# Adversarial review log — inflammation-depression **Case:** inflammation-depression (`hsef_v53_remaining_priority_06_10_2026-10-09/`) **Authoring model of record:** Sol 6.1. The package's own `review_metadata.reviewer` is "Codex" and `provenance_header.generated_by` is "Codex; HSEF v5.3"; recorded, not edited. **Adversarial reviewer:** Claude Opus 5.5 (Anthropic) **Date:** 2026-10-14 · **Round:** 2, group 1 ## Summary | Item | Before | After | |---|---|---| | Headline level | Γ — Mechanistically plausible, partial system support | Γ — unchanged | | Confidence | Moderate | Moderate — unchanged | | Causal role | reinforcing / stabilizing factor | unchanged | | Subclaim levels | SC1 Β · SC2 Ε · SC3 Γ | unchanged | | Aggregation | primary-subclaim (SC3); weakest-link Β, scope-matched Γ | unchanged | | Momentum / Stability / Divergence | Decisive work underway / Uncertain / Contested | unchanged | | Sources / claims | 56 / 60 | 56 / 60 | | Validator (v5) | PASS — 0 fail, 0 warn, 31 pass | PASS — 0 fail, 0 warn, 31 pass | **Findings by severity:** Critical 0 · Major 0 · Minor 2 · Editorial 1. Rejected findings: 8. **PMIDs:** 56/56 reference PMIDs verified; 6/6 prose- and scorecard-only PMIDs verified (three Published Errata, two superseded preprints of one reference, one trial protocol); 0 failures; no retraction or expression of concern in any of the 62 records. **Claims spot-checked:** 12. Eleven matched exactly; the twelfth produced finding D-3. ## Findings | ID | Severity | Location | Original → corrected | Evidence | |---|---|---|---|---| | D-1 | Minor | Document-wide | 74 × `PMID` → `PMID ` | The whitespace-loss defect named in the round-2 brief; the document already used the spaced form 69 times in its reference list. | | D-2 | Editorial | Momentum paragraph | `DOI10.1016/j.bbih.2024.100871` → `DOI 10.1016/…` | Same defect class. The identifier itself is correct and resolves to the INSTA-MD protocol. | | D-3 | Minor | "Evidence and causal interpretation", immune-challenge paragraph | Sentence now states the positive limb first: "its prespecified primary contrast, the baseline-to-1.5-hour SHAPS change in the high-CRP versus low-CRP endotoxin groups, was significant, while the overall three-way primary interaction was not" | The trial's prespecified primary outcome was the SHAPS and its prespecified primary contrast was exactly that comparison, reported as significant. The original prose named only the null limb, so a prose-only reader could conclude the primary analysis failed. A balance gap, not a misstatement: the evidence card for that source already records the positive finding, and the rival-model entry credits the trial with separating genuine immune-triggered acute anhedonia from a purely shared-cause account. The correction makes the prose match the card. | Plus the mandated banner under the `> Sources:` line and a new `## ADVERSARIAL REVIEW AND APPROVAL` section before `## REFERENCES`. **No level changed.** D-3 does not move a letter: the review already treats the challenge as acute-susceptibility evidence that cannot establish sustained-MDD initiation, which is what holds SC3 at Γ. ## What this review gets right and was checked hardest The aggregation is the exact pattern the round-2 brief warns about — a subclaim set whose highest member is an Ε — and it is handled correctly rather than misread. The pre-registered rule is primary-subclaim, the primary subclaim (SC3) is the registered claim verbatim, and SC2's Ε is explicitly scope-restricted to prolonged interferon treatment in a defined medical setting, so it cannot set the broadly scoped headline. Scope-matched returns Γ for the same reason, weakest-link returns Β, and all three are reported. The Momentum artifact is likewise the opposite of boilerplate: INSTA-MD is named with registry ID, protocol PMID (verified), the date the registry was read, the fact that the registry's last verification predates that read, and an explicit refusal to assert completion or results. ## Rejected findings R-1 … R-8 are tabulated inside the review. The ones worth the lead's attention are R-2 (the Ε-subclaim aggregation, above), R-3 (Momentum), and R-6 — the errata and the two preprints of reference 11 appear nowhere in the numbered list, which is correct ledger discipline rather than an omission. ## Files changed in this case folder - `inflammation-depression.review.md` — edited (3 corrections, banner, approval section) - `validator.txt` — new (v5 validator output for the revised edition) - `ADVERSARIAL_REVIEW.md`, `adversarial-findings.json` — new - `sol61-original/` — pre-edit copies of `inflammation-depression.review.md`, `inflammation-depression.scorecard.json`, `validation-report.json`, `consensus-validation.json`, `SHA256.txt` `inflammation-depression.scorecard.json` was **not** changed: no level, confidence, count or role moved. It does carry the same `DOI10.1016/...` fusion in its Momentum artifact and justification strings, and its `what_it_does_not_show` text for the high-CRP challenge is the source of D-3's wording; the lead should decide whether to mirror those two fixes into the JSON or regenerate. ## Package-level propagation needed from the lead No headline string, letter, confidence, count or title changed, so `OVERVIEW.md`-class values are not stale. Needed: 1. Replace the `inflammation-depression.review.md` digest in this case folder's `SHA256.txt` and in any package `MANIFEST.json` / `BATCH_COMPLETION.json` entry that carries it: `87ea8f5285c17043e69b8952173137b23271985aa88238e2d4f3bbdb69e4dda4` → `13f128d497aeebd050d4e5176849fd44952cd487b1b646a2f29264ead7f91a52`. 2. **Batch-wide:** the `PMID` fusion is present in all five case files of `hsef_v53_remaining_priority_06_10_2026-10-09` (dopamine-schizophrenia 73, hiv-aids 52, inflammation-depression 74, lipid-heart-diet 55, telomere-shortening 59) and in the scorecard JSONs, alongside the spaced form in the same documents. A regeneration pass over the batch is the clean repair. No `propagate_to` or `divergent_copies` entries exist for this slug. ## Lead-flagged systematic artifacts — per-case disposition The lead reported four systematic artifacts found by other groups. All four were checked against this case. | Artifact | Present here? | Action | |---|---|---| | **(1) False approval boilerplate** — the Oct-9 `integrate_consensus.py` appends "The earlier adversarial approval refers to its original edition" unconditionally | N/A | This case is not under `consensus_consistency_audit_2026-10-09/revised/`; no spliced Consensus block exists. | | **(2) Confidence band justified by reviewer access depth** | **Yes** | Corrected as finding **D-5** in the scorecard rationale. **Band unchanged at Moderate** — argued both ways; the High text is not met. | | **(3) Renderer bug** — "Best-supported rival" printing `models_evaluated[0]`, often the nominal | No | `models_evaluated[0]` is the rank-1 non-nominal model and the prose names it correctly. | | **(4) Reviewer-notes contradiction and false provenance clauses** | **Yes** | Corrected as finding **D-4**: the false "existing HIV draft consulted after a source-based checkpoint" clause was removed. It is batch boilerplate, accurate only for `hiv-aids`. | Per the lead, the scorecard `review_metadata.reviewer` string was treated as house provenance and left untouched. ## Lead policy applied — confidence bands and momentum (2026-10-14) - **Confidence two-step test:** step 1 (strike access depth as a determinant) applied; step 2 (re-derive) -> **Moderate stands**, band unchanged. - **Momentum stricter rule** ("Decisive work underway" requires work actually being run; a completed publication never satisfies it): **holds**. - Discordant MR, few trials with pleiotropic agents, one prospectively stratified test, and CRP as a reporter readout all survive step 2. Momentum rests on INSTA-MD, recorded as RECRUITING at the read date. Full derivation is recorded in the review's `## ADVERSARIAL REVIEW AND APPROVAL` section and in `adversarial-findings.json` -> `lead_policy_2026_10_14`. ## Consensus question for the lead In adults with major depressive disorder stratified **prospectively** by a pre-fixed baseline inflammatory marker threshold, does an anti-inflammatory or immune-targeted augmentation produce a significant treatment-by-stratum interaction on depressive symptoms? A replicated yes lifts SC3 off Γ; the only prospectively stratified trial in this evidence base (PREDDICT, one drug, n=119) is null.