# Adversarial review log — peptide-stacks-recovery

Authoring model: **Sol 6.1**. Adversarial reviewer: **Claude Opus 5.5 (Anthropic)**. Date: **2026-10-14**.
Round 2; group 12 of `NEW_REVIEW_INVENTORY_2026-10-14.json`.

## Summary

| Item | Value |
|---|---|
| Level before | Γ — Mechanistically plausible, partial system support / Moderate |
| Level after | Γ — Mechanistically plausible, partial system support / Moderate (unchanged) |
| Findings: Critical | 0 |
| Findings: Major | 0 |
| Findings: Minor | 3 |
| Findings: Editorial | 1 |
| Rejected findings | 9 |
| PMIDs checked / failed | 56 / 0 |
| Load-bearing claims spot-checked | 13 |
| Validator | PASS (0 fail, 0 warn, 30 pass) — identical before and after. |
| Approval | **Approved as revised.** |

## Findings

Full table (AR-1 to AR-4), the nine rejected findings and the residual limitations are in the `## ADVERSARIAL REVIEW AND APPROVAL` section of `peptide-stacks-recovery.review.md`. In brief:

- **AR-1 (Minor).** Confidence rationale cited reviewer access depth; rewritten on evidence-density and reproducibility grounds. Band stands at Moderate. Lead two-step test (2026-10-14) applied: step 1 strikes access depth as a determinant; step 2 re-derives the band from what remains, and it stands at Moderate because the review documents substantive evidence-quality reservations of its own, which are legitimate determinants.
- **AR-2 (Minor).** Divergence "Contested" given its required artifact naming the two conflicting framings.
- **AR-3 (Minor).** Two empty `DOI: .` fields (references 8 and 42) repaired to `DOI: not available.`; verified that PubMed carries no DOI for either record.
- **AR-4 (Editorial).** 22 fused or mis-spaced tokens separated across the review and the score card.

## Safety and efficacy attribution audit

This case concerns unapproved consumer products, so the attribution of harm and benefit was audited specifically. No correction was needed in either direction. The high-dose growth-hormone mortality result is reported accurately (two parallel randomized trials, 247 Finnish plus 285 other European patients, 39 versus 20 percent and 44 versus 18 percent in-hospital mortality) and explicitly bounded to prolonged critical illness; the review states on its own initiative that no unverified CJC-associated death allegation is used. Harms are not inflated (the rat tendon-bone finding is held to its model and high dose; the product-identity report is held to one sampled product). Benefits are not understated: the real positives are retained, including the 3.9 percent sprint gain in 96 athletes with its loss after washout, perioperative growth hormone improving walking distance in 33 hip-replacement patients, and capromorelin improving tandem walk and stair climb in 395 older adults with mild functional limitation. The net-benefit subclaim stays at Α in both directions — no measured exact-stack benefit-harm comparison, and no universal toxicity conclusion.

## Propagation

The two byte-identical copies listed in the inventory were verified identical before the edit and overwritten with the finished canonical file afterwards. Hash compare confirms all three copies match:

- `hsef_v53_traffic_ranks_06_10_2026-10-13/peptide-stacks-recovery/peptide-stacks-recovery.review.md`
- `hsef_v53_topical_all_22_2026-10-13/cycles/hsef_v53_traffic_ranks_06_10_2026-10-13_share/peptide-stacks-recovery/peptide-stacks-recovery.review.md`
- `hsef_v53_traffic_ranks_06_10_2026-10-13_share/peptide-stacks-recovery/peptide-stacks-recovery.review.md`

review.md sha256 `9d7df60d1ce1631d7765aab2c2ea2f52678f75db30804286a476ad9e4a17d7a4`; scorecard.json sha256 `f0075ac2de85c3f7f52bd353d57062caf83ac1c06cc56c0baaec5873cac895de`.

Per the lead's propagation rule (2026-10-14), propagation was applied to **every** artifact that was byte-identical before the edit, not only to the `review.md` named in the inventory. Both edited artifacts (`peptide-stacks-recovery.review.md` and `peptide-stacks-recovery.scorecard.json`) were confirmed byte-identical across all three copies beforehand — each copy's own `SHA256.txt` records exactly the pre-edit canonical hashes (`2bdd6f12...` review, `e5e61250...` score card) — and both were propagated, together with `ADVERSARIAL_REVIEW.md` and `adversarial-findings.json`. A `sol61-original/` backup of both files now exists in each of the three copies. A full walk of the three folders confirms all 17 common files are byte-identical and that neither share copy holds a file the canonical lacks. No other artifact in this case was edited, so nothing else could diverge; the sibling files still carrying the AR-4 rendering defects remain identical across copies and are listed above for the lead. Each copy's `SHA256.txt` is equally stale in all three and was left as a pre-review snapshot rather than rebuilt.

## Strings for the lead to propagate to the remaining case and package artifacts

No level, confidence, role, source count or claim count changed, so no package-level scoring string needs updating. The AR-4 rendering repairs were applied only to `peptide-stacks-recovery.review.md` and `peptide-stacks-recovery.scorecard.json`. The same defective strings remain in the sibling case artifacts `CONSENSUS_CROSSCHECK.md`, `RESEARCH_TIMELINE.md`, `research-timeline.json`, `evidence.json`, `adjudication.json`, `adjudication-draft.json`, `enrichment-draft.json` and `consensus-validation.json`, and in the derived `web-page.md` / `video-insert.md` where they appear.

| Old | New |
|---|---|
| `over4–24` | `over 4–24` |
| `and7864103` | `and 7864103` |
| `ROMANA26906526` | `ROMANA PMID26906526` |
| `Rubric06` | `Rubric 06` |
| `and26-week` | `and 26-week` |
| `of25.3 and32.6` | `of 25.3 and 32.6` |
| `blinded96-athlete` | `blinded 96-athlete` |
| `Universal cap4` | `Universal cap 4` |
| `examined2026` | `examined 2026` |
| `the2026` | `the 2026` |
| `thymosinbeta4` | `thymosin beta4` |
| `a3.9percent` | `a 3.9 percent` |
| `GH:39` | `GH: 39` |
| `VO 2 max` | `VO2 max` |
| `DOI: . Access` | `DOI: not available. Access` |

**Approval:** Following adversarial review and the corrections listed above, the revised document is approved. — Claude Opus 5.5 (Anthropic), 2026-10-14.