Research Synthesis: Acute Exercise Effects

Artifact

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

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Reconcile in-text statistics with source bundle excerpts: either confirm the Harris 2008 IL-6 statement against the actual bundle excerpt or remove/relocate it. Audit all 'representative statistic' lines for trace accuracy to bundle text.
2. Reframe the research question to a more falsifiable, specific claim (e.g., effect on a specific endpoint in a specific population) so the manuscript answers rather than deflects.
3. Disaggregate the 26-source 'Contextual Adjacent Evidence' bucket into at minimum cognitive, immune/inflammation-adjacent, vascular/hemodynamic, and nutrition-interaction sub-classes, or explicitly justify the lumping and discuss what is lost.
4. When citing class-level direction profiles, qualify the statements to reflect that 21/26 sources in the largest class are coded 'unclear'; do not state 'Negative signals appear in contextual other' without that caveat.
5. Tighten the Cross-Domain Synthesis and Discussion so generic boundary-condition paragraphs are either removed or rewritten with corpus-specific examples; the 'Metabolic-Functional Tradeoff Framework' should be referenced where it applies or removed.
6. Justify the next-study design thresholds (n≥200, ≥12 months) from the cited corpus or mark them as expert-derived rather than corpus-derived.

Major issues

• The research question ('do findings...support a decision-grade conclusion') is essentially answered in the negative throughout the manuscript, but the question itself is broad enough that the synthesis reads more like a refusal than an answer. The paper does not commit to a falsifiable specific question about a specific effect, dose, or population.
• Several source-level statistics appear to be drawn from or attributed to bundle abstracts but the in-text quotes do not always match the bundle excerpts. For example, the manuscript states Harris 2008 [bundle:41] reports 'Elevated (P < 0.001) concentrations of IL-6 following moderate (50% VO(2)) and high (75% VO(2)) intensity acute exercise' but the bundle excerpt for Harris 2008 reports a 24% FMD increase in active group (P=0.034) vs a 32% decrease in inactive group (P=0.010); the IL-6 statement does not appear in the bundle excerpt. This raises a traceability concern under exact-statistics calibration rules.
• The manuscript categorizes the entire 26-source 'Contextual Adjacent Evidence' bucket as a single outcome class, but this class includes cognitive function trials (Austin 2019, Johnson 2019a/b, Hatch 2021), T-cell/AhR immunobiology (Schenk 2021), nutrition interactions (Bauer 2025, Rohling 2021, Kappus 2011), marathon stress (Rohling 2021), and others. Lumping these into one bucket collapses clinically distinct endpoints and weakens the cross-domain synthesis claim of 'high-density pairwise disagreement.'
• The Direction-profile counts in the Results table (e.g., contextual_adjacent_evidence positive=0, negative=1, null=4, unclear=21) imply that 21 of 26 sources in that class are coded 'unclear,' which makes any class-level directional claim weak. The paper nonetheless makes directional statements ('Negative signals appear in: contextual other, immune') based on those counts, which understates the underlying uncertainty.
• The Cross-Domain Synthesis and Discussion sections are heavily templated and contain generic sentences (e.g., 'Population is the first boundary on transfer,' 'Dose and schedule form a separate boundary') that are not tied to specific findings from the cited corpus. Several paragraphs read as reusable boilerplate rather than corpus-specific interpretation.

Minor issues

• The 'Source-context map' separates infectious-disease/immunology, skeletal/muscle, aging/geroscience, and oncology/cancer contexts, but the 'aging/geroscience' slice contains only 2 sources (Heselton 2024 and White 2021) while 'contextual adjacent evidence' (n=26) is by far the largest class — this asymmetry is not discussed.
• Reference list format mixes full first names with abbreviated forms and includes both DOI and PMID where available; minor inconsistency.
• Several source titles include non-ASCII characters (e.g., 'HEIGHTENED STRESS IN FOOD CHOICES' all-caps, 'RGS16 – CXCL12') that may reflect encoding artifacts.
• The 'Metabolic-Functional Tradeoff Framework' is a paper-level organizing claim but is not explicitly cross-referenced in the Cross-Domain Synthesis or Conclusion, weakening its role.
• The next-study design recommendation specifies ≥200 participants/arm and ≥12 months follow-up but does not justify these thresholds from the cited corpus.
• The Conclusion restates the bounded-claim language multiple times without tightening the operational implications for the stated research question.

Reviewer note

The manuscript is a structured research synthesis on acute exercise effects across 43 included sources and five outcome classes, with explicit directness coding, a quantitative evidence index, cross-domain synthesis, and a bounded conclusion. These structural features place it above a minimal synthesis and meet several recommended depth sections. Strengths: (i) Methods are explicit about directness coding, evidence tier, risk-of-bias framework, and outcome-class assignment. (ii) The Quantitative Evidence Index / source-level table and the Evidence Landscape provide clear traceability. (iii) Cross-domain tensions are surfaced (e.g., Kristiansen 2026 vs Zhang 2025 on inflammation) rather than smoothed away. (iv) Hedging language is appropriate and proportionate to the source quality. (v) Sources cited in the manuscript match corresponding bundle entries by year and title, and DOIs/PMIDs resolve to plausible records. Weaknesses: (i) At least one in-text statistic (Harris 2008 IL-6 finding) does not appear in the corresponding bundle excerpt, raising a traceability concern under exact-statistics calibration. (ii) The largest outcome class ('Contextual Adjacent Evidence,' n=26) collapses heterogeneous endpoints (cognition, immunobiology, nutrition, marathon stress), weakening class-level directional claims. (iii) The Conclusion and Discussion rely on templated boundary-condition paragraphs that are not corpus-specific. (iv) The research question is answered in the negative throughout rather than interrogated; the synthesis reads as a refusal rather than an answer. (v) Some class-level directional statements ('Negative signals appear in contextual other') understate the underlying 'unclear' majority in that class. Verdict: The paper is structurally close to a gatekeeper-tier artefact but is undermined by the traceability issue and the over-aggregation of the largest outcome class. Bounded edits — statistic audit, class disaggregation, and tightening of generic synthesis paragraphs — would likely bring it to accept quality. Recommendation: revise.

Panel metadata

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

Route: fallback_tiebreak_failed_conservative

Prompt: reviewer-v12-grounded-integrity