Adjacent Evidence Brief: Mitochondrial DNA damage

Artifact

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

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Reconcile the source-admission funnel with an explicit, auditable arithmetic (or clearly tabulate the non-additive diagnostic states with a glossary) so the 15-source base can be independently verified.
2. Map every admitted bundle source to its appearance in the Findings Map (or explicitly mark sources not surfaced in prose with justification), and surface Chakraborty 2026, Chan 2012, and any other under-cited sources in the appropriate outcome-class section.
3. Recode effect directions against bundle excerpts, surfacing explicit positive/negative findings where the cited source clearly states them (e.g., Pena 2024 reversal of mtDNA damage; Ng 2019 null lifespan relationship; Shimizu 2026 PUFA-induced damage) rather than defaulting to 'unclear'.
4. Add an explicit Tensions and Gaps subsection enumerating the cross-source disagreements visible in the bundle (mFRTA vs. Ng 2019; PUFA direction; cancer-muscle aging acceleration; HIV-associated brain mtDNA damage).
5. Verify and correct the representative statistics cited in the Findings Map against bundle excerpts, especially the Pena 2024 p-value which appears anomalous.
6. Standardize outcome-class and directness terminology across the manuscript and bundle.

Major issues

• The source bundle contains 15 entries but only 11 are actually cited in the manuscript prose; four (Chakraborty 2026, Chan 2012, Picca 2019, Picca 2020 are cited, leaving Chakraborty 2026 uncited in the Findings Map while other sources are present). Several sources cited in the Findings Map sections (e.g., the muscle function section lists only Picca 2019, Picca 2020, Luo 2024) miss other bundle members, creating an incomplete mapping between cited prose and the 15 admitted sources.
• The admission funnel is internally contradictory: the funnel table lists 47 candidate union, 18 classified, then several non-additive buckets whose arithmetic does not reconcile (e.g., 'Mixed partial-or-none: 14' plus 'Partial-only: 2' plus 'Strict high-confidence: 1' totaling 17, not 18 classified). The manuscript acknowledges non-additivity but does not provide a clarifying reconciliation, undermining auditable scope.
• The direction-coding scheme labels nearly all findings 'unclear' even when bundle abstracts show clear positive or negative direction (e.g., Pena 2024 explicitly states mtDNA damage was 'reversed to healthy control levels'; Ng 2019 explicitly states 'no simple relationship between mtDNA damage and lifespan'). This conservative coding is internally inconsistent with the bundle evidence and suppresses legitimate heterogeneity the map should surface.
• The Findings Map reports 'no load-bearing cross-study disagreements' but the source bundle contains explicit tensions: Ng 2019 finds mtDNA damage does not determine lifespan (against mFRTA) while other sources frame mtDNA damage as causally implicated in aging phenotypes; Shimizu 2026 reports PUFA-rich diet increases mtDNA damage while typical health messaging treats PUFA as protective. These tensions are not surfaced.
• The cardiometabolic row claims 'negative signal in 1/1 sources' but classifies this as 'mechanistic' while other outcome classes distinguish 'indirect' from 'mechanistic'; directness categories are applied inconsistently across rows.

Minor issues

• Several bundle members are dated 2026 (Hsiao 2026, Shimizu 2026, Chakraborty 2026) which is plausibly post-dating the stated retrieval window of 2026-06-28 but warrants explicit handling.
• The cited representative statistics (e.g., Pena 2024 p = 0.92 reported as a significant source statistic) need verification against bundle excerpts; a p = 0.92 is typically non-significant, suggesting either a coding error or misalignment between the manuscript statistic and the cited source.
• Search queries include 'aging decline AND aging AND human' which is redundant and not well-formed; this should be cleaned up for auditability.
• The Tensions and Gaps section is thin (one paragraph) and does not explicitly enumerate the heterogeneity surfaced in the bundle.
• Outcome class labels in the bundle (e.g., 'deficiency_prevalence') differ from manuscript labels ('Deficiency Prevalence'), requiring consistent terminology for auditing.

Reviewer note

Evidence-map review of 'Mitochondrial DNA damage.' The scope is explicitly bounded (15 sources, 387 claims, no direct interventional hard-endpoint evidence), which is appropriate for the topic. Source attribution is present at the per-finding level, and the findings map separates outcome classes without pooling into one clinical effect — a structural strength. However, several issues prevent an accept: (1) the admission funnel is not arithmetically auditable as presented; (2) at least one admitted bundle source (Chakraborty 2026) is not surfaced in the Findings Map, while a few others are inconsistently cited; (3) direction coding is overly conservative — the bundle clearly shows positive and null directions in several sources that are coded 'unclear,' suppressing legitimate heterogeneity the map should surface; (4) the claim of 'no load-bearing cross-study disagreements' contradicts visible tensions in the bundle (e.g., mFRTA vs. Ng 2019 lifespan null; PUFA-protective vs. PUFA-damaging). The Limitations section is honest about evidence-role imbalance and endpoint heterogeneity, and the explicit evidence-honesty note is a strength. The Tensions and Gaps section is present but too thin. Overall: competent but fixable with bounded edits to the funnel, direction coding, and tension enumeration — revise rather than reject, since the structural skeleton is sound and the cited sources are real and traceable.

Panel metadata

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

Route: consensus

Prompt: reviewer-v11-research-synthesis