Created:INDEX.md (18-row canonical Pages table), wiki/overview.md (draft, 27 verified PMIDs), OPEN-QUESTIONS.md (7 seed questions), and this log.
Searches run (live PubMed E-utilities esearch/esummary): global stroke burden; acute ischemic stroke definition and treatment; NINDS/ECASS alteplase and time-to-treatment meta-analysis; AcT tenecteplase; MR CLEAN, HERMES, DAWN and DEFUSE 3 thrombectomy; stroke-unit care; ICH guideline and INTERACT2; secondary-prevention guideline, PROGRESS, CHANCE/POINT, ARISTOTLE, ECST and DEFENSE-PFO; stroke rehabilitation standards; small-vessel disease; premature stroke; neuroprotection.
Verification: every cited PMID was returned by live PubMed in this session; claims were limited to the retrieved article's scope.
Next: full build queued per CONDITIONS-ROADMAP.md; build remaining canonical pages, literature and patient-voice layers, verify trials against ClinicalTrials.gov, and independently audit claim–citation pairs.
Built all 18 canonical wiki pages: extended overview.md; added classification/workup, epidemiology/burden, penumbra biology, cerebral small-vessel disease, acute ischemic stroke, ICH, SAH, systems of care, secondary prevention, cardioembolic/cryptogenic stroke, rehabilitation, outcomes, guidelines, biomarkers, trial landscape, patient experience, and red-flags pages. All remain status: draft pending the separate audit required by the playbook.
Built the literature layer: 399-record cross-indexed BIBLIOGRAPHY.md; five landmark notes (NINDS alteplase, HERMES, DAWN, INTERACT2, organized stroke-unit care); a 29-record guideline registry with supersession chains, disagreements and watch list; 196-line statistics quick reference; and four patient-voice files.
Rebuilt the research frontier:OPEN-QUESTIONS.md now has 16 Tier 1 questions, eight Tier 2 questions, stable OQ IDs, design implications, and a 12-row “Dots not yet connected” table.
PubMed searches: more than 300 live E-utilities searches across definitions/mimics/imaging; global and regional burden; ischemic biology/collaterals/BBB; thrombolysis/thrombectomy; ICH/SAH; prevention/AF/ESUS/PFO/carotid disease; small-vessel disease; rehabilitation/outcomes; systems/prehospital care; guidelines; biomarkers; and qualitative patient/caregiver research. Final whole-condition ESummary retrieval resolved all 399 distinct PMIDs.
ClinicalTrials.gov: live v2 condition searches covered acute ischemic stroke, ICH, and rehabilitation. All 30 distinct NCT IDs used in the condition were then re-fetched from their individual v2 study endpoints; 30/30 resolved.
Patient voice: 12 organization sites and two public story hubs returned HTTP 200 and were catalogued with access dates. Themes are aggregate, public-source only, and each has at least two sources.
Could not verify through direct site fetch: AHA professional pages, American Stroke Association, Stroke Association UK, National Aphasia Association, and Aphasia Recovery Connection returned HTTP 403; several legacy India/South Africa/SAFE domains failed DNS; guessed story URLs returning 403/404 were not used. These failures are explicitly quarantined in the registry/patient-voice coverage notes; PubMed-indexed AHA/ASA guideline identity remains live-verified through E-utilities.
Validation: required page sections/frontmatter present; wiki page lengths within 150–400 lines; inline/reference PMID sets match; local Markdown links resolve; NCT IDs resolve; no unresolved verification markers remain; no prohibited paths were changed and no Git command was run.
Next: run the build playbook's independent audit in a fresh session, re-check each claim–citation pair and quoted number, fix any substantive mismatch, and only then promote passing pages to curated and roadmap status to audited.
Scope completed: audited all 18 canonical wiki pages and every literature artifact: the 407-record bibliography, five landmark notes, 31-row guideline registry, statistics tables, and all four patient-voice files. Checked 942 wiki-body PMID mentions against the retrieved records, plus 488 page-reference entries and 239 PMID mentions in non-bibliography literature artifacts. All 407 distinct PMIDs resolved through live PubMed E-utilities; six group-authored records omitted from one batch response were re-queried individually and resolved. All 30 distinct NCT IDs (45 mentions) resolved through the ClinicalTrials.gov v2 individual-study endpoint.
Errors found and fixed — 14 correction sites: two repeated DAWN numerical errors (control utility-weighted mRS corrected from 2.1 to 3.4 in acute-ischemic-stroke.md and outcomes-and-prognostication.md); one wrong inline author label (Rensma → Fang for PMID 32715831); ten collective-author/metadata citation entries repaired across page references and BIBLIOGRAPHY.md (ECST, SPS3 blood-pressure and antiplatelet reports, AVERT, CAST, IST, WFNS, and NASCET); and one stale guideline-map assertion corrected after the new AHA/ASA adult rehabilitation guideline was located.
Fresh searches behind absence/gap claims: reran PubMed searches for post-stroke insomnia incidence, blood biomarkers replacing imaging, human no-reflow treatment phenotypes, aSAH external-ventricular-drain weaning, 2025–2026 stroke guidelines, and clinical neuroprotection trials. Converted the insomnia, blood-biomarker, EVD-weaning, and neuroprotection statements into positively stated evidence gaps dated 2026-08-30, adding current effect sizes and boundaries rather than retaining timeless absence claims.
Current evidence added (all live-retrieved): the AHA/ASA 2026 adult rehabilitation guideline (PMID 42657476), the SVIN 2026 distal/medium-vessel thrombectomy focused guideline (PMID 42404835), a 2026 post-stroke insomnia cohort (PMID 41924789), acute-vertigo blood-biomarker study (PMID 41961296), EVD-weaning meta-analysis (PMID 41949294), EnTRIPS post-conditioning RCT (PMID 42267436), phase IIb adjunctive normobaric-hyperoxia RCT (PMID 42403344), and preliminary DSA no-reflow prediction work (PMID 42184179). The bibliography therefore increased from 399 to 407 distinct records and the guideline registry from 29 to 31 rows.
Registry and web verification: every one of the 30 trial rows still matched live status, planned/actual enrollment, and phase. All 14 organization/story-hub URLs and the NICE, Australian living-guideline, and Canadian guideline endpoints returned HTTP 200 on 2026-08-30; access dates were refreshed. Landmark-note PMIDs and DOIs matched PubMed.
Validation passed: 18/18 pages remain within 150–400 lines; required frontmatter/TL;DR/open-questions/related-pages/references sections are present; every page's body PMID set exactly matches its references; all local wiki links target canonical pages; all used PMIDs occur exactly once in the bibliography and every bibliography record has a downstream use; no duplicate PMID entry, unresolved link, verification marker, or substantive citation mismatch remains.
Disposition: promoted all 18 pages to status: curated; updated INDEX.md and CONDITIONS-ROADMAP.md to audited. Nothing remains flagged for citation resolution. Continuing scientific uncertainties are retained as dated open questions or guideline watch-list items, not verification failures.
Disposition: started with 12 uncited dot rows; rewrote 3 (D1, D2, D7) with numerical findings on both sides and retained their IDs; deleted 9. No previously cited dot row was changed (there were none in this table).
D3 deleted: the proposed preference-weighted large-core junction was already addressed by a large-core thrombectomy cost-utility model using seven mRS states and utility weights; its SELECT2-based analysis reported a 1.24 quality-adjusted-life-year gain (PMID 39479933). This is not an untouched literature gap, although patient-specific preferences remain a separate question.
D4 deleted: the proposed ICH-specific coordinated care bundle has been tested directly. INTERACT3 randomized implementation across hospitals and found better six-month mRS distribution with the bundle (common OR for poor outcome 0.86, 95% CI 0.76–0.97; PMID 37245517). Crossref lists an erratum for this article, not a retraction or expression of concern.
D5 deleted: its competing-risk restart rule restates OQ-8's design (MRI hemorrhage markers plus AF embolic risk), while the named OQ-10 records do not quantify AF burden as asserted by the dot.
D6 deleted: OQ-14 already explicitly calls for caregiver intervention stratified by survivor dependency and cognition, so this is a restatement of that open question, not a separate junction.
D8 deleted: the aphasia side is supported in the wiki, but neither its cited OQs nor the retrieved candidate records establish the quantitative claim that most digital follow-up assumes fluent language. No replacement estimate was verified.
D9 deleted: the proposed combination was already studied in PRESERVE, an intensive-versus-standard blood-pressure randomized trial in patients with lacunar stroke and cerebral small-vessel disease that measured MRI white-matter injury and cognition (PMID 34044580). Further treatment targeting remains open but is not the untested junction stated in the row.
D10 deleted: the condition's Indigenous incidence review supports disparities, but the named OQ-16 candidates address global burden and low/middle-income stroke services, not a quantified absence of Indigenous governance in stroke-system trials. A scientific statement recommending community-led action does not supply that missing trial comparison.
D11 deleted: selecting a no-reflow phenotype and proving target engagement is already OQ-18/OQ-3's proposed design; the cited review does not quantify the asserted neuroprotection-trial failure mechanism as a separate finding.
D12 deleted: a multicenter vocational rehabilitation RCT already used 12-month return to work as its primary endpoint (64.2% versus 59.4%, adjusted OR 1.12, 95% CI 0.75–1.68; PMID 39614629). The original blanket claim about trials using only 90-day mRS was too broad, and the remaining outcome question repeats OQ-23.
Searches behind retained scoped absences (live PubMed E-utilities esearch, 2026-10-07): D1: stroke[Title/Abstract] AND collateral*[Title/Abstract] AND (transfer[Title/Abstract] OR transport[Title/Abstract]) AND (augmentation[Title/Abstract] OR support[Title/Abstract] OR intervention[Title/Abstract]) → 5 records; all five abstracts were read, including observational transfer/imaging-decay work, and none tested collateral-support treatment during interhospital transfer. D2: thrombectomy[Title/Abstract] AND ("intra-arterial thrombolysis"[Title/Abstract] OR "intraarterial thrombolysis"[Title/Abstract] OR "intra-arterial lysis"[Title/Abstract]) AND ("blood-brain barrier"[Title/Abstract] OR permeability[Title/Abstract] OR MMP-9[Title/Abstract]) → 1 record, an imaging study without biomarker-guided adjunct lysis. D7: stroke[Title/Abstract] AND fatigue[Title/Abstract] AND (rehabilitation[Title/Abstract] OR therapy[Title/Abstract]) AND (dose[Title/Abstract] OR schedul*[Title/Abstract]) AND (work[Title/Abstract] OR vocational[Title/Abstract]) → 5 records; all five abstracts were read, including fatigue-tailored walking but none with a sustained-work outcome. These are PubMed-scoped findings, not claims of absence across all databases or trial registries.
Integrity and limits: every PMID newly written in the table or this entry was fetched and read through live PubMed E-utilities; PubMed publication types were checked for Retracted Publication and Expression of Concern, and each article's own DOI was checked through Crossref updated-by. No cited record carried either flag. DOI extraction was corrected during the audit to use the article's own PubmedData/ArticleIdList rather than DOI IDs embedded in its references. D8's digital-access estimate and D10's stroke-system governance denominator could not be resolved and were removed rather than inferred. Next pass should revisit the three retained scoped gaps with newly published trials and broader registries before making any stronger absence claim.
2026-10-07 — Independent audit of the uncited-dot grounding pass (auditor: Grok; grounder: Codex)¶
Scope: Codex's 2026-10-07 entry was the pass under audit. The public copy of the table still had the original 12 uncited rows (D1–D12). Codex rewrote D1, D2, and D7 and deleted D3–D6 and D8–D12. No row in this table carried a PMID or NCT ID before that pass. This audit re-fetched every PMID Codex kept or used to justify a deletion, re-read the abstracts, checked PubMed publication types for Retracted Publication and Expression of Concern, and checked each article DOI in Crossref updated-by. Absence queries were replayed in PubMed and, as title-and-abstract searches, in Europe PMC. Narrow ClinicalTrials.gov v2 query.term searches were run for the three junctions. WHO ICTRP was not queried: SOURCES.md marks that portal as JavaScript-driven and programmatically unproven. Cochrane CENTRAL, SciELO, Embase, and Scopus were not searched.
Rows checked: 12 (3 rewritten, 9 deleted). After this audit the table has 4 rows: D1, D2, and D7 repaired, D10 restored. No row was deleted by the auditor.
Citations that failed verification: 0 invented, mismatched, or retracted PMIDs. Every record Codex named exists, is the paper the citation implies, and matches the headline number once the population and interval are restored. Crossref updated-by was empty for all of them except INTERACT3.
Motivated anchoring: 2 rows. D1 turned an observational adjusted β into "shortened it by an adjusted 27.6 minutes" and dropped the interval (−51.2 to −3.9). The paper and the direction are right; the causal verb and the missing interval are not. D7 treated an implications sentence in the HiWalk acceptability paper (therapists tailored to fatigue) as a result, and its absence claim missed RETAKE, whose intervention description includes fatigue management and whose primary endpoint is 12-month return to work. D2 was a precision repair rather than a false citation: the MMP-9 diagnostic odds ratio and the POST-TNK percentages match, but "across 30 studies" overstated how many articles entered the MMP-9 pool, and the symptomatic-hemorrhage contrast was quoted as raw percentages without its adjusted risk ratio.
Numerical and population repairs:
D1. Ospel (PMID 34493575) does report 409 patients, median growth 4.74 mL/h (IQR 1.25–14.84), and collateral β −0.81 (95% CI −1.20 to −0.41). The sample is ESCAPE-NA1 patients with last known well under 6 hours and near-complete reperfusion. The negative collateral coefficient has the same sign as ASPECTS and the opposite sign to NIHSS and glucose in that model, which is the basis for reading it as better collateral status with slower growth. van Meenen (PMID 34512538) does report 198 transferred patients, median call-to-comprehensive-center time 162 minutes (IQR 137–190), and adjusted β −27.6 minutes (95% CI −51.2 to −3.9) for highest-urgency dispatch. The row now states that interval and calls it an association. One region is not a geography comparison; Codex had already dropped the original "varies by geography" wording, which the paper does not support.
D2. Krishnamoorthy (PMID 34569521) reports the MMP-9 pooled diagnostic odds ratio 29.571 (95% CI 17.750–49.267) and, in the same sentence, the ≥140 ng/mL threshold, a 29.5-fold symptomatic-hemorrhagic-transformation risk, and AUC 0.881. The 30 articles are the eligible biomarker set; MMP-9 is one of five meta-analyses inside it. POST-TNK (PMID 39804681) matches 49.1% (132/269) versus 44.1% (119/270), adjusted RR 1.15 (95% CI 0.97–1.36; P=0.11), and symptomatic intracranial hemorrhage 6.3% versus 4.4% (adjusted RR 1.43, 95% CI 0.68–2.99). The trial excluded prior intravenous thrombolysis. The row now says so.
D7. Chen's review (PMID 37674875) matches 5 RCTs, 626 participants, search through 4 January 2023, 3 trials with mood and fatigue components, and 2 of those reporting a higher return-to-work percentage. HiWalk (PMID 42803430) matches 91% (SD 13) attendance. Fatigue as a barrier is a result; therapist tailoring is in the implications text, and the row now labels it that way. Work was not an outcome of that nested study.
Absences:
D1, not stale after the boundary papers were named. PubMed replay of Codex's query returned the same 5 records (PMIDs: 41568583, 40985098, 40035294, 38189824, 31426729). Abstracts were read. Two are transfer-imaging observations (ASPECTS decay; hyperacute growth used to think about transfer). One is post-thrombectomy autoregulation and matched because "transfer function" is the autoregulation method. One is a Hungarian narrative review. One is a retrospective AI transfer-decision simulation. Europe PMC title-and-abstract search returned the same 5. A broader PubMed transfer-and-trial query returned 10, still with no collateral-support treatment. Seners (PMID 36748945) already measured growth during transfer: fast progression in 71% with hypoperfusion intensity ratio ≥0.40 versus 3% below 0.40 (n=233). RESIST (PMID 37787796) tested prehospital limb conditioning in 902 target-diagnosis patients (OR 0.95, 95% CI 0.75–1.20). The row now names both and keeps the narrower gap: a collateral-directed treatment during interfacility transfer. ClinicalTrials.gov query.term "stroke collateral transfer interhospital intervention" returned 0. That zero is the narrow query, not a registry-wide absence.
D2, scoped absence held; the selection literature was incomplete. PubMed replay returned the same 1 record (PMID 24948498), a flat-panel CT study, also the only Europe PMC title-and-abstract hit. A PubMed query for MMP-9 or blood-brain barrier with intra-arterial alteplase or tenecteplase and selection or prediction returned 0. ClinicalTrials.gov query.term "intra-arterial thrombolysis thrombectomy biomarker blood-brain barrier" returned 0. The row was stale only in omitting Chen's 7-trial meta-analysis (PMID 40506219): mRS 0–1 OR 1.44 (95% CI 1.21–1.72), symptomatic hemorrhage OR 1.15 (95% CI 0.75–1.75), with signals by dose, eTICI, NIHSS, and etiology. Those are clinical subgroups. The marker-guided trial is still unfound in the queries above. No retraction or expression of concern on PMID 40506219 or PMID 24948498.
D7, stale. The 5-record replay matched (PMIDs: 42803430, 37486748, 32876580, 25920538, 2035427). The last is a 1991 milrinone heart-failure trial, a query false friend. Europe PMC title-and-abstract search returned 4 of the 5 and did not surface a new dose-scheduling trial. A broader PubMed fatigue-and-vocational query returned 13, including the RETAKE intervention description (PMID 41930667) and POSITIF (PMID 35866206). RETAKE's primary result is 12-month return to work, 64.2% versus 59.4%, adjusted OR 1.12 (95% CI 0.75–1.68) (PMID 39614629), and the development paper lists fatigue management among components added after piloting. POSITIF collected return to work after a telephone fatigue intervention and reported no between-group difference, without giving the percentages. The repaired junction is narrower: rehabilitation dose changed in response to fatigue, with sustained employment as the outcome. ClinicalTrials.gov query.term "stroke fatigue vocational return to work rehabilitation dose" returned 0.
Deletions upheld:
D3. Gao's cost-effectiveness model (PMID 39479933) uses a 7-state mRS Markov model, SELECT2 probabilities at 3 months, and published utility weights, and reports a mean gain of 1.24 QALYs. That is a preference-weighted large-core result. The utilities were taken from the literature, which is the limit Codex already noted. Not retracted. Deletion stands.
D4. INTERACT3 (PMID 37245517) is a stepped-wedge cluster RCT of an ICH care bundle and reports a common odds ratio for poor outcome of 0.86 (95% CI 0.76–0.97). Crossref updated-by is an erratum (DOI 10.1016/S0140-6736(23)01420-4; PMID 37453751, publication type Published Erratum). The Europe PMC full text of that erratum corrects the spelling of two appendix investigator names and does not change the outcome. Deletion stands.
D5. The proposed competing-risk restart rule repeats the design already written in OQ-8. The OQ-10 candidates are a TOAST classification paper and a cryptogenic-stroke review; they do not quantify AF burden. A nearby cohort (PMID 32829908; n=65, 35 with possible or probable CAA) found ICH recurrence of 3.2 per 100 patient-years against a HAS-BLED expectation of 1.9, and no new integrated rule. Not retracted. Deletion stands. This was a restatement, not a failure to search.
D6. OQ-14 already asks for a caregiver trial stratified by survivor dependency and cognition, and already states the pooled neutral result. Deletion as a restatement stands.
D8. The "most digital follow-up assumes fluent language" claim still has no denominator. A small internet-use survey (PMID 30982360; 25 people with aphasia, 17 without) found aphasia associated with difficulty in technology-based written communication and found age the stronger predictor of internet use. It does not measure clinical follow-up systems. Not retracted. The deletion was an inability to verify a quantitative claim, and the row was not groundable from what this session retrieved. It stays deleted.
D9. PRESERVE (PMID 34044580) randomized 111 people with symptomatic lacunar infarct and confluent white-matter hyperintensities to standard (n=56) or intensive (n=55) blood-pressure targets. The primary endpoint was diffusion-tensor white-matter mean diffusivity at 24 months (P=0.92). Cognition was a secondary endpoint. PubMed's abstract is the graphical-abstract placeholder; the design and null primary result are in the Crossref abstract. Not retracted. The junction, a blood-pressure trial aimed at small-vessel imaging progression and cognition, has been done. Deletion stands.
D11. "Enroll only a validated no-reflow phenotype and prove target engagement" is the design already stated in OQ-3 and OQ-18. Deletion as a restatement stands.
D12. RETAKE (PMID 39614629) uses 12-month return to work as its primary endpoint, so the blanket claim that trials center on 90-day mRS is false. A five-year early-supported-discharge abstract (PMID 35988246) reports predictors (social problems OR 0.02, 95% CI 0.00–0.22; breadwinner OR 13.79, 95% CI 2.46–77.52) and uses "sixty percent (n=49)" for the analysis subsample, not as a return-to-work rate. No verified longitudinal rate was available to rebuild the row. Deletion stands. Both papers were checked; neither is retracted.
D10 restored. Codex's stated reason was that the incidence review and the World Stroke Organization statement do not quantify an absence of Indigenous governance in stroke-system trials. That reason is right about those two documents and wrong as a decision to drop the row. Balabanski (PMID 38354325) supplies the incidence ratios and the CONSIDER reporting gap. Healing Right Way (PMID 41131785) is a published stepped-wedge cluster RCT in Aboriginal adults with stroke or traumatic brain injury: 108 enrolled of a target 312, no EQ-VAS difference (p=0.83), processes of care higher by an adjusted 6.8% (95% CI 0.40%–13.26%). It is post-injury cultural security and coordination, with a standard quality-of-life outcome. A PubMed search for an acute thrombectomy, stroke-unit, prehospital, or systems trial in Indigenous, Aboriginal, or First Nations populations returned 0. The restored row states those two findings and does not repeat the unquantified "rarely." Both papers, and the statement (PMID 40415682), had no retraction flag and no Crossref updated-by entry. ClinicalTrials.gov query.term "stroke Indigenous Aboriginal governance" returned 0 on the same narrow basis as the other registry queries.
Dot identity: D1, D2, D7, and D10 each pair different findings. None is the same claim in two wordings. D5, D6, and D11 remain deleted because they restated an open question's own design.
Novelty-gate judgement: The four-row table can be used as scoped, sourced leads. It cannot be treated as proof that nothing exists outside the named queries. D7's original absence was stale, and D10 had been dropped even though Healing Right Way was retrievable. The gate should re-run the searches, including registries this session did not cover, before converting any of these junctions into a claim that the literature is empty.
2026-10-20 — Periodic literature sweep (author: Claude; independent audit pending)¶
Window and scope. The window is publication dates 2026-08-28 to 2026-10-20. This is the condition's first periodic sweep since the 2026-08-30 audit. I read CONVENTIONS.md, CLAUDE.md, SOURCES.md, this log, INDEX.md and OPEN-QUESTIONS.md before writing, and changed only stroke files. Queries, translations, returned IDs, registry snapshots and per-record retraction checks are in the search manifest.
Sources actually queried.
PubMed E-utilities. The unfiltered stroke/ICH/SAH window query returned 5,207 records. That set was not screened record by record. Screening ran instead on 15 topic queries restricted to RCT, meta-analysis, systematic-review or guideline publication types (or "trial/randomized" in the title), plus guideline, top-journal, statistics and two Cochrane queries. Together these returned 692 distinct records, all of which were title-screened, with abstracts read for about 40.
Targeted PubMed follow-ups: factor XIa trials, and an August-inclusive Cochrane query.
Europe PMC: 14 topic queries with FIRST_PDATE in the window. 299 distinct non-PMID records (preprints and PMC-only) were title-screened.
Crossref:updated-by checks without mailto, and one DOI title search. Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL is unset.
ClinicalTrials.gov v2: all 30 cited NCT records plus 8 registrations of new trials.
Web probes: guideline sites and the 14 patient-organisation URLs.
WHO ICTRP and ISRCTN returned JavaScript or cookie shells with no usable results. The EU CTR returned an empty response. ChiCTR, CTRI and CTIS were not queried. SciELO/LILACS and Cochrane CENTRAL could not be searched (HTTP 403). Every absence below is therefore an absence in the named sources only.
Cochrane priority.
One new in-window stroke review: Cardiorespiratory training for people with stroke (Kramer 2026, PMID 42742033). It is one of three reviews replacing Saunders 2020, which this condition cites. It does not contradict the wiki, but it sharpens it: high-certainty null effects on death and on recurrent vascular events, and fitness and walking-speed gains below the review's minimal clinically important differences.
Two August reviews published after the 2026-08-30 audit's searches were caught up: brain–computer interface training (PMID 42626977) and swallowing therapy (PMID 42573047).
The in-window carotid imaging-model record (PMID 42713886) is a protocol and was not used.
Most important finding: a pre-window gap. OCEANIC-STROKE (NEJM, April 2026; PMID 41985132) was absent from the condition. It is the first positive phase 3 trial of factor XIa inhibition added to antiplatelet therapy after non-cardioembolic stroke. It surfaced only because an in-window meta-analysis (PMID 42748394) and an in-window bleeding analysis (PMID 42866879) are driven by it. All three were added to secondary prevention, with the meta-analysis's own caveat that it is not independent replication.
Survived triage: wiki edits on six pages, all demoted to draft with last-curated 2026-10-20.
secondary-prevention.md: OCEANIC-STROKE efficacy and bleeding (the lower-GI excess included); FXIa meta-analysis; RECAP-ICH individual-patient meta-analysis (HR 0.62 for recurrent stroke); one new open question.
intracerebral-hemorrhage.md: 2026 NCS/SCCM focused update (conditional preference for 4F-PCC over andexanet; platelet-transfusion recommendations); RECAP-ICH pointer.
acute-ischemic-stroke.md: ATTENTION LATE (late-window tenecteplase before basilar EVT: 30% vs 30%, neutral); new "Medium and distal vessel occlusion" subsection with DISCOUNT (stopped for futility and excess hemorrhage; as-treated sICH 11% vs 3%), tied to the existing SVIN guideline; one new open question.
clinical-trials-landscape.md: TL;DR qualified; new subsection on LAIS (loberamisal; mRS 0–1 RR 1.24) and ERASE-STROKE (oral edaravone; OR 2.12), both flagged as unreplicated single-country trials; STRATEGY (tirofiban in branch atheromatous disease, neutral) and the colchicine-after-ICH pilot (feasibility only) added; four registry rows updated (next bullet).
rehabilitation-and-recovery.md: cardiorespiratory Cochrane appended to the Saunders paragraph; BCI row added to the upper-limb table; new short "Swallowing" section from the dysphagia Cochrane, with certainty grading foregrounded.
subarachnoid-hemorrhage.md: unruptured-aneurysm prevalence meta-analysis (3.9% reference prevalence, rising to 6.6% in 2016–2022 on angiographic imaging); 2026 NCS seizure-prophylaxis guideline set against AHA/ASA 2023; one new open question.
ClinicalTrials.gov changes since the 2026-08-30 audit. These are registry facts, not results:
NCT04667078 (P2Y12 plus EVT): recruiting → suspended. The registry states futility at a prespecified interim analysis.
NCT06437431 (glenzocimab, large core): → withdrawn. The sponsor ceased activities; 0 enrolled.
NCT06146790 (distal medium-vessel EVT): → unknown, last verified 2024-08.
NCT05413733 (internet neuropsychological rehabilitation): → completed, with 21 actual against 150 planned and no posted results.
The other 26 rows matched their wiki status. No cited trial newly posted results.
New-trial registrations matched their publications: DISCOUNT NCT05030142 (registry "unknown", estimated 488); ATTENTION LATE NCT05701956; LAIS NCT06517173; OCEANIC-STROKE NCT05686070; ERASE-STROKE NCT06648304; DETERMINE NCT04352296; colchicine pilot NCT05159219. These NCT IDs were verified but are cited in the wiki through PMIDs, not as NCT rows.
Guidelines, statistics, patient organisations.
Guideline registry: 31 → 38 rows: NCS/SCCM antithrombotic-associated ICH; NCS aSAH seizure prophylaxis; Chinese ICAS guideline; WSO/APSO/PAVIN ICAS consensus; Spanish palliative-care guideline; AHA health-literacy statement and WSO vascular-dementia statement (both marked adjacent). Two new documented disagreements were added (factor-Xa reversal; aSAH seizure prophylaxis). Albers et al.'s Proposed Corrections to the 2026 AIS guideline (PMID 42535283) is recorded as existing only; it has no abstract and was not read, so its content is not characterized.
STATISTICS.md: new §11a (aneurysm and small-vessel-disease marker prevalence) and a caveat row from the GBD-versus-register concordance study (Youkee 2026, PMID 41631666; only 22% of GBD 2021 incidence estimates matched registers). No new global GBD 2023 all-stroke paper was found in PubMed in the window; only regional or risk-factor GBD secondary analyses, which were not added.
Patient organisations: all 14 URLs returned HTTP 200 (the 2026-08-30 failures were not recurrent). The WSO homepage gave the 2026 World Stroke Day theme (symptom awareness, #ActFAST) and announced the World Stroke Congress for 21–23 October in Seoul; a dated note was added to organizations.md.
OPEN-QUESTIONS.md: OQ-1 sharpened by DISCOUNT; OQ-12 widened by OCEANIC-STROKE; new OQ-25 asks whether the 2026 neuroprotection results replicate. INDEX.md and BIBLIOGRAPHY.md updated (407 → 431 records; 24 added, each with live-computed cited-by; Majidi's cited-by refreshed).
Full-text honesty (FAILURE-MODES rule). Only three new records could be read in full: STRATEGY, the colchicine pilot and the NCS seizure guideline. Their numbers were checked against the Europe PMC full text. All other numbers are from PubMed abstracts and are labelled "abstract-level" inline. JAMA, doi.org→JAMA and the Cochrane Library returned HTTP 403; the PMC deposits for the JAMA and Cochrane records were metadata-only, and Europe PMC fullTextXML returned HTTP 500 for them. No claim that a paper omits something was made.
Citation controls. All 24 added PMIDs, plus the screened-but-not-added SINGLE-AF record, were checked through PubMed publication types and CommentsCorrections, and through Crossref updated-by. None carried a retraction, expression of concern or erratum. First-author initials, volumes and pages were taken from the live efetch, not typed from memory.
Screened and withheld.
SINGLE-AF (NEJM; PMID 42663303; AF primary prevention, outside this condition's after-stroke scope).
DETERMINE (neutral individualized BP during EVT; PMID 42585603), deferred to the next sweep for a full-text check.
ANGEL-TNK and TESLA one-year reports: no abstract, full text 403.
A 62-patient PCSK9 ICAS surrogate trial; a 121-patient magnesium-in-aSAH RCT; the GBD pediatric-burden paper; many single-centre rehabilitation RCTs.
A second phase III tirofiban trial in branch atheromatous disease (Europe PMC PMC13676718, Lancet Reg Health West Pac): no PMID, no DOI in Europe PMC, no Crossref match. Not citable yet.
Pre-existing issues noticed, not fixed (outside this sweep's remit; flagged for the auditor).
12 PMIDs used in OPEN-QUESTIONS.md by the 2026-10-07 dot-grounding pass are not in BIBLIOGRAPHY.md (e.g., 34493575, 39614629, 41131785).
INDEX.md still counts 12 "Dots-not-yet-connected" junctions, where the table now has 4.
clinical-trials-landscape.md's reference list skips number 4.
Several bibliography cited-by fields list INDEX.md or carry inline note text.
Validation. All six edited pages: every body PMID appears in References and vice versa; 167–189 lines each; frontmatter reads status: draft and last-curated: 2026-10-20. All relative Markdown links in the condition resolve. The manifest sits at the one permitted path, and tools/build_public_site.py maps literature/sweeps/*.json. No strict build or publish was run. No page was promoted by its author.
Next sweep / auditor.
(1) Retrieve full texts for LAIS, ERASE-STROKE, DISCOUNT, ATTENTION LATE, OCEANIC-STROKE, RECAP-ICH and the cardiorespiratory Cochrane, and check every abstract-level number against its table.
(2) Chase World Stroke Congress (21–23 Oct 2026) late-breakers, the Albers proposed-corrections content, and any AHA/ASA response.
(3) Watch for publication of the NCT04667078 futility result, and for a PMID/DOI for the Lancet Reg Health West Pac tirofiban trial.
(4) Re-check whether OCEANIC-STROKE prompts an AHA/ASA or ESO secondary-prevention update.
(5) Appraise DETERMINE and the ANGEL-TNK/TESLA one-year reports.
(6) Retry ChiCTR (both 2026 neuroprotection trials are Chinese) and the modern EU CTIS.
(7) Repair the pre-existing bibliography/INDEX inconsistencies listed above.
2026-10-27 — Periodic literature sweep (author: Codex; independent audit pending)¶
Window and scope. Read CONVENTIONS.md, CLAUDE.md, SOURCES.md, the prior log, INDEX.md and OPEN-QUESTIONS.md. Repeated the requested publication window 2026-08-28 through 2026-10-27, rather than shortening it to the 20 October sweep. Changes are confined to stroke. Exact queries, PubMed translations, returned IDs, trial snapshots, full-text table locators and retraction checks are in the search manifest, at the prescribed published path.
PubMed E-utilities actually queried. The broad stroke/ICH/SAH publication-date query returned 5,771 IDs; this broad set was not screened record by record. Nineteen prioritized queries covered classification/workup, burden, penumbra, small-vessel disease, thrombolysis, thrombectomy, ICH, SAH, systems, prevention, cardioembolism, rehabilitation, outcomes, biomarkers, patient experience, guidelines, major journals and statistics. Together with the Cochrane queries they returned 756 distinct IDs; 755 journal records were parsed through efetch. One non-PubmedArticle record was not used. The 65 IDs newly returned versus the previous sweep's prioritized queries were title-screened, alongside all Cochrane records and the major-journal shortlist; abstracts were inspected selectively. Additional E-utilities searches chased ENRICH AF, OPTEC, BRANT and guideline corrections; the narrowly scoped in-window ENRICH-AF and OPTEC queries returned zero, which does not exclude congress reports. Previously uncatalogued citations were fetched separately for inventory repair, without changing their scientific claims.
Cochrane first. Seventeen CDSR queries included an individual check for each of the 15 major topic groups and two broad stroke queries (publication date and indexing date). Five distinct records were returned, the same five already found on 20 October: the cardiorespiratory-training review, carotid imaging-model protocol, and less condition-specific hormone-therapy, community-falls and sickle-cell transfusion reviews. No additional stroke-specific Cochrane review survived triage. CENTRAL and Cochrane Library full text were not searched (HTTP 403 access limitation); PubMed indexing was the review-search route.
Europe PMC actually queried. Fourteen topic queries were retrieved in full, including 308 distinct records without a PMID field; the non-PMID trial/review/guideline title shortlist was inspected. Three supplementary searches covered workup, penumbra and guidelines; the broad workup and guideline supplements were capped at 1,000 records and were not exhaustively screened. Europe PMC's authoritative availability metadata was checked for the selected new reports and prior full-text follow-ups. The manifest distinguishes returned records, screening and full-text reading; retrieval is not appraisal. Identifiers otherwise appearing only in Europe PMC were additionally resolved through live PubMed ESummary before publication of the manifest.
Crossref and retraction controls. All five newly retained records and twelve bibliography repairs were checked against live PubMed publication types/CommentsCorrections and Crossref updated-by. None had a retraction or expression-of-concern flag. One Crossref 429 response was retried successfully before adding that record. The original rehabilitation guideline's Crossref record does have a correction, dated 26 October, and was treated as such. Crossref was queried without mailto; Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL is unset. OpenAlex and DOAJ were not queried.
Important resolved follow-up: BRANT. The previously unresolved Europe PMC-only phase III tirofiban record now has a matching Crossref DOI and a live-resolved PubMed identifier (Li 2026, PMID 42894564). Europe PMC search metadata still omitted the identifier fields; the full-text XML supplied the identifier, then live PubMed efetch verified it. The exact-phrase PubMed search itself returned zero, illustrating why a search miss must not be called non-indexing. Full Methods, Results and Discussion were read, and Table 2 checked: mRS 0–1 was 181/258 versus 178/255, adjusted RR 1.00 (95% CI 0.89–1.12); available-case major bleeding was 0/259 versus 1/245. Added a concise trial-landscape entry, preserving the absence of concomitant oral antiplatelets during infusion, exclusion of thrombolysis/EVT, and the distinction between neutral superiority and equivalence. Its registration resolved live (NCT06037889); 516 registered actual participants and 513 in the paper's full-analysis set are different denominators, explained by three consent withdrawals.
Two important new publications, metadata alerts only. LATE-MT (Yang 2026, PMID 42929656) and MILD-MT (Liu 2026, PMID 42929671) were published on 23 October. Both registrations were fetched live (NCT05326932; NCT06179017): completed, no posted results. Europe PMC lists subscription access; both NEJM full-text requests returned HTTP 403. Added publication/selection alerts to acute ischemic stroke and the trial landscape, and new OQ-26. Their abstract outcome estimates were not copied into the synthesis. No landmark note was created before full-text appraisal.
Caregiver synthesis sharpens, rather than closes, OQ-14. Read Lin et al.'s full text (PMID 42909033), including Table 2. Its remote-delivery subgroup reports seven studies/612 caregivers, SMD 0.47 (95% CI 0.18–0.75), I² 64.45%, with positive values favoring burden reduction. The delivery-mode interaction is not significant (P=0.556). Added this alongside the earlier neutral syntheses in the patient-experience page, statistics table and OQ-14; explicitly retained heterogeneity, effect-direction and trial-set reconciliation, and the exploratory nature of comparisons between delivery modes. This is not proof of the best intervention package or a prevalence estimate.
ClinicalTrials.gov v2 sweep. All 37 previously cited distinct IDs, including seven used only in the prior log rather than the original 30-row table, were re-fetched from individual study endpoints; the three new registrations above bring this session to 40. PROTECT-2 (NCT05789823) changed from recruiting to completed, with 160 actual participants (update posted 21 October). Its row and surrounding text were updated. No previously cited registration newly posted results; none of the 40 fetched records has posted results. NCT07187687 has a newer update date (20 October) but unchanged status/enrollment in the tracked fields. The other 36 previously cited statuses match the 20 October snapshot. Registry completion is not an efficacy result.
Non-US registry coverage, with actual results and limits. WHO ICTRP now returned usable tables: stroke 139 records/134 trials; stroke AND China 6/6; stroke AND India 3/2; edaravone 4/3; stroke AND rehabilitation 17/17; loberamisal zero in this portal query. First result pages included ChiCTR, CTRI, ACTRN and JPRN identifiers and were inspected; not all pages or historical versions were reviewed. The small, often COVID-related result sets and the failure to retrieve a known recent trial by drug name prevent any claim of complete registry coverage or regional absence. Direct ChiCTR returned a compressed JavaScript challenge, ISRCTN a cookie/JavaScript holding page, CTIS an application shell, and CTRI refused the connection; none supplied usable direct search results. SciELO and LILACS could not be searched (HTTP 403 access limitation); regional transportability remains unresolved. Embase, Scopus and Web of Science were not queried.
Society guidance. Read the ESO directory, Australian living-update page, Canadian homepage and NICE NG128 history, and fetched AHA's guideline directory. Catalogued the newly indexed AHA/ASA rehabilitation erratum (Richards 2026, PMID 42932410) as a companion correction, bringing the registry to 39 rows; the text is inaccessible (AHA HTTP 403), so no changed recommendation is inferred. Added alerts to guidelines and rehabilitation. The Australian dysphagia/weakness proposals are under consultation 24 September–30 October, not finalized recommendations. ESO lists a factor XI inhibitor expedited recommendation in development; OCEANIC-STROKE has not been assumed to have entered adopted guidance. These checks do not establish worldwide guideline completeness.
Registry/burden and patient organisations. The WSO fact-sheet landing page still points to its 2025 publication; SSNAP's public homepage was checked. No new table-verified global burden replacement was added. The newly indexed Indian device-specific thrombectomy registry was screened but its abstract figures were withheld pending full text. Added the caregiver effect-size row described above without replacing burden estimates. Read SAFE's 1 October announcement of a WSO Post-Stroke Checklist implementation survey, closing 30 November; catalogued it in organisations/sources and OQ-15 as implementation mapping, not outcome evidence. Re-read WSO's 2026 #ActFAST campaign page/news: multilingual toolkit and #SpotStroke quiz are available; no campaign-effect estimate was inferred. Only public organisation material was used; no survey response or private patient material was accessed.
Congress and outstanding full texts. Read the official WSC day 1–3 trial recaps and participant/program pages. ENRICH AF, OPTEC, intra-arterial-lysis subgroup reports and RAPID-M are next-search leads; conference summaries were not promoted to numerical wiki evidence. DETERMINE and Albers's proposed guideline corrections still returned publisher HTTP 403. Europe PMC still lists subscription-only access for LAIS, ERASE-STROKE, DISCOUNT, ATTENTION LATE, OCEANIC-STROKE and the cardiorespiratory Cochrane. RECAP-ICH, TESLA and ANGEL-TNK long-term reports remain on the full-text appraisal queue; this sweep does not revalidate the previous sweep's abstract-level numbers.
Inventory and status changes. Five new records plus the twelve previously logged omissions bring BIBLIOGRAPHY.md from 431 to 448. All seventeen added lines use session-retrieved metadata and retain topic organization. Corrected the index's stale dot count from 12 to 4 and the trial page's reference-numbering gap. Five wiki pages changed; each has last-curated: 2026-10-27 and status: draft. Eight wiki pages are now draft across the October sweeps. Updated the affected literature-layer dates and index audit labels. No author promotion or independent audit is claimed.
Validation. Condition-local checks passed: edited-page frontmatter/sections/lengths, body/reference PMID equality, sequential reference numbers, every maintained-content PMID present once in the bibliography, current index statuses/counts, all relative Markdown targets, ordered guideline rows, and complete dual-source retraction checks for all seventeen added bibliography records. All new NCT identifiers resolve in this session's v2 snapshots. Confirmed that tools/build_public_site.py publishes literature/sweeps/*.json; the manifest is at the exact permitted path. No global build, publish, deploy, git command, .git or .gitignore creation was performed.
Next sweep / independent auditor. (1) Obtain both new NEJM trial reports/protocols and check efficacy, mortality, hemorrhage, selection and analysis-population tables before changing indications or writing landmark notes. (2) Obtain the rehabilitation erratum text and determine affected recommendations; retry the Albers proposed-corrections paper and any AHA response. (3) Check final Australian dysphagia/weakness recommendations after 30 October and the ESO factor XI inhibitor work. (4) Chase citable primary publications/DOIs for ENRICH AF and OPTEC and the other WSC leads. (5) Watch for PROTECT-2 results and the P2Y12-adjunct futility publication. (6) Reconcile caregiver-review trial inclusion, effect orientation and risk-of-bias assessments before interpreting the newer signal as overturning earlier null findings. (7) Retry the previously listed full-text queue and direct non-US registries; seek broader WHO ICTRP retrieval before regional conclusions. (8) Check Post-Stroke Checklist survey reporting after its 30 November close, separating uptake mapping from an intervention trial.