v2 phase 5: final assembly + numbered references

[neuromechanist]

Summary

v2 final. Closes #13. Closes the v2 narrative-review redo.

What v2 Phase 5 did

1. Applied the 6 minor concerns (m1-m6) from the Phase 4 paper-review:

   • m1: Highlight Phase 1: Preprocess R3 (BIDS import, HPF, cleanline, channel rejection) #2 now names the four perspectives ("Psychophysics, action, language, emotion diverge on per-shot EEG predictions"; 76 chars).
   • m2: Abstract sampling-rate clause expanded for context ("the local 100 Hz cohort caps beta-band and gamma-band claims pending a 500 Hz validation").
   • m3: Section 4 Hickok critique substance named ("mu suppression also reflects general attention to moving stimuli rather than a one-to-one mirror-system signature").
   • m5: Section 1 meta-narration closing sentence dropped.
   • m6: Castelli2000MovementAM deduplicated against castelli2000heider (same DOI, two corpus-strand keys); refs.bib now 93 entries.
   • m4: Section 6 social-cognition density restructure deferred to human-author polish; restructuring requires content judgment beyond the skill pipeline.

2. Concatenated all per-section files into a single submission-ready manuscript.md (frontmatter + Highlights + Abstract + sections 1-7 + Box 1 + Trends Box + Outstanding Questions + Glossary + Figure legends). Body sections demoted from H1 to H2 per markdownlint MD025.

3. Converted 100 in-body [CiteKey] and [Key1; Key2] placeholders to Cell Press numbered form [N] and [N,M]. 82 unique cited keys ordered by first appearance.

4. Built references.md with 82 numbered entries in Cell Press house format from the 93-entry refs.bib.

Final manuscript metrics

Element	Value	TiCS target
Main text (sections 1-7)	3393 words	~3500-4000
Abstract	116 words	80-120
Highlights (5 bullets)	max 80 chars each	<=85
Trends Box	191 words	~200-250
Outstanding Questions Box	7 questions	5-7
Glossary	15 terms	10-15
Box 1	161 words	~180
Figures	4 (300 dpi, 174mm Cell two-column, 6pt min font)	3-5
References	82 numbered	80-120

v1 -> v2 changes summary

v1 was produced by bypassing /manuscript:manuscript-formatting and /manuscript:manuscript-writing (the user's primary objection). v2 invoked every named skill via the Skill tool and produced both:

1. The submission-ready manuscript (functionally equivalent in argument to v1, but produced through the named skill pipeline with each skill's outputs and gaps documented honestly).

2. The skill-pipeline demo observable. Each phase PR body documents what the skill provided, what it did not, what calibration was applied. Three notable findings worth recording from the v2 demo:

• manuscript:manuscript-formatting does not encode TiCS Forum Review conventions (Highlights, Trends Box, Outstanding Questions, Glossary). These are TiCS-specific decisions applied on top of the skill.
• manuscript:manuscript-writing IMRAD templates apply to Sections 1 (Introduction template) and 7 (Discussion template); perspective-body sections 2-6 have no IMRAD analogue and use Writing Principles only.
• figures:scientific-figure provides the pipeline (Plan -> Build -> Compose -> Validate -> Export) but the documented "user units = mm; font-size=N is N pt" convention is ambiguous; renderers (rsvg-convert) treat unit-less font-size as user units, not pt. Workaround: pt-viewBox with mm width/height.
• Additionally, rsvg-convert blocks external image refs (CVE-2023-38633 mitigation); base64 data URIs are the cross-renderer workaround for icon embedding.

The humanizer pass was a no-op: the upstream style-guide layer (CLAUDE.md) plus the v1 self-review carry-forwards already prevented all 29 AI-writing patterns from entering the prose.

Test plan

• All 6 paper-review minor concerns (m1-m6) addressed or explicitly deferred
• manuscript.md assembled with all per-section files concatenated
• [CiteKey] -> [N] conversion complete (100 in-body cites; 82 unique keys numbered by first appearance)
• references.md built and matches in-body numbering
• refs.bib deduplicated (93 entries; Castelli dupe removed)
• Word counts inside TiCS Forum Review budget
• All Highlights bullets <= 80 chars
• Pre-commit (markdownlint, typos) passes
• CHECKLIST.md tracks every TiCS submission requirement
• Post-PR: human-author final polish (m4 social-cognition restructure; ORCID; cover letter; .docx export)

Manifest

manuscript/narrative-review/
  manuscript.md            <- 5024-word single submission file
  references.md            <- 82 Cell Press-format numbered references
  refs.bib                 <- 93 BibTeX entries
  highlights.md, abstract.md, sections/*.md, boxes/*.md, glossary.md, figures.md
  STRUCTURE.md, frontmatter.yaml, CHECKLIST.md
  figures/                 <- 4 final figures + panels + icons + qa reports
  reviews/                 <- paper-review + humanizer logs