# Adversarial review log — creatine-cognition

Authoring model: **Sol 6.1**. Adversarial reviewer: **Claude Opus 5.5 (Anthropic)**. Date: **2026-10-14**.
Round 2; group 7 of `NEW_REVIEW_INVENTORY_2026-10-14.json`.

## Summary

| Item | Value |
|---|---|
| Level before | Ε — Context-specific causality established / Moderate |
| Level after | Ε — Context-specific causality established / Moderate (unchanged) |
| Findings: Critical | 0 |
| Findings: Major | 2 |
| Findings: Minor | 2 |
| Findings: Editorial | 0 |
| Rejected findings | 6 |
| PMIDs checked / failed | 61 / 0 |
| Load-bearing claims spot-checked | 15 |
| Validator | FAIL (1 fail, 0 warn, 33 pass) — `v5.lay_summary` only; identical before and after (validator 2.4 enforcing a v5.4 rule on a v5.3 package). |
| Approval | **Approved as revised.** |

## Findings

| ID | Severity | Location | Original → corrected | Basis |
|---|---|---|---|---|
| AR-1 | Major | Prose review and score card, throughout | 141 numeral-adjacent fused tokens repaired ("includes61" → "includes 61", "preregistered123-person" → "preregistered 123-person", "at5g/day" → "at 5 g/day", "no24week" → "no 24-week", "sixweeks" → "six weeks", "reducedATP" → "reduced ATP", "olderwomen" → "older women", and so on) | The known "randomize3808" rendering-defect class, at scale. Identifiers (study_id, DOI, PMC, PMID, NCT, rubric filenames, subclaim labels) were excluded and verified unchanged. No wording or claim altered. |
| AR-2 | Major | AT A GLANCE table; final-score block; score card momentum | Momentum "Decisive work underway" → "Mechanistic only"; artifact rewritten | All three named works (PMIDs 42075005, 40854087, 42578920) are completed published reports in new populations; none supplies the review's own quoted Ζ requirement (independent multi-site preregistered replication, blinding-integrity reporting, a mechanism-narrowing result). The Stability entry already said no such pending result exists, contradicting the Momentum value. Momentum modifies no letter. |
| AR-3 | Minor | Final-score block, Divergence line | Prose given the artifact the score card already carried (the two conflicting framings) | Divergence requires a checkable artifact; "Contested" may lack an implied level and arithmetic but must still name the split. Value unchanged. |
| AR-4 | Minor | Score card `final_score.confidence_rationale` | Access depth removed as a confidence determinant | The framework ties confidence to evidence density and reproducibility; the boilerplate rationale named "incomplete access" as a determinant. Rewritten to rest on density, replication and heterogeneity. Band re-tested against both band texts and unchanged. Moderate stands: multiple independent trials with some replication, but the largest preregistered trial conflicts with the early large effect and heterogeneity is substantial. |

## Rejected findings (checked, no change)

1. The level. Δ rejected because the context boundary is mapped, not untested (negative controls on both sides); Ζ rejected because the positive human results are one oral-supplement modality with a shared expectancy/practice/endpoint-selection failure mode, while spectroscopy is target engagement and the congenital/animal work is out of scope. Ε / Moderate stands.
2. Aggregation. Primary-subclaim pre-registered, SC2 named, alternatives reported (weakest-link Γ, scope-matched Ε); SC2's scope matches the central claim's, so this is not the "highest subclaim" misread.
3. Fourteen reference years one later than the PubMed e-pub year — print/e-pub convention.
4. Collective-author leading-comma defect, empty `DOI: .` fields, retrieval handles, "PMID not supplied" strings — screened, none present.
5. Rubric routing through rubric07 + rubric06 for a nutritional cognitive claim — adaptation declared, literal criteria quoted unchanged.
6. Lead-reported artifacts screened: the renderer defect that prints the nominal as the best-supported rival does not apply here (`models_evaluated[0]` is the rank-1 non-nominal model); no false-approval boilerplate (not a revised edition); no reviewer-notes/provenance contradiction. The band was re-tested and Moderate stands.

## Attribution audit

15 load-bearing claims checked against the primary records; all matched, including the ones most liable to drift: Rae 2003's 45 vegetarians at 5 g/day for six weeks (14561278); the 123-person preregistered crossover's BDS-borderline, RAPM-null, no-vegetarian-advantage result (37968687); McMorris benefit confined to complex central-executive tasks (16416332, 17046034); hypoxia attention restoration with 9.2% brain-creatine rise (25632150); the rested young-adult nulls (18579168, 37759877); the 56-woman four-arm older-women null (24098469); regionally dependent reversible brain-creatine rise (10484486); PCr/ATP shift driven by falling ATP, not an ATP boost (17185404); 36-woman menopause formulation trial (40854087); exploratory middle-age trial (42578920); and the single 20-patient AD cohort behind three reports (40395689, 41737290, 40520504). No citation inversion found.

## Residual limitations

The same builder defect also produced word-to-word fusions with no numeral anchor ("wellpowered", "lowerbrainstores", "restedomnivoretrial", "samepersonbrainmeasurement", "codedcreatine", "andwithouttraining"), concentrated in the evidence cards, reconciliation records and timeline. These cannot be repaired by exact substitution without risking a change of meaning and are referred to the package builder; the document is cleaner but not fully clean. Validator FAIL is a framework-version mismatch.

## Strings for the lead to propagate to package-level files

| Old | New |
|---|---|
| `Decisive work underway` (creatine-cognition row, Momentum column) | `Mechanistic only` |

No level, confidence, role, source count or claim count changed. If OVERVIEW.md / READER.html / EDITORIAL_HANDOFF.md reproduce any of the fused-token prose repaired under AR-1, those strings should be re-rendered from this corrected review rather than patched individually.

**Lead policy recorded (confidence bands, 2026-10-14).** The lead adjudicated the question raised in this pass and set a two-step test for the corpus. Step 1: strike access depth as a determinant everywhere — how much of a paper the reviewer could read is neither evidence density nor reproducibility, and it never justifies a band. Step 2: re-derive the band from what remains; if access depth was the only thing holding the band down and the cited evidence meets the High text, raise it, but if the review documents substantive evidence-quality or reproducibility concerns (indirect effects, incomplete washout, residual confounding of the mechanism, reporter rather than endogenous readouts) those are valid determinants and the band stands. A reviewer's uncertainty about the evidence is a legitimate reason for Moderate; a reviewer's uncertainty about their own reading access is not. This pass's rationale rewrite is step 1; the band decision below is step 2. Step 1 applied (rationale rewritten under AR-4). Step 2: band stands at Moderate on valid determinants — the largest preregistered trial conflicts with the early large effect and methodological heterogeneity is substantial, which is the Moderate text, not access depth.

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