criteria as the abandoned homogeneity strategy; Hsiung's 40% two-year VCI-ND conversion and 14% reversion
`epidemiology-and-burden.md`	5	Hisayama incidence (9.5/1,000 py, six times Rotterdam/NEDICES despite stronger ascertainment); the internally inconsistent Chinese prevalence set (VCI 1.54% ≤ VaD 1.6%); Cao's male-predominant VaD prevalence against EURODEM's sex-neutral incidence; the only VaD-specific societal cost estimate (23% above AD)
`cerebral-small-vessel-disease.md`	6	the three-way perivascular-space contradiction (count vs volume vs DTI-ALPS); Framingham's multimarker score (HR 1.76, c 0.82–0.83, indistinguishable from a risk questionnaire); microinfarct prevalence by pathology (62% of VaD brains); LADIS cognitive reserve; Persyn's 31-locus GWAS
`cerebral-amyloid-angiopathy.md`	6	Case's quantified MCI prevalence (79%) and vascular cognitive profile; Pfeifer's CAA × AD interaction (45.9% vs 16.6% CASI deficit) against Sin's null mediation; Charidimou 2013 and Roongpiboonsopit's siderosis recurrence curves (74% at 4 years disseminated; 7% at 6 months); a new section on iatrogenic CAA — the only transmissible substrate in this condition
`monogenic-and-inherited-forms.md`	6	Rutten's 12-year onset gap by EGFr position, which is the mechanism generating the 40-fold prevalence gap; Liem's 7-year longitudinal finding that WMH change is unrelated to cognitive decline in CADASIL; Brice's post-50 non-linear trajectory; a direct HTRA1-vs-NOTCH3 imaging comparison; the COL4A1/2 epilepsy phenotype; the nearest randomized Fabry ERT evidence
`pathophysiology.md`	5	Kisler's capillary-level pericyte mechanism and Miners' human tissue dissociation of pericyte loss from barrier leakage; Montagne's age-dependent hippocampal BBB breakdown; Claassen's four-system decomposition of "cerebral autoregulation"; Santisteban's venous and lymphatic neurovascular ageing
`mixed-pathology-and-alzheimer-overlap.md`	5	Boyle 2013/2021 — eleven pathologies explain 43% of decline variance and the cerebrovascular block explains 3–8%; a Chinese community brain bank where LATE-NC exceeds ADNC; Jellinger's 1,700-autopsy age gradient; Barnes's sex-patterned AD+CVD
`subtypes-and-clinical-syndromes.md`	4	strategic lacunes in a population sample (brainstem for non-amnestic MCI); tract-specific strategic WMH outperforming total WMH volume; CREST-2's memory-predominant non-lateralized deficit against the haemodynamic account; Swiss-AF's 85%-silent infarcts under anticoagulation
`cognitive-profile-and-assessment.md`	4	Shi's MoCA cut-point analysis (optimum 20/19); the Cochrane IQCODE threshold problem; Demeyere's OCS-vs-MoCA head-to-head exposing MoCA's left-hemisphere bias; a BPSD-by-subtype table (apathy 62% in subcortical VCI, hallucinations 26.6% in mixed)
`neuroimaging.md`	4	Kapeller's collapse of visual-rating agreement for progression (κ 0.19–0.39) against baseline (0.59–0.78); Wahlund's MTA rating beating its own volumetry; ASL spatial coefficient of variation outperforming mean perfusion; 7T lenticulostriate pulsatility present independently of visible SVD and absent in the MCA
`biomarkers.md`	5	Qu's NfL rate-of-change signal and Prapiadou's finding that GFAP predicts SVD markers nine years ahead while NfL does not; Skillbäck's 1,861-patient albumin-quotient series; the counterintuitive direction of CSF MMP-2; OCT-angiography against sub-visible white-matter damage
`post-stroke-cognitive-impairment.md`	4	Pendlebury's dissociation of delirium (independent of white-matter disease) from infection (conditional on it); Fang's 15.5% covert incident lesions at 6 months; the post-ICH trajectory (32–37% at 5 years) and Taiwan's subtype hazard ordering
`prevention-and-risk-factor-control.md`	5	He 2025 — the first adequately powered randomized trial with dementia as its primary outcome to be positive (RR 0.85 on a 22 mm Hg contrast in 33,995 adults); HYVET-COG; Gottesman's midlife diabetes hazard approaching APOE ε4; Smith's age-window attributable fractions (2–8% after 80); Chinese PAF composition
`treatment.md`	5	Black 2003 and the 54-week open-label extension; a new section on exercise, whose ADAS-Cog effect (−1.71) exceeds donepezil 5 mg; citicoline's post-stroke attention-executive signal
`guidelines.md`	4	EAN 2020 (vascular risk as management, anticoagulation without imaging conditioning); the retired AAN MCI guideline that grades exercise above drugs; ACR imaging appropriateness; Alzheimer's Association DETeCD-ADRD
`clinical-trials-landscape.md`	0 new PMIDs, 9 new NCTs	a fresh ClinicalTrials.gov v2 sweep that found the phase-3 successor to LACI-2 the 2026-08-31 sweep recorded as absent (IMPACT, NCT07252544, n=3,156), plus a cilostazol phase-4, a CADASIL BBB-endpoint phase-2, and a covert-brain-infarct phase-3
`red-flags-and-safety-concerns.md`	4	the anticholinergic prognostic-versus-intervention asymmetry (102,684 vs 299 participants); randomized antihypertensive deprescribing (no harm, and no fall reduction); financial capacity and exploitation
`patient-experience-and-advocacy.md`	4	the caregiver-burden reversal by disease severity; the invisibility problem created by successful reperfusion; Argentine cost data replicating the US hospital-heavy profile
`overview.md`	8 new to the condition (13 reference entries; five were already cited on other pages)	controversies table expanded from 8 to 13 rows; prevention/treatment table extended with the 2025 BP trial, exercise, and the lipid null

Literature layer¶

BIBLIOGRAPHY.md: found and closed a pre-existing gap — 118 records were cited in wiki pages but absent from the bibliography — and added the 81 records introduced by this pass. 199 entries added; wiki-to-bibliography coverage is now complete (0 cited-but-unlisted). Entries are grouped by first-citing page in the house format with full cited by lists.
STATISTICS.md: 23 → 34 sections, adding country-specific VaD prevalence/incidence, cost of vascular dementia, population attributable fractions by age window, randomized blood-pressure prevention with achieved contrasts, exercise and non-drug intervention, BPSD by subtype, additional criteria performance, additional SVD markers, imaging measurement reliability, post-stroke covert injury and delirium, mechanism/biomarker quantities, and neuropathology variance explained. Twelve new rows in "known conflicts and caveats", including the internally inconsistent Chinese prevalence set and the Hisayama incidence gap.
REGISTRY.md: 21 → 25 records (EAN 2020, AAN 2018 retired, ACR 2024 update, Alzheimer's Association DETeCD-ADRD 2025), with per-record notes, five new disagreement rows (exercise-versus-drugs; anticoagulation conditioning; guidelines not yet reflecting the 2025 BP trial; who owns imaging selection; primary-care pathways), and five watch-list additions.
OPEN-QUESTIONS.md: OQ-2 sharpened and partly closed by the 2025 BP trial; six new questions (OQ-31 to OQ-36) on exercise-versus-drug head-to-head, anticholinergic deprescribing, which PVS measurement carries information, the Hisayama incidence gap, GFAP as the earliest signal, and whether haemorrhagic and non-haemorrhagic CAA are one disease; ten new junctions (D18–D27); four questions moved to the closed/downgraded table.

Searches run¶

Roughly 90 PubMed E-utilities queries across every page's topic (criteria validation and reliability; secular trends and country-specific epidemiology; cost; SVD burden scores, perivascular spaces, microinfarcts, GWAS; CAA prevalence, siderosis recurrence, inflammation, iatrogenic transmission; CADASIL natural history and genotype-phenotype, HTRA1, COL4A1/2, Fabry; neurovascular coupling, BBB permeability, pericytes, autoregulation, glymphatics; mixed pathology, LATE-NC, hippocampal sclerosis, non-Western autopsy cohorts; strategic lesions, carotid stenosis, heart failure, atrial fibrillation; MoCA/MMSE/OCS/IQCODE accuracy, apathy and BPSD; visual rating reliability, MTA, ASL, 7T perforators, FDG-PET; plasma NfL/GFAP, albumin quotient, MMPs, retinal imaging; post-stroke trajectories, delirium, ICH, thrombectomy; HYVET, SCOPE, ARIC, Chinese BP trials, smoking, physical activity, attributable fractions; memantine, nimodipine, cilostazol, citicoline, donepezil, exercise; guideline bodies), plus three live ClinicalTrials.gov v2 condition queries and eight single-study API fetches.

What was deliberately left alone¶

literature/notes/ — the five landmark notes were not touched. None of this pass's additions is a practice-changing landmark of the kind the notes are reserved for, with the possible exception of He 2025 (the positive cluster-randomized BP trial); a note for it is a reasonable follow-up but it is one trial, open-label, and not yet replicated or subtype-adjudicated.
literature/patient-voice/ — the four files were not touched. Organization URLs still need re-fetching at the next audit; this pass added no non-journal sources.
Page structure — no page was restarted, no verified content deleted, no section removed. Every edit is an insertion or an extension of an existing sentence, except where an explicitly stale claim needed correcting (see below).
Five orphan references in overview.md (PMIDs 20660506, 31248555, 31294622, 37222252, 8094895) and one each in mixed-pathology-and-alzheimer-overlap.md (31248555) and monogenic-and-inherited-forms.md (20851424) sit in reference lists without a body citation. These predate this pass; they were left in place rather than deleted, since removing pre-existing content is outside this pass's brief. They are flagged here for the auditor.

Corrections made to existing content¶

clinical-trials-landscape.md carried a dated absence claim — that no registered phase-3 successor to LACI-2 existed. A repeat registry query two days later found IMPACT (NCT07252544). The claim was corrected in place and re-labelled as a query-coverage artifact, and the general lesson (registry absence claims decay within days) was added to the statistics conflicts table.
prevention-and-risk-factor-control.md and overview.md framed "does blood-pressure lowering prevent dementia?" as an open two-sided controversy. That framing is no longer accurate for all-cause dementia at large achieved contrasts and was rewritten around He 2025, with the residual open components (target, treated populations, subtype adjudication) stated explicitly.
monogenic-and-inherited-forms.md carried a dated evidence-gap sentence about Fabry enzyme replacement. It was replaced with the nearest randomized evidence (Fellgiebel 2014, white-matter lesion stability in a phase-4 subgroup, NNT 8) plus a re-dated absence statement for cognitive endpoints.
Reference lists were renumbered on every edited page, and one missing reference entry was added to clinical-trials-landscape.md. Every page now has zero body-cited-but-unreferenced PMIDs.

Follow-up for the independent auditor¶

Re-fetch all 416 wiki PMIDs and the 9 new NCT records; check each new numeric claim against its abstract, particularly the effect sizes in the He 2025, Liu-Ambrose, Paradise/Hong/Hilal, Boyle, and Pfeifer passages.
Re-run the searches behind the absence claims this pass makes or preserves: no head-to-head exercise-versus-cholinesterase trial; no randomized cognitive-endpoint trial in Fabry disease; no contemporary VaD-specific societal cost estimate; no validated patient-reported outcome for sporadic SVD; no post-ICH trial with a cognitive endpoint. Stale absences are the most common real error.
Decide whether the five orphan overview references should be removed or given body citations.
Consider whether He 2025 warrants a literature/notes/ entry.
Re-fetch the patient-voice organization URLs, which have not been checked since the build.

Could not verify in this pass: the vascular-dementia-specific effect of the 2025 cluster-randomized blood-pressure trial (it reported all-cause dementia only); whether the Hisayama-versus-European incidence gap has ever been formally decomposed (no such analysis was found); contemporary societal cost of vascular dementia outside the 2003 Swedish and 2011 Argentine estimates.

2026-08-31 — Independent audit of the deepening pass (Grok auditor; Claude author)¶

Audited the new material Claude added to all 19 wiki pages (previously curated content was not re-litigated except where a new claim attached to it). Did not treat the deepening pass as trustworthy. Re-fetched 217 wiki PMIDs via PubMed E-utilities esummary (all 217 resolved) plus 66 priority abstracts and targeted MEDLINE XML for numeric claims. Re-fetched 25 NCT IDs via ClinicalTrials.gov v2 (all resolved; statuses and enrollments match the trials table). Re-ran absence searches for confirmatory Cerebrolysin/Actovegin trials, LACI-3, VASCOG/VICCCS/VasCog-2-WSO autopsy validation, Fabry ERT cognitive prevention, and VaD-specific behavioral RCTs.

Claim–citation pairs checked. Principal new tables and quantitative paragraphs on treatment, nosology, guidelines, prevention, epidemiology, SVD, biomarkers, CAA, mixed pathology, trials, PSCI, monogenic disease, neuroimaging, cognitive assessment, subtypes, red flags, and patient experience. Landmark numbers that matched their abstracts include: Orgogozo/Wilcock memantine; Cheng 2026 16 trials/5,668 and memantine MD −2.20; Shi 2022 SMD −1.11/−1.99 and CDR-SB/EXIT25; Dang 194 RCTs/21 drugs; Guekht Cerebrolysin −6.17 and 67.5% vs 27.0%; ARTEMIDA −2.3; Jia NBP −2.46/−1.39 and 57.1% vs 42.1%; Kwan NBP MD −1.07 (Jia included as putative antiplatelet); Wieland SKT −1.86 and ADL −0.19 with I² 96%/91%; CONIVaD 15% nimodipine adherence; LACI-2 6-month Bath 2026; COMCID −1.8 vs −1.3 at 50 mg bid; OxHARP pulsatility/CVR/perfusion; Gold 1997/2002 and Bacchetta 42%; Chui κ 0.30–0.61; Syst-Eur 50%/55%; Charisis d −0.40; US POINTER 0.029 SD/y; LatAm-FINGERS 0.11 SD/y; Reddin OR 0.96; Guo HR 0.74/0.58; Ott/NEDICES/EURODEM/Honolulu incidence; Liang HR 5.90/1.33 and VaD −1.27 y; Mobley HR 1.68; Boyle 94%/22–100%; Lamar 80%/37%/63%; Yu 1,633/280/five profiles; RUN DMC HR 1.76; Sargurupremraj OR 1.43 vs HR 1.02; Cho 1 in 277; Hack OR 10.81; CROMIS-2 9.8 vs 2.6 per 1,000 py and HAS-BLED C=0.41; SoSTART/COCROACH; CAA recurrence 23%; PRESTIGE-AF siderosis HR 7.7; Edinburgh 12/16/26%; Rabin n=1,722; LAAO 5.16/2.73; MarkVCID PSMD/free-water/CVR/WMH ICC; Kim Dice 0.659/0.722 vs human 0.744; Lo STROKOG trajectories; Kusec C=0.76; Andrew 84%.

Confirmed author decisions. Wieland's dementia global-clinical-status estimate (MD −0.06, 95% CI −1.00 to −0.20) has the point estimate outside its interval; it remains omitted, now with an explicit note. Li 2025's Ginkgo MoCA interval (MD 1.29, 1.24–1.35) is implausibly narrow and was correctly not quoted.

Errors found and fixed - Ding 2026 (PMID 41962914) and Cunningham 2021 (PMID 34028812) both report dementia OR 0.89 but are not the same four-trial set: Ding includes PROGRESS (prior stroke/TIA; 16,823 participants: Syst-Eur, PROGRESS, SCOPE, HYVET-COG); Cunningham is restricted to people without prior cerebrovascular disease (236/7,767 vs 259/7,660). Corrected on the prevention page and STATISTICS.md. - Zhang 2026 IRVR PACC contrast (PMID 41870419): the published abstract prints "0.1 units; 95% CI −0.3 to 0.03", which places the point estimate outside the interval. Arm means (+0.2 IRVR vs +0.3 no-IRVR) imply −0.1. Corrected on the prevention page and STATISTICS.md. - Cerebrolysin Cochrane (PMID 31710397): 6 trials/597 were eligible, but SMD 0.36 is 3 trials/420 and RR 2.69 is 2 trials/379. The deepening pass attributed both estimates to the 6/597 set. Corrected on treatment.md and OPEN-QUESTIONS.md. - Stale absence: a VaD-specific BPSD RCT exists (Pessoa 2026, PMID 41277041; n=30, CBD 300 mg/day × 4 weeks, NPI/BPRS reduced, no cognitive/functional change). Added to treatment.md and BIBLIOGRAPHY.md. The "no VaD-specific behavioral RCT" sentence is retired. - Kwan 2022 (PMID 35833913) numbers for NBP were correct but the framing "Cochrane re-analysis" omitted that NBP entered an antithrombotic-therapy review as a putative antiplatelet. Clarified.

Padding removed. Unused reference-list entries (no body citation) deleted and said so: Markus 2022 on treatment.md; O'Brien and Battle on cognitive-profile-and-assessment.md; Hankey 2013 on post-stroke-cognitive-impairment.md; duplicate Bhanu 2021 on red-flags-and-safety-concerns.md; duplicate Wardlaw plus unused McShane on guidelines.md; Liu 2023, Debette 2010, and Wardlaw 2019 on neuroimaging.md; O'Brien and Gorelick on patient-experience-and-advocacy.md; Pearce, Wardlaw 2019, and Greenberg 2020 on subtypes-and-clinical-syndromes.md; Pantoni and Kalaria on pathophysiology.md; Wardlaw 2013, Iadecola 2019, and Greenberg 2020 on biomarkers.md.

Absence searches re-run (2026-08-31). No VASCOG/VICCCS/VasCog-2-WSO autopsy validation (0 hits). No confirmatory Actovegin trial. Cerebrolysin 2020–2026 hits were a review and an introduction, not a new eligible RCT. No LACI-3 on ClinicalTrials.gov (only LACI-1 NCT02481323 and LACI-2 NCT03451591). No Fabry ERT randomized cognitive-prevention trial. Covert-SVD antiplatelet "do not recommend" remains supported.

ClinicalTrials.gov. All 25 tabulated NCT IDs resolved. FINGER (NCT01041989) and ASPREE (NCT01038583) remain registry-status "unknown" despite published results. No registered phase-3 LACI-2 successor. The ISRCTN coverage limit stands.

Promotion. All 19 wiki pages set to status: curated after the corrections above. INDEX.md updated.

Remaining limits (not unmarked errors). Regulator labels were not independently retrieved. Li 2025's Ginkgo MoCA interval is still not quoted. The Korean NOTCH3 genetic prevalence (1 in 277) is ancestry-specific and already so labelled. Several pages remain in the 136–200 line range with 15–38 unique PMIDs.

Searches run. PubMed E-utilities esummary on 217 wiki PMIDs; efetch abstracts for 66+ priority records and XML for Saito, Cui, Yu, Lamar, Rodrigues, Rabin, Kim, Ding, Zhang, Pessoa; esearch for butylphthalide Cochrane, VASCOG autopsy validation, Cerebrolysin 2020–2026, Actovegin confirmatory, LACI-3, Fabry ERT cognition, VaD behavioral RCTs. ClinicalTrials.gov v2 for 25 NCT IDs plus condition queries (vascular cognitive impairment; cerebral small vessel disease; LACI-3).

Follow-up. Next sweep should watch for a published correction of Zhang 2026's IRVR interval, any independent replication of Pessoa 2026, LACI-3 if it appears on ISRCTN, and any vascular-dementia-specific drug-label change (retrieve the label, not a review).

2026-08-31 — Deepening pass (Claude)¶

What this pass was. Not a sweep and not a rebuild. The condition was structurally complete but thin at 125 lines and ~15 distinct citations per page, against reference conditions running 155–177 lines and 21–37 citations. This pass went page by page through all 19 wiki pages, ran substantially more literature searches than the build did, and extended pages with what was missing: landmark trials with actual effect sizes and intervals, competing guideline positions with their grades, quantified epidemiology in place of qualitative statements, deeper mechanism, and explicit controversies cited on both sides. Nothing verified was deleted and no page was rewritten from scratch.

Result. 19/19 pages extended; total wiki lines 2,376 → 3,009 (mean 125 → 158); distinct PMIDs per page mean ~15 → ~22; distinct verified PMIDs across the condition 121 → 218.

Page	Lines (before → after)	PMIDs (before → after)	Principal additions
`treatment.md`	133 → 200	21 → 38	MMM 300/MMM500 memantine trials with numbers; the raw-scale-vs-SMD measurement dispute (Cochrane vs Shi 2022); the Western-null vs Asia-positive split (194-RCT Bayesian NMA); a table of agents with one positive pivotal trial and no confirmation (Cerebrolysin, Actovegin/ARTEMIDA, butylphthalide, EGb 761); CONIVaD adherence finding; cilostazol stroke-benefit vs null COMCID; LACI-2 6-month results; OxHARP
`prevention-and-risk-factor-control.md`	116 → 153	20 → 31	Syst-Eur as the field's outlier with three competing explanations; Cochrane and drug-class NMA against it; SPRINT global-SVD-factor dose-response; AF rhythm-control/ablation observational signal; definitive lipid-lowering null; a FINGER-replication table (US POINTER, LatAm-FINGERS, the null 2×2 exercise/IRVR trial) with the active-control problem; ACHIEVE
`epidemiology-and-burden.md`	154 → 189	13 → 21	Community incidence table (Rotterdam, NEDICES, EURODEM, Honolulu-Asia); the sex-neutrality of vascular-dementia incidence; mortality and survival meta-analysis; global informal-care cost; secular-trend systematic review; the AF/low-conventional-risk vascular-dementia excess
`nosology-and-diagnostic-criteria.md`	153 → 192	20 → 24	Full autopsy-validation table for ADDTC/NINDS–AIREN/DSM-IV/ICD-10/HIS, interrater κ, the 0.20–0.25 sensitivity of the "probable" tiers, mixed-dementia misclassification rates, and the 42%-no-stroke finding in the oldest old
`cerebral-small-vessel-disease.md`	127 → 156	20 → 28	RUN DMC 14-year progression-to-dementia data; the two-sample-vs-individual-level Mendelian randomization contradiction; two independent drug-target MR analyses nominating calcium-channel blockade; the null perivascular-space/cognition result; disappearing acute cortical microinfarcts; MINERVA
`biomarkers.md`	122 → 160	11 → 19	MarkVCID staged-qualification section with all four validated kits and their numbers; the plasma NfL/p-tau217/GFAP dissociation between vascular and Alzheimer pathology
`clinical-trials-landscape.md`	120 → 185	13 → 28	Live 2026-08-31 ClinicalTrials.gov sweep adding eight studies with a coverage-limit note; a "what the failures taught" table across six trials; an endpoint-readiness table grading six candidate markers
`overview.md`	139 → 177	30 → 49	Eight-row index of live controversies with citations on both sides and page pointers; community incidence and survival anchors
`mixed-pathology-and-alzheimer-overlap.md`	124 → 150	13 → 17	Boyle person-specific attributable fractions (22–100%); Lamar's finding that single cerebrovascular pathologies do not accelerate decline; Yu's five empirical neuropathologic profiles
`monogenic-and-inherited-forms.md`	119 → 152	12 → 16	Clinical-vs-genetic CADASIL prevalence (~40-fold gap); the three-tiered NOTCH3 EGFr risk classification superseding the 1–6/7–34 dichotomy; aggregation-not-signalling mechanism
`red-flags-and-safety-concerns.md`	127 → 149	16 → 20	CROMIS-2 absolute ICH rates and the HAS-BLED C-index of 0.41; basal-ganglia perivascular spaces as an ICH marker; SoSTART and the COCROACH IPD meta-analysis
`cerebral-amyloid-angiopathy.md`	123 → 148	16 → 21	23% recurrence with its predictors; PRESTIGE-AF imaging subanalysis (siderosis stratifies, "probable CAA" does not); Edinburgh CT criteria for MRI-poor settings; the CAA→plaque→tau mediation result; LAAO evidence
`post-stroke-cognitive-impairment.md`	123 → 150	12 → 17	STROKOG trajectory classes and long-term decline rates; the executive-function exemption; failure of general-population risk models (CAIDE AUC 0.53) and the stroke-specific replacement
`guidelines.md`	117 → 152	14 → 18	Four-column comparison of ESO–EAN, ESO covert-SVD, Korean CPG, and the Asian EGb 761 consensus with each body's stated position per question; the two irreconcilable disagreements named explicitly
`pathophysiology.md`	121 → 138	15 → 19	Periventricular-vs-deep WMH mechanistic split and the perilesional penumbra; 7 T CADASIL evidence that vascular dysfunction is locally causal and globally uninformative; OPC-recruitment failure as the rate-limiting repair step
`neuroimaging.md`	117 → 141	12 → 15	Automated WMH segmentation benchmarked against human inter-rater reliability on 8,421 patients, with the volume-vs-voxel distinction and the uncertainty index; ML systematic review's performance-vs-rigour gap
`cognitive-profile-and-assessment.md`	118 → 140	11 → 15	Direction-specific cutoff non-transportability (MoCA lower in less-WEIRD populations); Oxford Cognitive Screen development and independent psychometric critique; the OCS-based prediction model
`subtypes-and-clinical-syndromes.md`	111 → 141	19 → 25	Multivariate lesion-symptom mapping for "strategic" infarcts with its language-testing caveat; asymptomatic carotid stenosis as the cleanest hypoperfusion test; the pathology finding that single substrates may not count; CAA recurrence anchor
`patient-experience-and-advocacy.md`	112 → 136	12 → 16	Quantified unmet need (84%, peaking in year two); the clinician-described VCI "no-man's land"; psychoeducation evidence concentrated on aphasia; CADA-PRO and the absence of a validated sporadic-SVD patient-reported outcome

Searches run. Roughly 60 PubMed E-utilities searches across pharmacotherapy (memantine, Cerebrolysin, actovegin, nimodipine, cilostazol, ginkgo, butylphthalide, network meta-analyses), prevention (Syst-Eur, HYVET, Cochrane antihypertensives, multidomain trials, statins, AF rhythm control, hearing, CPAP), epidemiology (population incidence cohorts, secular trends, mortality, cost), criteria validation, SVD mechanism and genetics, biomarker qualification, imaging automation, CAA recurrence and anticoagulation trials, monogenic prevalence and genotype-phenotype, screening accuracy, and patient experience. Two live ClinicalTrials.gov v2 condition queries (vascular cognitive impairment; cerebral small vessel disease; lacunar stroke) plus eight single-study record fetches.

Literature layer. BIBLIOGRAPHY.md 138 → 294 lines: 110 records added in 13 new topic groups, each with tags and cited-by pages; all 110 were re-resolved through live efetch before writing. STATISTICS.md 108 → 356 lines: fourteen new quick-reference sections (9 → 23 total) (community incidence; mortality and cost; post-stroke trajectory and prediction; criteria-versus-autopsy accuracy; large-series neuropathologic composition; SVD markers and progression; biomarker validation; haemorrhage risk and anticoagulation; expanded prevention and drug trials; imaging automation; monogenic frequency; screening across populations; patient-reported burden), and nine new rows in the known-conflicts table. guidelines/REGISTRY.md 78 → 103 lines: three guideline records added (ESO–EAN 2021, Asian EGb 761 consensus 2019, Korean Dementia Association 2025), with per-record notes and six new disagreement rows naming the contradictions explicitly.

OPEN-QUESTIONS.md 122 → 206 lines: five existing Tier 1 questions (OQ-1, OQ-2, OQ-5, OQ-6, OQ-7) sharpened with the new evidence; five Tier 1 questions added (OQ-21 Asia-Pacific pharmacology, OQ-22 unconfirmed pivotal trials, OQ-23 surrogate mediation, OQ-24 raw-vs-standardized effect framing, OQ-25 single-vs-mixed cerebrovascular pathology); five Tier 2 added (OQ-26 to OQ-30); a new table recording six questions the pass closed or downgraded; and seven new junctions D11–D17.

Verification. Every PMID and NCT written in this pass came from a live PubMed E-utilities or ClinicalTrials.gov v2 query in this session. After writing, all 218 distinct PMIDs appearing anywhere in the condition were re-fetched in a single sweep and all 218 resolved. All 19 wiki pages were set to status: draft for re-audit. Internal link check: 0 broken relative links across wiki pages, INDEX, OPEN-QUESTIONS, and the literature layer.

What I deliberately left alone.

literature/patient-voice/ (README, organizations, themes, sources) — organization entries require live site fetches with recorded access dates and carry the ethics rules in CONVENTIONS §3. This pass was a literature deepening, not a patient-voice re-verification, and stale organization URLs should be re-fetched rather than inherited.
literature/notes/ — the five landmark notes are correct and their papers remain landmarks. Several 2024–2026 papers surfaced in this pass are strong candidates for new notes (Boyle 2018 PMID 29244218 for person-specific attribution; Jacob 2023 PMID 37073486 for SVD progression causality; Lo 2022 PMID 34775838 for the post-stroke decline curve; the COCROACH IPD meta-analysis PMID 37839434), but writing them is a separate task.
The Alzheimer's-disease border. No Alzheimer-side content was added beyond what the overlap and CAA pages need; amyloid/tau staging and anti-amyloid therapy remain owned by conditions/alzheimers-disease/.
INDEX.md page list — unchanged; the canonical 19-page structure was adequate and the brief was to add substance, not structure.
Pages I judged closest to complete at their current length: cerebral-amyloid-angiopathy.md and subtypes-and-clinical-syndromes.md were already dense for their scope, and received targeted additions (recurrence quantification; lesion mapping and the single-substrate pathology finding) rather than proportional expansion.
Three sourced-but-unresolvable gaps were left as stated gaps rather than filled: regulator label histories for vascular-dementia drugs (not retrievable from PubMed); ISRCTN-registered trials, which the ClinicalTrials.gov sweep cannot see and which the LACI-2 registration shows is a real coverage limit — this is now stated in the trials page rather than left implicit; and prospective validation of VASCOG/VICCCS/VasCog-2-WSO against autopsy, which does not exist and is recorded as OQ-1's sharpened form.

Follow-up for the auditor. (1) Re-fetch every PMID and check claim-to-record alignment, with particular attention to the numbers in the new comparison tables in treatment.md, nosology-and-diagnostic-criteria.md, guidelines.md, and STATISTICS.md. (2) Two published abstracts contain internally inconsistent figures that I handled by omission rather than correction — the Cochrane ginkgo dementia global-clinical-status estimate (point estimate outside its stated interval, PMID 41641880) and the Li 2025 Ginkgo MoCA interval, which is implausibly narrow (PMID 41404461); the latter is cited with an explicit caveat, the former is not cited. Confirm both decisions against the full texts. (3) Verify the ClinicalTrials.gov table by re-querying, since registry status changes. (4) Consider promoting pages to curated only after the numeric spot-checks pass.

2026-08-31 — Independent audit (Grok auditor; Codex author)¶

Audited all 19 wiki pages and the literature layer. Did not treat the Codex build as trustworthy. Re-fetched 80 original PMIDs (all resolved) plus 41 additional PMIDs added during audit (121 distinct PMIDs in the condition after expansion). Re-fetched 12 original NCT IDs (all resolved) plus 5 landmark NCT IDs (SPS3 NCT00059306, SPRINT NCT01206062, FINGER NCT01041989, ASPREE NCT01038583, LACI-2 NCT03451591). Checked claim–citation pairs on every page for existence, author/year/journal match, direction, and quoted numbers.

Claim-citation pairs checked: every inline PMID/NCT on the 19 wiki pages, OPEN-QUESTIONS, bibliography, notes, guideline registry, statistics, and patient-voice files (well over 200 pairs). Numeric claims that matched their abstracts include Schneider 38.0/30.0/12% and OR 2.8; Pendlebury 14.4/9.1/7.4/41.3%/3.0% per year and 93% variance; Barbay 53.4/36.4/16.5%; Hughes OR 0.93 and ARR 0.39%; SPRINT MIND HR 0.83/0.81; Peters OR 0.87 and 10/4 mm Hg; Battle ADAS-Cog MDs and AE ORs; Boston v2.0 sensitivity/specificity table; Freitas AUCs; FINGER 0.022 and 7% vs 1% AE; Wilkinson −1.65/−2.09 and withdrawal rates; Román −1.156; Auchus −1.8 vs −0.3; Mok 30% vs 15%; van Middelaar nine trials RR 0.93 (0.84–1.02).

Errors found and fixed - Systematic wrong-journal metadata (PMID existed; journal/title often did not): 40955506 JAMA Neurol not Stroke; 39822128 Alzheimers Dement not Can J Neurol Sci; 35969390 JAMA Neurol not Stroke; 35699195 JAHA not Neurology; 29927845 J Hypertens not Hypertension; 30452613 Fam Pract not BMC Health Serv Res; 31050033 Int J Geriatr Psychiatry not Int Psychogeriatr; 32637417 Front Med not Clin Rehabil; 35797006 Int J Stroke not Alzheimers Dement; 35914668 Neuroimage not Neuroimage Clin; 37405677 J Thromb Thrombolysis not Age Ageing; 37539725 Alzheimers Dement not Ageing Res Rev; 38226361 Cereb Circ Cogn Behav not Br J Occup Ther; 38415688 JCBFM not Ageing Res Rev; 39666076 J Neurol not Front Aging Neurosci; 40156276 Alzheimers Dement not Neurology; 19300631 Neuropsychiatr Dis Treat not Acta Neurol Scand; 20851424 J Neurol Sci not J Neurol; 11794448 J Geriatr Psychiatry Neurol not Neurol Clin; 31074290 Aging Ment Health not Int J Geriatr Psychiatry; 34786789 J Clin Nurs not J Adv Nurs; 32813338 JBI Evid Synth not Dementia. - Filler 2024 (PMID 38101426): RR 3.10 (2.77–3.47) is for baseline cognitive impairment → post-stroke dementia, not a combined “acute impairment/dementia” endpoint; PSCI RR is 2.00 (1.66–2.40). Corrected on wiki pages and STATISTICS.md. - NCT02993367 is phase 2/3, not phase 2. - Hughes primary analysis is 12 trials/92,135, not the 14-trial eligible set, for the OR 0.93. - All six [unverified] flags resolved: CAA-ri immunosuppression (Regenhardt 2020, PMID 32568365); ARIA-E 12.6% lecanemab / 24.0% donanemab (van Dyck 2023, PMID 36449413; Sims 2023, PMID 37459141); CADASIL thrombolysis remains case-level (Pescini 2023, PMID 36255541); COL4A1 haemorrhage associations with trauma/anticoagulation in 52 carriers (Lanfranconi 2010, PMID 20558831) with no pregnancy-trial magnitude; Fabry organ therapy exists but no RCT of cognitive-dementia prevention as of the 2026-08-31 search; psychotropic death/stroke signals from dementia-wide RCTs (Schneider 2005, PMID 16234500; Herrmann 2005, PMID 15697324). Regulator labels were not independently retrieved; dated as 2011 AHA/ASA “no FDA-approved VCI drug” plus 2021 Cochrane “no recommended pharmacological treatment.”

Absence searches re-run (2026-08-31): licensed/approved VCI-specific drugs; aspirin for covert SVD/WMH; CAA-ri treatment; ARIA magnitudes; COL4A1 pregnancy; Fabry ERT cognition; CADASIL thrombolysis; antipsychotics in VaD; SPS3/LACI-2/PRESERVE/PROGRESS/VITATOPS/ASPREE/preDIVA. Covert-SVD antiplatelet “no” remains supported by ESO 2021 (PMID 34414301) and ASPREE (PMID 32213642). No VCI-specific drug-approval trial overturning Battle 2021 was found.

Expansion: added live-fetched landmark papers (NINDS-AIREN, ADDTC, NINDS-CSN, Wolters, Debette, Pantoni, Wardlaw 2019, Iadecola, Snowdon, Esiri, Kalaria, Joutel, LACI-2, PASTIS, SPS3, PROGRESS, VITATOPS, preDIVA, ASPREE, SPRINT MRI, PRESERVE, CAA-ri, ARIA trials). Patient-organization sites re-fetched: original five remained HTTP 200; alz.org and stroke.org 200; stroke.org.uk still 403.

Promotion: all 19 wiki pages set to status: curated because every cited record resolved and matched after correction, with no remaining unverified markers. CONDITIONS-ROADMAP.md set to audited.

Remaining limits (not unmarked errors): many pages are 112–155 lines with 11–30 unique PMIDs — comparable to the Alzheimer's reference condition, below the 150–400 / 25–60 playbook ceiling. Chen 2024 CADASIL thrombolysis is a research letter. Fabry cognitive-prevention and COL4A1 pregnancy-specific magnitudes remain dated evidence gaps. Page density was expanded but not rewritten to 400 lines.

Searches run: PubMed E-utilities esummary/efetch on all original and new PMIDs; esearch for criteria, epidemiology, SVD, CAA-ri, ARIA, COL4A1, Fabry, CADASIL thrombolysis, antipsychotics, SPS3, LACI-2, PRESERVE, nimodipine, VITATOPS, PROGRESS, ASPREE, preDIVA, SPRINT MRI, PASTIS; ClinicalTrials.gov v2 for 17 NCT IDs and a VCI condition query; organization homepage fetches.

Follow-up: next sweep should watch VasCog-2-WSO implementation, LACI-3 if registered, Boston v2.0 memory-clinic validation, and any vascular-dementia-specific drug label change (retrieve the label, not a review).

2026-08-31 — Full build (Codex author; independent audit pending)¶

Built all 19 canonical wiki pages from the INDEX plan; every page remains status: draft.
Re-verified all eight seed anchors through live PubMed E-utilities and ran topic searches covering criteria, epidemiology, post-stroke cognition, SVD/STRIVE, CAA, inherited disease, mechanisms, cognition, imaging, biomarkers, prevention, treatment, guidelines, safety, and patient experience.
Queried ClinicalTrials.gov v2 live for vascular cognitive impairment and tabulated 12 resolved NCT records with status, phase, and enrollment.
Built the bibliography, five landmark notes, guideline registry, statistics file, and four-file patient-voice layer.
Fetched six organization sites directly: five HTTP 200 entries were retained and one HTTP 403 site was quarantined.
Replaced the seed questions with 20 tiered questions and 10 “Dots not yet connected” junctions.
Updated INDEX counts/reading paths/statuses and changed the roadmap state to built.

Follow-up for independent auditor: Re-fetch every PMID/NCT and verify claim-to-record alignment, full citation metadata, line-density expectations, web organization status, and all [unverified] statements. Several pages prioritize scoped, quantitative synthesis but fall below the nominal 150-line reference target and should be expanded during audit before promotion.

Could not verify: regulator-specific non-approval history for vascular-dementia drugs; detailed Fabry/COL4A1 treatment and pregnancy risks; exact CAA-inflammatory treatment effects; anti-amyloid ARIA risk magnitudes. These were omitted or marked [unverified].

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.
Defined the canonical 19-page list in INDEX.md (the canonical list for every condition after the two founding ones is its own INDEX Pages table, per CONVENTIONS.md §5). The list was designed against the shape of this condition's literature, not copied from another condition.
Ran live PubMed scoping searches and recorded the resolved anchor records in INDEX.md. Every PMID in the seed came from a live query in that session; none was written from memory.
Wrote six seed open questions, each tied to an anchor record, to be replaced by a full tiered agenda plus dots table at build time.

Deliberate scoping decision. Alzheimer's disease and vascular dementia are curated as two conditions with an explicit shared border rather than as one. They have distinct literatures, criteria sets and trial pipelines, but mixed pathology is the common case at autopsy, so each condition carries a dedicated overlap page cross-linking the other, and neither may assert a clean separation the pathology literature does not support.

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.