# Adversarial review - resident-brain-microbiome (2026-10-09 revised edition)

Authoring model: Sol 6.1. Adversarial reviewer: Claude Opus 5.5 (Anthropic). Date: 2026-10-14.

Level before to after: Α Moderate to Α Moderate (unchanged). Momentum: Decisive work underway to Mechanistic only.
Findings: 0 Critical, 2 Major (one a revision-introduced regression), 1 Minor, 1 Editorial. PMIDs verified 59/59. Claims spot-checked: 12, six of them against retained original full texts.
Validator (framework/5.4-registry-compatibility): FAIL - 1 fail, 1 warn, 33 pass; **identical on the pre-edit copy**, so the single FAIL (v5.lay_summary, a v5.4-only requirement a v5.3 review cannot satisfy) and the v5.claim_form phenomenon WARN are pre-existing and untouched.

## Crux check: contamination controls

The review's treatment of negative/blank controls and of the contamination counter-explanation matches what the cited papers report, on every point checked, and in each case it takes the conservative reading. Verified against the retained full texts: the AlzPI edge/core study's four blank extraction controls and 86-samples/31-specimens design (the review follows the paper's Methods, which disagree with the paper's own abstract); the host-aware null's mock-community recovery failure (Listeria absent, Bacillus under-recovered, no mechanical lysis), used to bound the null in both directions; the 2013 transfer's heated/unheated donor assignment, which makes the review's donor-confound objection a real design limitation; Emery 2017's own statement that normalized re-amplified libraries were not designed to measure bacterial numbers; Alonso 2018's multiple PCR negative controls, laminar-flow sterile handling and 4.5 h mean postmortem interval, with the review correctly hedging that *sequenced* blank libraries remain unverified; and the salmonid study's perfusion, lysis-buffer-only negative plates, CSF controls, non-matching environmental taxa and two-fish comparative genomics.

| # | Severity | Location | Original to corrected | Evidence |
|---|---|---|---|---|
| 1 | **Major - regression** | spliced Consensus provenance section; consensus-addendum.md | "The earlier adversarial approval refers to its original edition" to "No earlier adversarial approval is attached to this case" | No earlier approval exists anywhere for this case; spliced in unconditionally on 9 October and absent from the earlier edition. |
| 2 | **Major** | AT A GLANCE; Final score / M-S-D; scorecard momentum | "Decisive work underway" to "Mechanistic only" | The artifact named only completed publications and conceded no recruiting trial; the framework requires work being run that could cross the next threshold. Momentum modifies no letter. |
| 3 | Minor | Score card final_score.confidence_rationale (x3) | access depth struck; band re-derived from evidence quality | Lead policy two-step test; documented evidence-quality concerns are valid determinants, so Moderate stands rather than rising to High. |
| 4 | Editorial | inline citation runs (6 prose, 8 score card) | "PMID33771222,25387460,..." to "PMIDs 33771222, 25387460, ..." | Known rendering defect; no citation added, removed or re-ordered. |

Notably **not** findings: the "Best-supported rival" field correctly names the rank-1 non-nominal model, and the REVIEWER-NOTES / Consensus contradiction is absent from this case.

**Divergent copy left untouched:** hsef_v53_popularity_ranks_31_35_2026-10-08/resident-brain-microbiome/resident-brain-microbiome.review.md (it carries the same Momentum error and compressed PMID runs, but not the false-approval clause).

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