v2 phase 4: paper-review + humanizer

[neuromechanist]

Summary

v2 Phase 4: invoke /manuscript:paper-review + /manuscript:humanizer on the assembled per-section content. Closes #12.

Skill invocations

Skill	Loaded instructions	Applied to
/manuscript:paper-review	Review process, methodology checklist, output template (Synopsis / Critical / Major / Minor / Editor Note)	/tmp/assembled_for_review.md (concatenated per-section files)
/manuscript:humanizer	29 AI-writing patterns + research-writing calibration	All per-section files individually

Paper-review result

Recommendation: minor revision (TiCS equivalent).

• Critical issues: 0
• Major concerns: 0
• Minor concerns: 6 (highlight specificity, abstract sampling-rate clause clarity, Hickok-critique substance, Section 6 social-cognition density, Section 1 meta-narration, Castelli citation slug dedupe)

All F1-F5 critical findings from the prior v1 self-review are addressed in prose: F1 verbatim phrasing in 5a, F2 Codispoti citation in Section 6, F4 Petroni + pipeline anchor in 7.2 + Box 1, F5 topographic-and-band rejection region in 7.3. Plus I2 (narrative-position objection in 7.4), I3 (Hickok-critique named, though thin), I4 (cohesion bridges), I5 (5a + 5b split with parallel thesis sentences), m8 (Klin autism in limitations), m9 (explicit latency windows).

Humanizer result

No rewrites required. Scanned all 14 per-section files against the 29 humanizer patterns. Zero matches across:

• Pattern 1 (significance inflation), 3 (-ing endings), 4 (promotional), 7 (AI vocabulary), 8 (copula avoidance), 10 (rule of three), 14 (em-dashes), 18 (emojis), 19 (curly quotes), 20 (chatbot artifacts), 23 (filler), 24 (excessive hedging), 25 (generic positive conclusions), 27 (persuasive tropes), 28 (signposting)

Two scientific uses of "robust" (Section 4 hippocampal responses; Section 6 alpha desynchronisation) are citation-anchored technical descriptions, not significance inflation. Methods-adjacent passive voice (Box 1 pipeline; Section 3 paragraph 3 eye-tracker constraint) is acceptable per skill's research-writing calibration. Single hedges in Section 7 ("may therefore be confounded", "is the response") are legitimate research-writing usage.

Why humanizer was a no-op

The manuscript was drafted in Phase 2 against an upstream style-guide layer (CLAUDE.md bans em-dashes, emojis, AI attribution) plus v1 self-review carry-forwards already addressing the worst tells. The humanizer pass functions here as an audit confirmation rather than a rewrite step. This is the demo's observable: when upstream discipline is in place, the finishing-pass skill becomes redundant; when it isn't, the finishing-pass skill catches what the upstream missed.

Output

• manuscript/narrative-review/reviews/internal-review.md (1875-word peer-review-style critique)
• manuscript/narrative-review/reviews/humanizer-log.md (audit-confirmation log)

Test plan

• Both skills invoked via the Skill tool (loaded full instructions)
• Paper-review applied the Synopsis / Critical / Major / Minor / Editor Note template
• Humanizer scanned 14 files against 29 patterns
• All F1-F5 carry-forward checks verified verbatim in prose
• Pre-commit (markdownlint, typos) passes
• Phase 5 (final assembly) will compile per-section files into a single submission package + apply the 6 paper-review minor concerns