# Adversarial review log — microbiome-causality (revised edition)

Authoring model: **Sol 6.1**. Adversarial reviewer: **Claude Opus 5.5 (Anthropic)**. Date: **2026-10-14**.
Round 2 of the adversarial pass (brief: `deliverables/ADVERSARIAL_REVIEW_BRIEF_2026-10-06.md`, ROUND 2 section).

| Item | Value |
|---|---|
| Level before | Γ — Mechanistically plausible, partial system support |
| Level after | Γ — unchanged |
| Confidence | Moderate — unchanged |
| Role | reinforcing / stabilizing factor — unchanged |
| M / S / D before | Decisive work underway / Stable / Contested |
| M / S / D after | **Mechanistic only** / Stable / Contested |
| Subclaims | SC1 Β, SC2 Ζ, SC3 Ε, SC4 Γ — all unchanged |
| PMIDs checked | 67 / 67 resolve to the stated title, first author, journal and year |
| PMIDs failed | 0 |
| Retractions / EoC | none found |
| Load-bearing claims spot-checked | 15 |
| Findings | 0 Critical, 0 Major, 2 Minor, 1 Editorial (13 instances) |
| Validator | PASS — 0 fail, 0 warn, 30 pass (`validator-adversarial-2026-10-14.txt`) |

## Revision-regression check

This file is the revised edition of the earlier website review. Diffed against the earlier edition at
`hsef_v53_popularity_ranks_41_45_2026-10-08/microbiome-causality/microbiome-causality.review.md`: the revision
added only the "Consensus provenance reconciliation — 9 October 2026" block and changed nothing else.
**One regression was introduced by the revision** — finding C3, the false "earlier adversarial approval"
clause, which the splice inserted unconditionally into a case that has no earlier adversarial approval. The
Momentum defect (C1) and the rendering defects (C2) were inherited from the original edition.

Two systematic artifacts reported by the lead were checked and are **absent** here. The reviewer-notes
contradiction is absent: the preserved METHODS AND CONSENSUS CONTRIBUTION and REVIEWER NOTES sections both
affirm Consensus use, so nothing contradicts the spliced validation block and no scoping edit was needed.
The "Best-supported rival" renderer bug is absent: `models_evaluated[0]` is "Diet/substrate-dependent
microbial contribution" with `is_nominal: false`, and the nominal ("Primary gut-microbiome driver") is the
fifth entry, so the prose names the correct rank-1 rival.

## Findings

| ID | Severity | Location | Change |
|---|---|---|---|
| C1 | Minor | AT A GLANCE table; Final score / Momentum; score card `extended_scoring_MSD.momentum` | "Decisive work underway" → "Mechanistic only". The artifact itself ended "no claim of a currently recruiting trial", and Stability adds that no named pending result is claimed — both contradict the framework's definition, "Someone is running an experiment that could cross the next threshold." The four named studies are completed placebo-controlled trials in established disease, none of them the before-disease factor manipulation that this review's own SC4 Δ requirement names. Momentum modifies no letter. |
| C2 | Editorial | Throughout (13 instances) | "Universal cap5" → "Universal cap 5"; "phase2 placebo-controlled" → "phase 2 placebo-controlled" (×2); "ErratumIn:" → "Erratum in:" (×8), two of which also read "PMIDnot supplied" → "PMID not supplied". |
| C3 | Minor | Consensus provenance reconciliation block, final paragraph; `consensus-addendum.md` | "The earlier adversarial approval refers to its original edition" → "No earlier adversarial approval is attached to this case". False provenance boilerplate inserted unconditionally by the 9 October revision script. Verified before changing: the earlier edition carries no approval section, `reviews/microbiome-causality/` carries no red-team record, and this slug is not among the thirty reviews carrying a 2026-10-06 approval. A revision-introduced regression. |

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

Two-step test as issued by the lead. **Step 1, strike access depth as a determinant.** The score card's `confidence_rationale` named reviewer access ("incomplete access set the confidence"). How much of a paper the reviewer could read is neither evidence density nor reproducibility and never justifies a band, so the sentence was rewritten to drop it and to say explicitly that access depth is recorded per ledger row and is not a determinant. **Step 2, re-derive from what remains.** What remains is substantive and is documented in the review's own text: defined rodent perturbations replicate across distinct mechanisms, but the human record is mixed and the transfer literature shows demonstrated rather than hypothetical reproducibility problems — engraftment without functional effect, and housing- and diet-dependent reversals — with heterogeneous products, comparators and endpoints across the 67 reports. Reproducibility is demonstrably not well established, which is what the High text requires. The band therefore **stands at Moderate**; access depth was not the only thing holding it down.

**Momentum re-checked against the same clarification** (a completed publication or trial never satisfies "Decisive work underway"): this was finding C1, corrected from "Decisive work underway" to "Mechanistic only" earlier in this pass on exactly that reasoning — the four named studies are completed placebo-controlled trials. The lead's clarification confirms the correction.

## Rejected findings (checked, no change)

1. **Headline Γ** — the THEMATIC hepatic-encephalopathy recurrence result is the strongest case for Δ, but the contrast is a secondary endpoint whose pooled any-FMT comparison is post hoc, in patients who already had cirrhosis on maximal therapy. Treating established disease does not test upstream primacy.
2. **Scope-matched aggregation** — checked against the known misreading. SC2 (rodent) and SC3 (prespecified diagnoses under intervention) are both scope-restricted relative to the registered broad-primacy claim; only SC4 matches, and the declared alternatives (weakest-link Β, primary-subclaim Γ) are correct. The derivation's "preregistered human-primacy gate" phrasing reads like primary-subclaim language, but both rules return Γ.
3. **SC2's Ζ** — independence is recorded with three named lines and the distinct systematic error each excludes; they do not share a failure mode.
4. **Cap 5 cited for SC4** — the cap-precedence paragraph resolves the collision explicitly in writing, which is what the framework requires.
5. **Comma-run PMID lists** — used consistently as this document's citation convention, not a sporadic rendering failure; house style is outside the remit.
6. **"remains Gamma" spelled out in prose** — the normative descriptor rule governs the letter and descriptor fields, which are correct.
7. **Collective-author reference defect** — none present.
8. **Divergence "Contested" without arithmetic** — Contested is defined as the case where no single implied level applies.
9. **Eighteen PMIDs shared with `gut-microbiome-systemic-causality`** — see below.
10. **Confidence band Moderate justified partly by access depth.** The score card's `confidence_rationale` reads "67 primary reports with 8 selected partial-body examinations; density, heterogeneous controls and incomplete access set the confidence, not correctness", which names reviewer access as a confidence input although the framework defines confidence by evidence density and reproducibility. The band itself was independently re-derived against the High text ("extensive independent replication across labs, models and assay modalities; reproducibility well-established") and **Moderate is correct**: this review documents demonstrated reproducibility failures in the transfer literature and a mixed human trial record. **Superseded 2026-10-14:** the lead has since directed that the rationale be rewritten per case; that edit is finding C4 and is described under “Lead policy applied” above. The band is unchanged.

## Cross-review check with gut-microbiome-systemic-causality

The two reviews register different claims — compositional primacy across "systemic and neurobehavioural disease" here, versus "incident or progressive disease across human distant-organ systems" there — and each is scored against its own. Eighteen of this review's 67 references also appear among that review's 38; the shared records are used at the same scopes in both (for example the segmented-filamentous-bacteria arthritis experiment supports a rodent subclaim in each). **Neither borrows the other's evidence to clear a threshold.** In particular, neither lets the recurrent *C. difficile* literature carry a systemic claim: this review names it an out-of-scope calibration comparator, and the other scores it as an explicitly out-of-scope subclaim with the index's own instruction not to let it carry the broad claim. Both land at Γ with the rodent subclaim at Ζ and bounded human contribution at Ε, which is consistent rather than borrowed.

## Files changed in this case folder

- `microbiome-causality.review.md` (banner + ADVERSARIAL REVIEW AND APPROVAL section + C1–C3 corrections)
- `microbiome-causality.scorecard.json` (C1: momentum value, artifact and justification)
- `consensus-addendum.md` (C3; sha256 `978821860f5bcfe43e2ec15849c6cf89c2d69802ae55ab38584474266a525f4a`)
- `validator-adversarial-2026-10-14.txt` (new)
- `SHA256.txt` (refreshed; keeps the package's two-entry shape, so the addendum hash is recorded here instead)
- `sol61-original/` holds the pre-review copies of the review, score card, consensus-addendum.md, SHA256.txt and validation-report.json.

The divergent earlier edition at `hsef_v53_popularity_ranks_41_45_2026-10-08/microbiome-causality/` was **not** touched.
Package-level files (`README.md`, `COMPLETION.json`, `FINAL_VERIFICATION.json`) were **not** edited.

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