# Adversarial review log - vaccine-autism-avoidance

Authoring model: **Sol 6.1**. Adversarial reviewer: **Claude Opus 5.5 (Anthropic)**. Date: **2026-10-14**.
Round 2 of the Sol 6.1 HSEF v5.3 adversarial pass.

| Item | Value |
|---|---|
| HSEF level before -> after | A (Speculative / narrative) -> A (unchanged) |
| Confidence before -> after | High -> High (unchanged) |
| Causal role | epiphenomenon (unchanged) |
| Findings | Critical 0 - Major 0 - Minor 4 - Editorial 0 |
| PMIDs verified | 62/62 reference PMIDs plus 4 in-text identifiers; 0 failures |
| Claims spot-checked | 16 load-bearing numeric/design claims against original abstracts; all matched |
| Validator | FAIL on `v5.lay_summary` only - the shipped validator enforces the v5.4 lay-summary rule against a self-declared v5.3 review. Cross-version artifact, not a defect of this review. 32 checks pass, 0 warnings. |

## Findings

| ID | Severity | Location | Correction |
|---|---|---|---|
| AR-1 | Minor | Throughout | ~127 renderer-dropped spaces repaired (numerals, units, acronyms, PMID link text, one Greek letter). Whitespace only. |
| AR-2 | Minor | REFERENCES 22, 25, 29, 35 | Empty `DOI: .` fields replaced with "No DOI in PubMed record." (verified absent in all four PubMed records). |
| AR-3 | Minor | Evidence cards [4], [22], [39], [45], [54], [55] | Erratum strings normalised; the four "ErratumIn: PMIDnot supplied" cases now carry the actual journal citation, verified as genuinely PMID-less in PubMed. |
| AR-4 | Minor | scorecard `final_score.confidence_rationale` | Reviewer access depth removed as a stated input to the confidence band, per the framework's Confidence band section. Band unchanged (High). |

No level, confidence, role, momentum, stability, divergence, subclaim level, source count or claim count changed.

## Rejected findings

Headline A (cap 2 is a ceiling, not an earned rung); SC4 at H (independence basis survives the v5.2 shared-failure test; Theta correctly withheld); aggregation arithmetic under all three rules; "Best-supported rival" (scorecard confirms `models_evaluated[0]` is the rank-1 non-nominal, so the known renderer defect does not fire); Momentum artifact named; no dropped collective authors; rubric-07 boilerplate declared rather than silently resolved; exclusion of the retracted PMID 21058170 and the 1998 Wakefield report from `sources_total`; no false-approval boilerplate and no REVIEWER NOTES / Consensus-section contradiction in this package.

## Harm-and-refutation check (requested for this case)

The refutation is stated, not hedged: the claim is graded at the lowest rung, the Falsification section records that the predicted dose, formulation and withdrawal signatures *fail* in the stronger clinical exposure tests rather than merely being untested, and the Consensus reconciliation names the retraction of PMID 21058170 and excludes the 1998 Wakefield report. Harm attribution is correct in both directions: measles and pertussis consequences are sourced to the actual exemption cohorts, undervaccination gradients and laboratory-confirmed outbreaks (PMID 11135778, 10404911, 19482753, 20048244, 24019039, 24913790, 27705270, 32160662, 28704350, 30073293) rather than to reviews; and the vaccine-side harms that are real - anaphylaxis at 1.31 per million doses, the MMR ITP window, the first-dose MMRV febrile-seizure excess - are retained at their own scope and explicitly not converted into autism evidence.

## Files changed

- `vaccine-autism-avoidance.review.md` (banner, ADVERSARIAL REVIEW AND APPROVAL section, AR-1/2/3)
- `vaccine-autism-avoidance.scorecard.json` (AR-4)
- `validator.txt` (new)
- `sol61-original/` holds the pre-review copies.

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