Hypothesis-Generating Brief: Therapeutic plasma exchange

Artifact

Living evidence brief from agent-v3-full-paper-live

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. For each of the seven outcome classes, list the specific directional findings from each source (or note explicitly that no directional signal was extracted for that source) so the 'no extracted directional signal' blanket can be replaced with source-level evidence attribution.
2. Resolve the duplicated/contradictory directional counts under 'Contextual Adjacent Evidence Outcomes' and verify that each subsection's counts sum to the stated n for that class.
3. Enumerate which five sources are classified as 'direct clinical' and which 23 are 'adjacent clinical', and state the criterion used for this classification so the directness counts are auditable.
4. Either remove the geroscience / anti-aging framing from the Evidence Landscape and Abstract, or restrict the longevity framing strictly to Fuentealba 2025 (the single source with biological-age endpoints) and treat the other 27 sources on their own clinical terms.
5. Expand the Tensions and Gaps section to name the specific conflicting findings already visible in the bundle (Fuentealba 2025 positive biological-age signal vs. Espana-Cueto 2025 null PCC trial vs. Salur 2026 null cast-nephropathy result vs. Boada 2020 mixed AD trial), so disagreement is surfaced rather than smoothed away.
6. Cite by name and integrate the remaining ~24 admitted sources into the Findings Map rather than admitting 28 sources and discussing only four.

Major issues

• Findings Map is presented as a table of outcome-class slices with no extracted directional signal in nearly every cell (e.g., 'no extracted directional signal in 10/13 sources', '3/3', '1/1'), so the actual reported findings — what any individual source claims and in what direction — are not surfaced. The 'strongest signal' column is uniformly empty across all seven rows, leaving the map with no mapped findings for a reader to evaluate.
• Several Findings Map subsections give contradictory or garbled numbers that cannot be reconciled with the stated per-class counts (e.g., the second 'Contextual Adjacent Evidence' paragraph gives directional=negative=1, null=10, unclear=2 and directness=direct=2, indirect=9, review=2 — totaling 13 sources — which is internally consistent, but the next paragraph at the same heading restarts with 'mixed=1, null=4, unclear=1, direct=1, indirect=4, review=1' totaling only 6 sources and no heading change; this appears to be either a fragment from another class pasted under the wrong heading or a duplicate block.
• The paper collapses therapeutic plasma exchange into a single 'longevity / anti-aging' framing despite the corpus being overwhelmingly about established clinical indications (TTP, MG, AD, sepsis, COVID-19, cast nephropathy, pediatric neuroimmune disease, etc.). The mapping of every source to a 'geroscience rationale' is an unsupported editorial framing — only Fuentealba 2025 directly targets biological-age endpoints in healthy adults, yet the abstract and Evidence Landscape apply a geroscience lens to the entire corpus.
• Outcome classes do not consistently correspond to the cited sources: 'Longevity' cites no source by name; 'Mortality and Survival' is acknowledged to be empty of retained narrative paragraphs; 'Dosing and Pharmacokinetics' cites only one veterinary case (Yeh 2026) but is framed as animal/preclinical evidence for a human-relevant dosing claim; 'Cardiometabolic' has only one source labeled direct but no narrative is provided.
• Source directness counts (5 direct clinical sources of 28 admitted) are reported in the abstract but the manuscript never enumerates which five sources are the direct ones, leaving the auditability of the directness classification unverified.
• The 'Tensions and Gaps' section is a single generic sentence ('Run adequately powered human studies...') that does not surface the specific contradictions visible in the corpus (e.g., Fuentealba 2025 biological-age improvement vs. Espana-Cueto 2025 null PCC trial vs. Salur 2026 null myeloma cast nephropathy result vs. Boada 2020 ADCS-ADL improvement).

Minor issues

• Several subsection headings (e.g., 'Longevity Outcomes', 'Cardiometabolic Outcomes') contain sub-blocks with no heading change but appear to be mis-copied paragraphs from other outcome classes; structural cleanup is needed.
• The 'Source attribution' check fails: 'Xu 2026' is cited as 'Xu 2026' but its DOI is 10.1002/iid3.70369 and source_type is 'corpus' rather than pubmed, yet it is treated equivalently to peer-reviewed sources in the Immune and Inflammation row; the bundle entry does not provide a normal bibliographic record.
• 'Eichinger 2025' is pediatric and 'Lee 2026' is pediatric, yet both are grouped under Immune and Inflammation with 'limited corpus depth' noted — the heterogeneity of population (pediatric vs adult vs veterinary) within a single outcome class is not surfaced.
• The Search Summary 'source admission funnel' table reports 28 admitted sources but the funnel itself is opaque: 'No extractable claims (24)' and 'None-only claim binding (11)' total 35 candidates from 37 classified, which does not reconcile with the stated 28 admitted sources without an explicit re-inclusion step.
• Abstract states 'high-density pairwise disagreement map' but the manuscript body never shows or quantifies this pairwise disagreement map — readers cannot audit it.
• The Source bundle contains 28 sources but the paper's narrative only cites a handful by name (Kimber 2026, Eichinger 2025, Xu 2026, Yeh 2026), leaving ~24 sources unmentioned in prose and untraceable to specific findings.

Reviewer note

This is an evidence map of therapeutic plasma exchange across 28 sources that, in principle, should be well-suited to the evidence-map format because the underlying literature is heterogeneous and does not collapse to a single positive claim. However, several execution problems prevent the manuscript from functioning as a faithful landscape. First, the Findings Map table has a uniformly empty 'strongest signal' column and reports 'no extracted directional signal' for nearly every cell, so the actual mapped findings — what each source claims and in what direction — are not surfaced. A reader cannot audit what the map is mapping. Second, the Internal Directional Counts in the Contextual Adjacent Evidence subsection are internally contradictory (one paragraph sums to 13 sources, the next to 6, with no heading change), suggesting a copy-paste error or misassigned block. Third, the paper imposes a geroscience / anti-aging framing on a corpus that is overwhelmingly about established clinical indications (TTP, MG, AD, sepsis, COVID-19, cast nephropathy, pediatric neuroimmune disease); only Fuentealba 2025 directly addresses biological-age endpoints in healthy adults. This is an unsupported editorial collapse of a heterogeneous corpus into a single narrative frame. Fourth, the directness classification (5 direct of 28) is reported but never enumerated, so its auditability is limited. Fifth, the Tensions and Gaps section is a single generic sentence that fails to surface the specific, visible contradictions in the corpus (Fuentealba positive vs. Espana-Cueto null vs. Salur null vs. Boada mixed). The Limitations section is reasonably specific and the Limitations on surrogate endpoints, missing geriatric-syndrome reporting, and absence of long-term mortality RCTs are appropriate. Source grounding is partially supported: the cited sources exist and are recent, and reference-only calibration applies, but ~24 of the 28 admitted sources are never mentioned by name in the manuscript body, so the source-attribution check fails for most of the bundle. The manuscript is not structurally broken — it has a coherent scope and a defensible search summary — but it is closer to a revise than an accept because the core deliverable of an evidence map (source-attributed findings, surfaced disagreement) is materially incomplete.

Panel metadata

Models: MiniMax-M3 + google/gemma-4-31b-it + mistralai/mistral-small-2603

Route: consensus

Prompt: reviewer-v11-research-synthesis