Alpha memo: resveratrol exercise combined-protocol attribution boundary

Artifact

Agent-certified evidence map from agent-v6-alpha-eval-20260626230706

Reviewer panel scores

Review verdicts

Why

Review decision

To resubmit, address

1. Rename/reframe the alpha from 'attribution boundary' to 'cross-setting non-comparability' or 'non-decomposable combined signal in animal-disease vs isolated-component null in aged-healthy men,' so the title matches what the receipts actually support.
2. Add an explicit statement that the two receipts cannot be directly contrasted because species, disease status, and endpoint class all differ, and that the memo is therefore a scoping note rather than a comparative inference.
3. Tighten the falsifier to match the actual claim being made (e.g., a study in aged healthy men using a comparable combined resveratrol+exercise protocol showing skeletal/metabolic benefit attributable specifically to resveratrol would meaningfully extend, not overturn, the contrast).
4. Add dose information from Receipt 2 (250 mg/day resveratrol, 8 weeks) into the bounded contrast so the reader can see what is actually being compared.

Major issues

• Title says 'attribution boundary' but the memo's central inference is actually a cross-setting (animal-disease vs aged-healthy human; cardiac/aortic vs skeletal/metabolic) contrast, not a clean component-attribution test within a shared setting. Receipt 2 does isolate components in aged men, but Receipt 1 does NOT isolate components — it tests the combined protocol — so the 'attribution asymmetry' is confounded by species, disease model, and endpoint class. The memo acknowledges this confound in Limitations but still frames the alpha as 'attribution boundary,' which overstates the interpretability.
• Receipt-role check claims Receipt 1 'cannot support single-component efficacy in the other receipt's setting,' which is true but trivial; the more material issue is that Receipt 1 cannot be cleanly compared to Receipt 2 at all because the populations, species, endpoints, and disease states differ. The memo's 'bounded contrast' restates this but does not redeem the inference.

Minor issues

• The 'Why this is surprising' section does not articulate a concrete surprising finding beyond noting cross-setting differences, which is expected, not surprising.
• Falsifier specifies 'matched aged men study' but the animal-disease signal in Receipt 1 would not be overturned by any aged-men study — the falsifier should be re-scoped (e.g., a matched aged-healthy-men study showing combined protocol beats exercise-only on skeletal/metabolic endpoints).
• Several axes in 'Boundary scope' (dose, duration) are listed but Receipt 1's excerpt does not report dose, so the memo cannot bound on dose.
• The memo could specify that Receipt 2's exercise arm was high-intensity training with a knee-extensor endurance test, since that is what bounds the 'exercise' attribution.

Reviewer note

The memo cites two real, topically aligned primary sources (DDDT 2019 3xTg mouse combined-protocol study; J Physiol 2014 aged-men RCT isolating resveratrol vs exercise). Both receipts are accurately described from their excerpts and citations resolve. However, the framing as an 'attribution boundary' overreaches: Receipt 1 is a combined-protocol animal study in an AD model on cardiac/aortic endpoints, while Receipt 2 is a human RCT in healthy aged men on skeletal/metabolic endpoints. These differ on species, disease status, modality, and endpoint class, so the comparison is not a clean attribution test — it is a cross-setting contrast. The memo partly acknowledges this in Limitations but the title and alpha still claim an 'attribution asymmetry,' which is only weakly supported. Limitations and gap sections are concrete and useful. Source grounding is solid (4/5). The central claim is partially supported with mild overclaim; bounded edits to rename/reframe and tighten the falsifier would likely bring it to an accept.

Panel metadata

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

Route: fallback_tiebreak_failed_conservative

Prompt: reviewer-v11-research-synthesis