# Adversarial review log — telomere-shortening

**Case:** telomere-shortening (`hsef_v53_remaining_priority_06_10_2026-10-09/`)
**Authoring model of record:** Sol 6.1. The package's own `review_metadata.reviewer` is "Codex" and
`provenance_header.generated_by` is "Codex; HSEF v5.3"; per the lead this is corpus-wide house provenance and was
left unedited.
**Adversarial reviewer:** Claude Opus 5.5 (Anthropic)
**Date:** 2026-10-14 · **Round:** 2, group 1

## Summary

| Item | Before | After |
|---|---|---|
| Headline level | Γ — Mechanistically plausible, partial system support | Γ — unchanged |
| Confidence | Moderate | Moderate — unchanged (rationale corrected) |
| Causal role | reinforcing / stabilizing factor | unchanged |
| Subclaim levels | SC1 Ε · SC2 Ε · SC3 Γ | unchanged |
| Aggregation | primary-subclaim (SC3); weakest-link Γ, scope-matched Γ | unchanged |
| Momentum / Stability / Divergence | Mechanistic only / Stable / Contested | unchanged (Divergence artifact added) |
| Sources / claims | 51 / 55 | 51 / 55 |
| Validator (v5) | PASS — 0 fail, 0 warn, 31 pass | PASS — 0 fail, 0 warn, 31 pass |

**Findings by severity:** Critical 0 · **Major 2** · Minor 2 · Editorial 0. Rejected findings: 8.

**PMIDs:** 51/51 reference PMIDs verified against PubMed; 3/3 prose-only PMIDs verified (all Published Errata,
correctly attached to parent rows rather than given rows); 0 failures; no retraction or expression of concern in any
of the 54 records.

**Claims spot-checked:** 12, all matched their source exactly.

## Findings

| ID | Severity | Location | Original → corrected | Evidence |
|---|---|---|---|---|
| E-1 | Minor | Document-wide | 59 × `PMID<digits>` → `PMID <digits>` | The whitespace-loss defect named in the round-2 brief; the reference list already used the spaced form 61 times in the same document. |
| E-2 | **Major** | "Registered claim and scope", Selection sentence | "; existing HIV draft consulted after a source-based checkpoint" removed | **A false provenance claim.** No HIV draft belongs to this case. The clause is batch boilerplate present in all five case files of this batch and accurate only for `hiv-aids`. Lead-flagged; another group applied the same correction to `dopamine-schizophrenia`. |
| E-3 | **Major** | `telomere-shortening.scorecard.json`, `final_score.confidence_rationale` | Access depth removed as a ground for the band; the rationale now rests on evidence density and reproducibility and says access depth is recorded separately | The framework states that confidence reflects evidence density and reproducibility, and that access depth is "a statement about what you did, not about the source". **Band unchanged** — see below. Lead-flagged. |
| E-4 | Minor | Divergence line | "No manufactured field consensus or arithmetic gap." → same verdict with the split named | The framework requires a checkable artifact for Divergence and allows "Contested" only where the field is genuinely split. The added clause uses only framings already in this review's evidence base. |

## Why the Moderate band survives (E-3 argued both ways)

The lead's rule is that the band is wrong if the cited evidence meets the High text — "extensive independent
replication across labs, models, and assay modalities; reproducibility well-established". It does not, at the
registered scope. The cellular and model-organism evidence replicates well, but that is SC1/SC2 scope. The evidence
bearing on SC3 — ordinary human aging — rests on a single mouse gene-therapy programme, one pharmacological
activator, an uncontrolled syndrome treatment, one randomized supplement trial with a telomere-length endpoint, and
mean-leukocyte-telomere-length Mendelian-randomization instruments whose functional and lifespan results are
discordant. Moderate stands; only the stated reason for it was wrong.

## Rejected findings

R-1 … R-8 are tabulated inside the review. Three worth the lead's attention:

- **R-3, collective-author defect: correctly handled here, and the upstream cause is visible.** Reference 11 is the
  Telomeres Mendelian Randomization Collaboration paper, whose PubMed record carries a CollectiveName; the PubMed
  tool returns its first author as an **empty object**, which is exactly what produces the ", Haycock PC …" defect
  elsewhere in the corpus. This review renders it correctly as "Telomeres Mendelian Randomization Collaboration,
  Haycock PC, Burgess S et al."
- **R-4, the `models_evaluated[0]` renderer bug: not present.** `models_evaluated[0]` is the rank-1 non-nominal
  model and the prose names it; the nominal sits at index 3, rank 4.
- **R-2, aggregation.** A Γ headline from a set containing two Ε subclaims is the brief's warned pattern, handled
  correctly: the pre-registered rule is primary-subclaim, SC3 is the registered claim verbatim, SC1 and SC2 are
  explicitly scope-restricted, and all three rule outcomes are reported.

## Files changed in this case folder

- `telomere-shortening.review.md` — edited (4 corrections, banner, approval section)
- `telomere-shortening.scorecard.json` — edited (E-3, one line: `final_score.confidence_rationale`)
- `validator.txt` — new (v5 validator output for the revised edition)
- `ADVERSARIAL_REVIEW.md`, `adversarial-findings.json` — new
- `sol61-original/` — pre-edit copies of `telomere-shortening.review.md`, `telomere-shortening.scorecard.json`,
  `validation-report.json`, `consensus-validation.json`, `SHA256.txt`

Line endings were preserved as CRLF throughout, so each edited file's diff against its backup is confined to the
corrected content.

## Package-level propagation needed from the lead

No headline string, letter, confidence band, count or title changed, so `OVERVIEW.md`-class values are not stale.
Needed:

1. Refresh the `telomere-shortening.review.md` and `telomere-shortening.scorecard.json` digests in this case folder's
   `SHA256.txt` and in any package `MANIFEST.json` / `BATCH_COMPLETION.json` entry that carries them (old → new in
   `adversarial-findings.json` → `changed_file_hashes`).
2. **Batch-wide, lead-flagged:** the false HIV-draft provenance clause and the access-depth confidence rationale are
   present across this whole batch, and the `PMID<digits>` fusion is present in all five case files
   (dopamine-schizophrenia 73, hiv-aids 52, inflammation-depression 74, lipid-heart-diet 55, telomere-shortening 59).
   `hiv-aids` is the one case where the HIV clause may be accurate and should be checked rather than stripped.

No `propagate_to` or `divergent_copies` entries exist for this slug.

## Lead policy applied — confidence bands and momentum (2026-10-14)

- **Confidence two-step test:** step 1 (strike access depth as a determinant) applied; step 2 (re-derive) -> **Moderate stands**, band unchanged.
- **Momentum stricter rule** ("Decisive work underway" requires work actually being run; a completed publication never satisfies it): **holds; positive contrast**.
- Small intervention cohorts, an uncontrolled syndrome treatment, mean-LTL reporter instruments and discordant lifespan results survive step 2; the good model-organism replication is SC1/SC2 scope, not the registered scope. Momentum is already "Mechanistic only" and its artifact is a completed publication - exactly why it cannot support "Decisive work underway".

Full derivation is recorded in the review's `## ADVERSARIAL REVIEW AND APPROVAL` section and in `adversarial-findings.json` -> `lead_policy_2026_10_14`.

## Consensus question for the lead

In adults **without** an inherited telomere biology disorder, does any intervention that measurably lengthens or
maintains telomeres improve a prespecified functional healthspan endpoint — not telomere length itself — in a
randomized trial? A yes is the first positive human functional perturbation and would move SC3 off Γ; the Mendelian
randomization lifespan and functional-aging nulls currently point the other way.