Research Synthesis: NAD+ Metabolism Effects

Artifact

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

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Rebuild the Findings Map to attribute each outcome-class signal to named sources (author-year) so the landscape is auditable per source, not just per class.
2. Recompute or disclose the basis for 'no extracted directional signal' coding; the source abstracts show clear directional findings (positive NAD+ elevation in Martens 2018, Conze 2019, Elhassan 2019; null functional outcomes in Connell 2021; null metabolic outcomes in Chen 2024 meta-analysis). Reconcile this with the abstract claim of 8/12 null-coded sources.
3. Expand the Tensions and Gaps section to name the specific disagreements present in the bundle — e.g., the divergence between biomarker-level NAD+ elevation (positive) and functional/clinical endpoints (null) across the NR/NMN trials, and the conflict between mechanistic/safety evidence and clinical efficacy evidence.
4. Remove or qualify the '0 cross-study disagreements' claim in the abstract, or replace it with an explicit description of the disagreements that are visible in the bundle.
5. Clarify whether the 12-source bundle includes only the cited references or whether the limitations-section citations (Ioannidis, Cruz-Jentoft, Studenski, Cesari) are out-of-corpus context; if out-of-corpus, state this explicitly.

Major issues

• The abstract and scope state '0 cross-study disagreements' and '0 non-orthogonal tensions' while the source bundle contains multiple sources with divergent signals (e.g., Martens 2018 and Conze 2019 suggest NR elevates NAD+; Connell 2021 shows no effect on mitochondrial or muscle function; Chen 2024 meta-analysis shows no significant benefit on glucose/lipid markers; Membrez 2024 supports a trigonelline-NAD+ axis in muscle). The map silently smooths these into a null/unclear aggregate, which is a form of heterogeneity collapse.
• Several sources in the bundle are classified as 'Contextual Adjacent Evidence' with 'no extracted directional signal' yet their abstracts report clear directional findings (e.g., Conze 2019: 22-142% dose-dependent NAD+ increase, safe; Martens 2018: NR well-tolerated, elevates NAD+; Elhassan 2019: NR elevated muscle NAD+ metabolome and anti-inflammatory signatures). The 'no extracted directional signal' coding appears to understate the actual source-level evidence and misrepresents the bundle to readers.
• The Tensions and Gaps section is generic and procedural ('Run adequately powered human studies...') rather than surfacing the specific disagreements visible in the bundle (e.g., the Martens/Conze/Elhassan positive NAD+ elevation findings vs. the Connell null functional findings vs. the Chen 2024 null metabolic findings). This is exactly the heterogeneity the evidence-map review is supposed to reward, and it is absent.
• Findings Map rows do not name the specific sources supporting each outcome class. With 12 sources and small n per class, the table should map source-level attribution (e.g., Cardiometabolic: Chen 2024, Katayoshi 2023, Martens 2018) so readers can audit which sources drive each signal claim.

Minor issues

• The abstract line 'with 0 cross-study disagreements' contradicts the evidence-honesty framing; if disagreements are coded as zero, this should be reconciled or the tension should be named explicitly.
• The Limitations section cites Ioannidis 2005, Cruz-Jentoft 2019, Studenski 2011, Cesari 2009, and Qader 2025 by name, but these are not in the admitted source bundle (n=12). This is acceptable as contextual reference but should be disclosed as outside-corpus citations.
• Search Summary inclusion of preprints and the funnel description are thorough, but the relationship between the 175-record union and the 12 admitted sources is not fully auditable from the manuscript alone.
• Several 'year 2026' entries (Holmes, Christen, Gao) have future-dated publication years relative to typical indexing; this is likely a versioning artifact but should be clarified.
• Outcome-class table repeats 'limited corpus depth in this outcome class' as the limitation for every row, which is not informative variation.

Reviewer note

This evidence map has a reasonable structure and an appropriate evidence-honesty framing, and it correctly avoids collapsing into a clinical recommendation. However, it materially understates the directional content of the source bundle: the bundle contains clear positive signals (NR elevates NAD+ in Conze 2019, Martens 2018, Elhassan 2019, Christen 2026) and clear null signals (Connell 2021 functional null; Chen 2024 meta-analytic null on glucose/lipids), which constitute a genuine and informative tension between biomarker-level and clinical-level evidence. The manuscript's claim of 0 disagreements and 8/12 null-coded sources is not faithful to the source bundle and silently collapses the most important heterogeneity in the NAD+ literature. The Findings Map also lacks source-level attribution, making the landscape non-auditable. The Limitations section is the strongest part of the manuscript. Recommendation: revise to (1) attribute findings to specific sources, (2) reconcile the directional coding with the actual source content, and (3) expand Tensions and Gaps to surface the biomarker-vs-clinical-endpoint divergence explicitly.

Panel metadata

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

Route: fallback_tiebreak_failed_conservative

Prompt: reviewer-v11-research-synthesis