Reviewed 24 October 2026. Three publications across two research programmes. The first two reuse one cohort. This is a new assessment of older work, not a new-research alert or an independent replication claim.
Longitudinal audiovisual study: 20 participants, 539 observations, nine informant-validated episodes in three people. Structured multi-informant review; baseline comparison used a random 20% sample of participant data with generalised estimating equations. Meaningful vocalisations increased across episodes, with heterogeneous features.
Added assessment: repeated observations are nested within people. Neither nine/539 nor three/20 is a population prevalence estimate. An episode-relative baseline does not establish restored global cognition or a mechanism. Detection, informant judgement, selective capture, clinical fluctuation and medication require detailed review. Blinding, preregistration, quantitative effect sizes and uncertainty were not verified.
Scope: complete PubMed abstract, metadata and disclosure; full publisher retrieval failed and PMC access was challenged. No full-text/data audit. No conflicts declared; funding not verified. Already appears as a metadata source in corpus ledger Batch 169, 2 October 2026. No new-source count claimed. Evidence type: preliminary behavioural observation. Next: assess validation criteria, clinical context and missed-event risk.
02 · REASSESSMENT OF A PREVIOUSLY SURFACED LEAD
Detection models: more segments do not create more participants
The analysis uses 1,283 segments from the same three episode-positive participants in the earlier 20-person study. Individual three-class models fitted better than models requiring common class definitions. A candidate class contained most previously validated episodes. Worse fit under shared classes is an informative constraint on a universal detector.
Added assessment: no independent replication or population classifier is established. Selecting episode-positive people limits evaluation in people without validated episodes. Temporal dependence, model selection and possible reference-label overlap require scrutiny. Held-out accuracy, sensitivity, specificity, uncertainty and external validation were not verified here; this is not a claim that none was performed.
Scope: publisher abstract/data statement and selected PMC introduction/methods, including explicit cohort linkage; later PMC retrieval was challenged. Supplements, code, raw recordings and complete results not audited. Publisher states the research was not preregistered. Data access requires IRB-approved procedures and agreements, not open download. Funding/conflicts unverified. Previously surfaced in user-provided cross-task context; no novelty claimed. Next: test a fixed detector in independent people, including episode-negative controls.
03 · HISTORICAL BACKLOG · PROTOCOL, NOT RESULTS
Caregiver stress and grief: a separate research question
The protocol reports recruiting 545 caregivers through five US registries, with monthly online reports for a year and caregiver outcomes assessed at baseline, six and twelve months. These are caregivers, not 545 objectively confirmed lucid cases. It plans to examine episode reports, predictors and associations with care decisions, burden and grief; no completed outcome effect is supplied by this protocol.
Added assessment: complements the Wisconsin observations through a different reporting design. Recall, digital/registry selection, attrition, expectancy and confounding may affect associations. Caregiver outcomes cannot alone identify neural mechanisms. Preregistration and analysis-plan timing unverified. No participant-level independence from other caregiver studies established.
Scope: PubMed abstract, affiliations and disclosure; BMJ returned 403 and PMC was challenged. No complete protocol/data audit. No conflicts declared; funding unverified. No exact identifier match in inspected ledger, but global novelty is unknown. Empirical results N/A. Next: locate separately published results and compare planned versus reported analyses. This protocol is neither positive nor negative survival evidence.
Coverage, duplication and next rotation
Compared identifiers and selected receipts in Corpus-Rotation-Ledger.md version 241 with available conversation context. Earlier original-study metadata and the already surfaced follow-up are explicitly reassessed. Three papers are not three independent replications. Broader cross-task reports and the live PRISM database were not comprehensively searched.
Checked primary surfaces: PubMed, SAGE, selected PMC text and Wisconsin’s public resource directory. OUP retrieval failed; BMJ full text returned 403; several PMC requests returned challenges. Targeted correction/retraction searches identified no verified notice in this bounded check; no exhaustive absence claim. No new LEAD results were verified. The 16-source rotation and all 155 Atlas families were not scanned.
Review Wisconsin full methods, validation labels and observation timing.
Audit LCA supplements/code and participant-level held-out validation.
Locate LEAD outcome publications and assess cohort overlap with earlier caregiver work.
Next topic rotation: controlled past-life-memory comparisons and counter-explanations, after identifier checks. No recurring task schedule changed.
