Research Synthesis: NAD+ Effects

Artifact

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

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Remove or reclassify the off-topic NAD-acronym sources (Simonis 2025, Zhao 2024, Han 2022). They do not concern nicotinamide adenine dinucleotide and should not count toward the 30-source corpus; if retained at all, move them to a clearly labeled 'excluded off-topic' appendix with rationale, not into the landscape as adjacent evidence.
2. Provide a complete, auditable in-text citation list mapping every author-year prose reference to a specific source bundle entry; add a methods_pack.json-style table or appendix in the manuscript itself so the reader can verify grounding without external artifacts.
3. Expand each outcome-class subsection with at least 1–3 sentence narrative findings for each retained source (direction, population, endpoint, effect size where available), instead of referring the reader to a table that itself reports 'unclear' for most rows. The map should actually map the findings.
4. In Tensions and Gaps, name specific studies and specific disagreements (e.g., Martens 2018 blood-pressure signal vs. Connell 2021 null muscle outcomes; Liao 2021 aerobic capacity gain vs. Martens 2018 neutral cardiovascular findings; Ren 2023 brain NAD+/NADH ratio in PD/MS vs. absence of replication; Curran 2025 preclinical infection benefit vs. no human infection RCT; positive cardiometabolic signals in Yu 2025/Pei 2024 vs. surrogate-endpoint caution).
5. Define the 'claim' and 'tension' counting units (claim = one extracted directional finding per source per endpoint; tension = pairwise directional disagreement) and report them in the methods, so the '915 claims' and '146 tensions' figures are interpretable.
6. Either add Ioannidis 2005 to the source bundle or replace the surrogate-endpoint argument with a different cited framework; do not invoke an uncited reference as a load-bearing methodological anchor.

Major issues

• Several cited sources in the manuscript are off-topic noise (Simonis 2025 on cirrhosis decompensation NAD vs AD; Zhao 2024 on acupuncture for anxious depression NAD vs non-anxious depression; Han 2022 on air pollution and NAD mortality in Beijing) — these use the acronym 'NAD' for non-nicotinamide-adenine-dinucleotide concepts and are miscategorized as 'Contextual Adjacent Evidence' for NAD+ effects. Their inclusion as landscape sources is a substantive source-fit defect that the evidence-honesty framing does not excuse, because they pollute the cited corpus with claims that have nothing to do with the topic.
• The source bundle also contains papers that are clearly about NAD+ but were not in the manuscript's author-year citation list (e.g., Holmes 2026, Christen 2026, Membrez 2024, Mevenkamp 2024, Yu 2025, Pei 2024, Bai 2022, Ministrini 2025, Liao 2021, Baichuan 2023, Elhassan 2019, Martens 2018, Airhart 2017, Vreones 2022, Curran 2025/2023, Connell 2021, Ren 2023, Simon 2024, Xue 2022, Okabe 2022, Yi 2022, Katayoshi 2023, Nazari 2022, Gao 2025/2026). The manuscript's text references many of these by author-year, but the in-text bibliography in the manuscript is incomplete, and the mapping from in-text citations to bundle entries is not auditable without the manifest.json.
• Findings Map tables are largely non-informative: directional signals are reported as 'unclear' for 3/3 muscle-function sources, 'unclear' for 1/3 cardiometabolic sources, and 'no extracted directional signal' for 8/16 contextual sources, and the narrative per outcome class is either empty ('See the structured evidence table…') or absent. The map does not actually report what each retained study found in narrative form, so the 'every mapped finding attributed to specific cited sources' check is only partially met.
• Tensions and Gaps section is generic: it lists categories (cardiometabolic, muscle function, contextual other) but does not surface specific contradictions between named studies (e.g., Martens 2018 vs. Connell 2021 on blood pressure/arterial stiffness vs. null muscle outcomes; Liao 2021 positive aerobic capacity vs. Martens 2018 neutral cardiovascular findings; Ren 2023 brain NAD+ engagement vs. absence of functional replication). The 146-tensions count is asserted but not substantiated with named-study examples.

Minor issues

• Abstract's '915 claims' and '146 non-orthogonal tension(s)' are reported without explanation; these counts should be defined operationally.
• Search summary lists databases and queries but the eligibility funnel shows 169 → 49 → 30 with multiple intermediate buckets whose meanings ('Mixed partial-or-none', 'Partial-only') are not defined for the reader.
• The Ioannidis 2005 surrogate-endpoint citation is invoked repeatedly but is not present in the source bundle; it should be added if central to the argument.
• Evidence-honesty framing is appropriate, but the manuscript still uses phrasing such as 'defines a bounded geroscience rationale' which risks implying more mechanistic-clinical convergence than the cited human evidence supports.
• Manuscript is an evidence map and is not required to converge on a single claim, but the abstract's second paragraph still makes a 'thesis' statement ('evidence for NAD+ Effects is context-dependent') that the body does not explicitly test or adjudicate.

Reviewer note

This is an evidence-map submission on NAD+ precursor effects that correctly avoids collapsing into a single clinical claim and includes an appropriate evidence-honesty frame. The scope (30 sources across cardiometabolic, muscle function, dosing/PK, longevity, frailty, immune/inflammation, safety) and search summary are reasonably well bounded, and the Limitations section is specific and material (no long-term hard-outcome RCTs, narrow populations, surrogate-endpoint caution, single-trial generalization, mechanism-to-clinic gap). However, several defects block an accept: (1) at least three cited sources (Simonis 2025, Zhao 2024, Han 2022) are off-topic noise that use 'NAD' as an acronym for non-nicotinamide-adenine-dinucleotide concepts and are miscategorized as contextual adjacent evidence for NAD+ effects; (2) the Findings Map tables are largely non-informative, reporting 'unclear' or 'no extracted directional signal' for most rows, and the narrative subsections are empty or refer the reader back to the same non-informative table, so the 'every finding attributed to specific sources' check is only partially met; (3) the Tensions and Gaps section is generic and does not name specific study-to-study contradictions; (4) the source-grounding of in-text author-year citations is not auditable in the manuscript itself (it depends on an external manifest.json). The claims that are made are appropriately hedged, and the evidence-honesty framing prevents overclaim, so claim_support is 'partially_supported' and overclaim is 'mild'. The manuscript is salvageable with bounded edits: remove the off-topic sources, populate per-source narrative findings, name specific tensions, and make the citation-to-bundle mapping explicit. Recommendation: revise.

Panel metadata

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

Route: consensus

Prompt: reviewer-v11-research-synthesis