# Adversarial review log — antimicrobial-amyloid (2026-10-09 revised edition)

- **Authoring model:** Sol 6.1 (HSEF v5.3 pipeline)
- **Adversarial reviewer:** Claude Opus 5.5 (Anthropic)
- **Date:** 2026-10-14 (round 2)
- **Canonical file reviewed:** `consensus_consistency_audit_2026-10-09/revised/antimicrobial-amyloid/antimicrobial-amyloid.review.md`
- **Level before → after (this round):** Β (Moderate, downstream consequence) → Β (Moderate, downstream consequence) — unchanged
- **Corrections applied:** 3 (0 Critical, 0 Major, 2 Minor, 1 Editorial)
- **PMIDs verified:** 61/61 title + first author + journal + year; 61/61 DOI strings; 0 failures; 0 retractions or expressions of concern
- **Load-bearing claims spot-checked:** 10 (all supported)
- **Status:** Approved as revised

## Standing of the 2026-10-06 approval

The 2026-10-03 edition was adversarially reviewed and approved on 2026-10-06 with 0 corrections. The 2026-10-09 revision is a **re-authored edition, not a patch**, and it dropped both the approval banner and the approval section. The earlier approval therefore did not transfer, and was not assumed: the current edition was re-verified from scratch before the record was restored.

### What the revision changed

| Item | 2026-10-03 (approved) | 2026-10-09 (current) | Assessment |
|---|---|---|---|
| Title | "Antimicrobial amyloid and the interpretation of AD deposits" | "Antimicrobial Amyloid Hypothesis" | Neutral |
| Headline | Γ — Mechanistically plausible, partial system support | **Β — Associational** | **Not a regression**; better grounded (see below) |
| Causal role | epiphenomenon (step 2) | **downstream consequence (step 1)** | **Not a regression**; corrected — a dependency *is* demonstrated (infection → deposition in the positive models), so step 2 cannot be the first match |
| Subclaims | 5 | 6 (adds explicit human SC6) | Improvement — the human leg is now graded in its own right |
| Sources / claims | 60 / 66 | 61 / 68 | Consistent with banner, ledger and provenance header |
| Reference set | — | 37 of 60 replaced; 23 shared | Legitimate re-scoping |
| Literature cutoff | 2026-10-03 | 2026-10-08 | Later, as expected |
| M / S / D | Mechanistic only / Stable / Contested | unchanged | Correct; artifact names real mechanistic work |
| Adversarial record | present | **dropped** | **Administrative regression — fixed here (finding C1)** |

**Why Γ → Β is defensible and was not reverted.** Both editions preregister weakest-link aggregation. This edition isolates the human leg as SC6 ("Human Alzheimer plaques form around real pathogens as a net protective response", scoped to biomarker-defined sporadic AD with infection ascertainment and a clinical horizon) and caps it at Β by quoting the rule *"association without temporal or perturbational proof caps at Β"*. The human evidence is pathogen/DNA colocalization and culture detection, against null detections and contamination/processing boundaries, with no study showing that plaque entrapment improves a clinical or tissue outcome. The framework's Β definition explicitly includes pathology-correlation claims, so the cap is correctly applied and weakest-link follows it. Changing the letter back would require abandoning the preregistered rule or lifting a correctly quoted cap, which the brief forbids.

**Detail lost without error.** The lanabecestat DAYBREAK amyloid-PET sign correction (+2.2 rather than −2.2 centiloids) and Soscia's "eight of twelve" organism count — both verified in the 2026-10-06 pass — are absent from this edition. Neither omission makes any current statement wrong (BACE-inhibitor evidence is retained via PMID 33049114; "several organisms" is accurate), so this is recorded as a loss of precision, not a correction.

## Findings

| ID | Severity | Location | Original → corrected | Evidence |
|---|---|---|---|---|
| C1 | Minor | Header blockquote; section before REFERENCES | Missing adversarial-review record restored, with the Γ→Β and role changes re-adjudicated rather than assumed | The 2026-10-09 revision dropped the 2026-10-06 banner and approval section while changing the headline letter, the causal role and most of the reference set. |
| C2 | Editorial | 23 lines | Lost inter-word spaces restored: "TASTYseries", "Prosswimmer2024", "Kumar2025", "whereas5XFAD", "twoHSV", "Phase2donanemab", "rank42", "atΒ", "retainΓ", "nativeAβ", "adjacentAβ", "oligodendrocyteAβ", "Consensus2Pro,0Deep,0API" | Known upstream builder defect. Whitespace-only repair verified by whitespace-stripped comparison; 5XFAD, Tg2576, EV-A71, miRNA-146a, HHV6A, APP/PS1 and 16S were protected from the repair. |

| C3 | Minor | scorecard `final_score.confidence_rationale` | Access depth removed as a stated basis for the band | The framework ties confidence to evidence density and reproducibility, not to the reviewer's access. Re-derived on legitimate grounds — 61 reports, but the decisive HSV experiments conflict, the human detection literature splits between positive and null/contamination-bounded series, and nothing has been reproduced in the Aβ-selective form the claim needs — the band remains **Moderate**; only its basis is corrected. Historical `score_before`/`score_after` snapshots inside `consensus_validation` left as-is. Lead policy of 2026-10-14 (two-step test) applied: access depth struck, band re-derived. Step 2 keeps **Moderate** — the decisive HSV experiments conflict, the human detection literature splits between positive series and null or contamination-bounded ones, and nothing has been reproduced in the Abeta-selective form the claim needs — which are legitimate determinants; the High band's "extensive independent replication across labs, models, and assay modalities" is not met, and access depth was not the only thing holding the band down. |

## Rejected findings (checked, no change)

1. Weakest-link aggregation applied as written; all three alternatives reported and all return Β; SC1–SC5 reported separately at Γ.
2. Β/Γ boundary for the conjunction — see above; the Γ legs are real but the conjunction is gated by its weakest link by preregistration.
3. PMID 32336252 typed "Comment" by PubMed but counted as primary — it is an original reanalysis of the primary sequencing data and is described as such.
4. Anti-amyloid RCTs used as counterevidence rather than positive crossings — correct handling.
5. PubMed title fields for references 8 and 21 drop italicised "Aβ1-42" spacing and the species name "E. coli"; the review prints the full correct titles. Standing convention.
6. Validator warning `v5.claim_form` ("entry_form is 'phenomenon'") — the registered claim states function **plus** consequence, the documented exception.
7. The 9 October 2026 provenance-reconciliation block inside the framework section — accurate as written (it states it changes no score or count, and that the earlier approval refers to the original edition) and retained.
8. The spliced-consensus-block versus preserved-notes self-contradiction reported elsewhere in this audit batch — **checked, absent here.** The preserved `METHODS AND CONSENSUS CONTRIBUTION` section already records signed-in Consensus browser use for discovery, counterevidence and attempted indexed full-text checks, so it does not contradict the spliced block. Nothing was scoped or deleted.
9. **Confidence band** — re-derived independently of access depth (finding C3); remains Moderate on density and reproducibility alone.
10. **Renderer defect (nominal printed as its own "Best-supported rival") — checked, absent.** `models_evaluated[0]` is rank 1 with `is_nominal` false; the nominal is rank 5.

## Validator

`hsef-hypothesis-review-v5/tools/validate_review.py` on the scorecard + revised prose: **33 pass, 1 warn, 1 fail**. The warn is `v5.claim_form` (rejected by standing convention); the fail is `v5.lay_summary`, marked by the validator as "mandatory from v5.4" and not applicable to this v5.3 package. Output in `validator.txt`; pre-review `validation-report.json` preserved in `sol61-original/`.

## Files changed

- `antimicrobial-amyloid.review.md` (banner, approval section, whitespace)
- `antimicrobial-amyloid.scorecard.json` — `final_score.confidence_rationale` only (surgical text edit, formatting preserved). Level, confidence band, role, momentum and counts unchanged.

## Propagation

The inventory records no `propagate_to` copies for this slug. The divergent copy at `deliverables/hsef_v53_popularity_ranks_41_45_2026-10-08/antimicrobial-amyloid/antimicrobial-amyloid.review.md` was **not** modified, as instructed. The earlier approved edition at `deliverables/hsef_v53_neurodegeneration_2026-10-03/antimicrobial-amyloid/` was left untouched; it keeps its own 2026-10-06 approval, which refers only to that edition.