# Adversarial review log — cancer-immunoediting (revised edition)

- **Authoring model:** Sol 6.1
- **Adversarial reviewer:** Claude Opus 5.5 (Anthropic)
- **Date:** 2026-10-14
- **Canonical file:** `consensus_consistency_audit_2026-10-09\revised\cancer-immunoediting\cancer-immunoediting.review.md`
- **Divergent copy, NOT edited:** `hsef_v53_popularity_ranks_36_40_2026-10-08\cancer-immunoediting\cancer-immunoediting.review.md` (the pre-revision edition; it still carries the F1 momentum defect and the F2-F5 rendering defects)
- **Level before → after:** Β — Associational → **Β** (unchanged)
- **Confidence before → after:** Moderate → Moderate (band unchanged; rationale rescoped, see F7)
- **Causal role:** reinforcing / stabilizing factor (unchanged)
- **Momentum before → after:** Decisive work underway → **Mechanistic only**
- **Stability / Divergence:** Stable · Contested (unchanged)
- **PMIDs checked:** 65/65 resolved (63 references + body-only 36993286, 42155448); 0 failures, 0 retractions or expressions of concern
- **Load-bearing claims spot-checked:** 9
- **Validator:** FAIL (1) + WARN (1) — `v5.lay_summary` (mandatory only from v5.4, which post-dates this v5.3 review) and `v5.claim_form` "entry_form is 'phenomenon'", which the standing conventions reject because the registered claim states the sequence **plus** the selection consequence. 33 checks pass. **Identical on the pre-edit file**, so neither is a regression. Output saved to `validator.txt`; the original package report is preserved at `sol61-original/validation-report.json`.

## Regression check against the earlier edition

A line-by-line diff of this revised edition against the 2026-10-08 edition shows the revision added
exactly one block, `### Consensus provenance reconciliation — 9 October 2026`, and changed nothing
else. **The revision introduced no error.** In particular, the self-contradiction reported elsewhere
in this batch — a spliced Consensus block beside preserved notes denying that Consensus work was
done — does not occur here: the preserved `## METHODS AND CONSENSUS CONTRIBUTION` section already
states that Consensus was used through the signed-in browser with zero API calls. Findings F1-F5 below are **inherited** from the earlier edition. F6 **is** a revision regression:
the splice asserted that an earlier adversarial approval exists for this case, and none does.

## Findings by severity

| Severity | Count |
|---|---|
| Critical | 0 |
| Major | 1 |
| Minor | 5 |
| Editorial | 1 |
| **Total corrections applied** | **7** |

## Findings table

| ID | Severity | Location | Correction | Evidence |
|---|---|---|---|---|
| F1 | Major | AT A GLANCE; Final score / M-S-D; score card `extended_scoring_MSD.momentum` | Momentum "Decisive work underway" → **"Mechanistic only"**, artifact rewritten | The framework defines the value as "someone is running an experiment that could cross the next threshold", but every named artifact is a completed published study, and the review's own Stability entry says no named pending result is asserted. Momentum modifies no letter. Inherited, not a revision regression. |
| F2 | Minor | Decisive-perturbation items [2], [6], [30], [31], [32], [33], [57] | The "shows" and "does not show" halves had been concatenated with no separator or terminal stop; a sentence break plus an explicit "Does not show:" lead-in was restored from each matching evidence card | Rendering defect. The halves are stored separately in the score card and separated correctly in the cards, so only the join was repaired. |
| F3 | Minor | Evidence card [51] and matching score-card fields | "ErratumIn: PMIDnot supplied." → "ErratumIn: PMID not supplied in the PubMed record." | Known builder defect; REVIEWER NOTES already records that the IFN erratum lacks a resolved PMID. |
| F4 | Minor | REFERENCE 45 | "DOI: ." → "DOI: none in PubMed record." | PubMed confirms PMID 10229818 carries no DOI. |
| F5 | Editorial | Final score / M-S-D | "steps3/4" → "steps 3/4" | Missing-space rendering defect. |
| F6 | Minor | Consensus provenance block (review + `consensus-addendum.md`) | "The earlier adversarial approval refers to its original edition" → "No earlier adversarial approval is attached to this case" | **Revision regression**, not inherited: the clause was spliced in on 9 October and is false here — the 2026-10-08 edition carries no adversarial approval section or banner. Rescoped, not deleted. |
| F7 | Minor | Score card `final_score.confidence_rationale` | Rationale rested partly on "incomplete access" → rewritten to rest only on evidence density and reproducibility | The framework ties confidence to evidence density and reproducibility, not reviewer access. The band was re-derived and **stays Moderate**: no replicated human inception dataset for the universal subclaim, repeated TRACERx/TCGA reuse, heterogeneous controls — the Moderate text, not the High text. |

Rejected findings — including the aggregation re-derivation against the index entry, the unenumerated
chromosomal-instability-passenger sixth model, the causal-tally phrasing, the erratum and preprint
linkages, the seven mid-list consortium author entries and the `review_metadata.reviewer` field —
are recorded in the `## ADVERSARIAL REVIEW AND APPROVAL` section of the review itself.

## Files changed in this case folder

- `cancer-immunoediting.review.md` — F1-F5, banner, `## ADVERSARIAL REVIEW AND APPROVAL`
- `cancer-immunoediting.scorecard.json` — F1 (momentum value and artifact), F3, F5 mirrored (no level, confidence, subclaim level, aggregation rule, role, stability or divergence value changed; CRLF line endings preserved)
- `validator.txt` — new
- `ADVERSARIAL_REVIEW.md`, `adversarial-findings.json` — new
- `sol61-original/` — pre-edit copies

## Propagation

`propagate_to` is empty for this slug, so nothing was copied. The 2026-10-08 edition is a declared
divergent copy and was deliberately left untouched.

## Approval

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