Paper Poster

Paper Poster: From Paper to Conference Poster

Generate a conference poster from: $ARGUMENTS

Context

This skill runs after Workflow 3 (/paper-writing). It takes a compiled paper and generates a print-ready poster for conference poster sessions. The poster extracts key content from the paper — it does not dump the full paper text onto a poster.

Unlike papers (dense prose, 8-15 pages), posters are visual-first: one page, 4 columns, bullet points only, figures dominant. A good poster tells the story in 60 seconds.

Constants

• VENUE = NeurIPS — Target venue, determines color scheme. Supported: NeurIPS, ICML, ICLR, AAAI, ACL, EMNLP, CVPR, ECCV, GENERIC. Override via argument (e.g., /paper-poster "— venue: ICML").
• POSTER_SIZE = A0 — Paper size. Options: A0 (841x1189mm, default), A1 (594x841mm).
• ORIENTATION = landscape — Orientation. Options: landscape (default), portrait.
• COLUMNS = 4 — Number of content columns. Typical: 4 for landscape A0 (IMRAD), 3 for portrait A0 (research consensus), 2 for portrait A1. Portrait A0 should NEVER use 4 columns — text becomes too narrow and unreadable.
• PAPER_DIR = paper/ — Directory containing the compiled paper (main.tex + figures/).
• OUTPUT_DIR = poster/ — Output directory for all poster files.
• REVIEWER_MODEL = gpt-5.5 — Model used via Codex MCP for poster review.
• AUTO_PROCEED = false — At each checkpoint, always wait for explicit user confirmation. Set true only if user explicitly requests fully autonomous mode.
• COMPILER = latexmk — LaTeX build tool.
• ENGINE = pdflatex — LaTeX engine. Use xelatex for CJK text.

💡 Override: /paper-poster "paper/" — venue: CVPR, size: A1, orientation: portrait, columns: 3

Optional: Style reference (— style-ref: <source>, opt-in)

Lets the user steer the poster's structural layout (panel-to-text ratio, figure density, caption length) toward a reference paper. Default OFF — when the user does not pass — style-ref, do nothing differently from before.

Only when — style-ref: <source> appears in $ARGUMENTS, run the helper FIRST:

# Resolve $STYLE_HELPER via the canonical strict-safe chain (see
# shared-references/integration-contract.md §2). Policy A — gate:
# unresolved helper means --style-ref cannot be satisfied, so abort.
cd "$(git rev-parse --show-toplevel 2>/dev/null || pwd)" || exit 1
if [ -z "${ARIS_REPO:-}" ] && [ -f .aris/installed-skills.txt ]; then
    ARIS_REPO=$(awk -F'\t' '$1=="repo_root"{print $2; exit}' .aris/installed-skills.txt 2>/dev/null) || true
fi
STYLE_HELPER=".aris/tools/extract_paper_style.py"
[ -f "$STYLE_HELPER" ] || STYLE_HELPER="tools/extract_paper_style.py"
[ -f "$STYLE_HELPER" ] || { [ -n "${ARIS_REPO:-}" ] && STYLE_HELPER="$ARIS_REPO/tools/extract_paper_style.py"; }
[ -f "$STYLE_HELPER" ] || {
  echo "ERROR: extract_paper_style.py not resolved at .aris/tools/, tools/, or \$ARIS_REPO/tools/." >&2
  echo "       Fix: rerun bash tools/install_aris.sh, export ARIS_REPO, or copy the helper to tools/." >&2
  echo "       --style-ref cannot be satisfied; aborting." >&2
  exit 1
}
STYLE_STATUS=0
CACHE=$(python3 "$STYLE_HELPER" --source "<source>") || STYLE_STATUS=$?
case "$STYLE_STATUS" in
  0) ;;                                       # use $CACHE/style_profile.md as structural guidance
  2) echo "warning: style-ref skipped (missing optional dep)" >&2 ;;
  3) echo "error: --style-ref source failed; aborting poster" >&2 ; exit 1 ;;
  *) echo "error: helper failed unexpectedly; aborting poster" >&2 ; exit 1 ;;
esac

Sources accepted: local TeX dir / file, local PDF, arXiv id, http(s) URL. Overleaf URLs/IDs are rejected — clone via /overleaf-sync setup <id> first and pass the local clone path.

Strict rules (full contract in tools/extract_paper_style.py docstring):

• Use style_profile.md to align figure-to-text ratio, caption length tendency, and paragraph density. Venue color scheme and column count above still take precedence — --style-ref only refines density tendencies.
• Never copy poster content, design elements, slogans, or section names verbatim from anything reachable through the cache.
• Never pass — style-ref (or the cache contents) to the GPT-5.4 reviewer sub-agent — the reviewer must judge the poster's clarity on its own merits.

Venue Color Schemes

Use deep, saturated colors for primary — pastel/light colors wash out on large posters viewed from distance. Each venue uses a 3-color system: primary (dark, for title bar), secondary (medium, for section headers), accent (contrast, for highlights).

⚠️ Color lesson: Never use light/pastel colors (e.g., #8B5CF6) as primary — they look washed out on A0 posters. Always use the darkest shade as primary for the title bar.

State Persistence (Compact Recovery)

Poster generation can be long. Persist state to poster/POSTER_STATE.json after each phase:

{
  "phase": 3,
  "venue": "NeurIPS",
  "poster_size": "A0",
  "orientation": "landscape",
  "columns": 4,
  "figures_selected": ["architecture.pdf", "results.pdf"],
  "codex_thread_id": "019cfcf4-...",
  "status": "in_progress",
  "timestamp": "2026-03-18T15:00:00"
}

On startup: if POSTER_STATE.json exists with "status": "in_progress" and within 24h → resume from saved phase. Otherwise → fresh start.

Critical LaTeX Architecture Decisions

⚠️ MUST use article class, NEVER beamer class. The beamer class consumes too many TeX grouping levels for its overlay/mode system. Combined with tcbposter's enhanced style on 8+ posterboxes, this triggers ! TeX capacity exceeded, sorry [grouping levels=255]. The article class + geometry package for custom page size is the correct approach. This was validated through 5 failed compilation attempts with beamer before switching to article.

⚠️ NEVER use adjustbox package. It may not be installed in minimal TeX distributions. Use plain \includegraphics[width=0.96\linewidth]{file} instead. Do NOT use max height option (requires adjustbox).

Template Foundation

\documentclass{article}
% A0 landscape: paperwidth=1189mm,paperheight=841mm
% A0 portrait:  paperwidth=841mm,paperheight=1189mm
\usepackage[paperwidth=1189mm,paperheight=841mm,margin=0mm]{geometry}
\usepackage{tcolorbox}
\tcbuselibrary{poster,skins,fitting}
\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{enumitem}
\usepackage[table]{xcolor}  % MUST use [table] option for \rowcolor in tables
\usepackage{lmodern}
\usepackage[T1]{fontenc}
\pagestyle{empty}

⚠️ NEVER use \usepackage[most]{tcolorbox} — it pulls in listingsutf8.sty which may not be installed. Always use \tcbuselibrary{poster,skins,fitting} explicitly.

⚠️ Use [table]{xcolor} not plain {xcolor} — needed for \rowcolor in benchmark tables. The colortbl package is loaded automatically by this option.

tcbposter Layout Rules (Critical)

⚠️ The #1 cause of poster failures is content overflow. tcbposter uses a fixed grid — content that exceeds the box is silently clipped with no compilation error. You will NOT see any warning; the poster will simply be cut off.

⚠️ The #2 cause is large whitespace gaps. Using too few rows (e.g., rows=5) creates 168mm per row on A0 landscape. If title text only needs 120mm, the remaining 48mm is wasted whitespace. Solution: use rows=20 for fine-grained control (42mm per row).

Grid System: rows=20 (Critical)

Use rows=20 for A0 landscape. Each row ≈ 42mm, giving precise control over section heights.

Recommended row allocation for 4-column A0 landscape:

Key principle: Always use between=rowN and rowM syntax (not below=name) for precise vertical placement. The below= syntax lets tcolorbox auto-place, which often leaves unwanted gaps.

Row Count Guidance

Portrait A0 Layout (3 columns, rows=20)

⚠️ Portrait A0 posters use 2-3 columns, NEVER 4. Research consensus: "Two columns is typical for a poster with a portrait orientation" (Colin Purrington, NYU poster guides). At 841mm width, 4 columns give only 195mm per column — too narrow for readable text at poster-session distance. **3 columns (260mm each) is the recommended default** for content-rich papers. Use 2 columns for simpler posters or when figures need more horizontal space.

For portrait posters (841x1189mm), use a 3-column, 3-row-band layout:

3-column portrait layout diagram:

┌─────────────────────────────────────┐
│         TITLE BAR (span=3)          │
├─────────────────────────────────────┤
│   Stat 1   │   Stat 2   │  Stat 3  │
├────────────┼────────────┼──────────┤
│ Background │  Method    │ Result 1 │
│ & Motiv.   │ (hero fig) │ (figure) │
├────────────┼────────────┼──────────┤
│ Contribu-  │ Equations  │ Result 2 │
│ tions      │ & Ablation │ (fig+tbl)│
├────────────┼────────────┼──────────┤
│ References │  Setup &   │   Key    │
│ + QR Code  │ Benchmarks │Takeaways │
└────────────┴────────────┴──────────┘

⚠️ All 3 columns in each row band share the same row boundaries. This ensures cross-column alignment. Never mix row6 to row11 in one column with row6 to row10 in another — it creates visual misalignment.

⚠️ Use spacing=0mm for tight layouts. Card separation is handled by card styles (left accent stripe, drop shadow), not grid spacing. Grid spacing > 2mm creates visible gaps between rows.

Modern Card Design System (Left Accent Stripe)

Instead of rounded boxes with colored headers, use a left accent stripe design. This is cleaner, more modern, and avoids the "PowerPoint box" look.

Define 4 card styles using the venue's 3-color system:

% Tinted card backgrounds (NOT pure white — adds warmth)
\definecolor{redbg}{HTML}{FFF5F3}     % warm pink tint for redcard
\definecolor{bluebg}{HTML}{F0F4FF}    % cool blue tint for bluecard
\definecolor{darkbg}{HTML}{FDF6F3}    % warm cream tint for darkcard
\definecolor{redtitlebg}{HTML}{FDEAE8}   % title bar tint
\definecolor{bluetitlebg}{HTML}{E4ECFF}
\definecolor{darktitlebg}{HTML}{F5E8E2}

\tcbset{
  redcard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=redbg,
    borderline west={5pt}{0pt}{secondary},
    left=16pt, right=14pt, top=4pt, bottom=4pt,
    fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{secondary},
    coltitle=secondary, colbacktitle=redtitlebg,
    toptitle=6pt, bottomtitle=6pt,
    titlerule=2pt, titlerule style={secondary!50},
    valign=top, drop shadow={opacity=0.18},
  },
  bluecard/.style={...same pattern with accent color and bluebg...},
  darkcard/.style={...same pattern with primary color and darkbg...},
  highlightcard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=primary!18!white,
    borderline west={6pt}{0pt}{primary},
    fonttitle=...\color{white}, colbacktitle=primary,
    ...
  },
}

Card assignment pattern (creates visual rhythm):

• redcard (secondary stripe): Background, Key Idea, Ablation, References, Setup
• bluecard (accent stripe): Result 1, Result 2, Benchmarks, Analysis
• darkcard (primary stripe): Contributions, Method
• highlightcard (primary fill): Key Takeaways / Conclusion

⚠️ Card backgrounds must NOT be pure white (#FFFFFF). Use subtle tints matching the card's color family. Pure white cards on a tinted poster background look disconnected. The tint should be barely visible but adds cohesion.

Figure + Caption Macro

Define a consistent macro for all figures to ensure uniform spacing:

\newcommand{\posterfig}[3]{%
  \centering\includegraphics[width=#1\linewidth]{#2}\\[3mm]
  {\fontsize{26}{32}\selectfont\color{textgray}\textit{#3}}\vspace{2mm}%
}
% Usage: \posterfig{0.96}{figures/results.png}{Caption text here.}

⚠️ Inconsistent figure-text spacing is the #1 visual flaw in generated posters. The \posterfig macro enforces uniform 3mm gap + 2mm bottom padding across all figures.

Content Colorbox Intensity

Inside cards, use \colorbox{color!N} for highlighted blocks. The intensity N must be 18-25% (not 8-12% which is too faint):

% TOO FAINT (invisible on print):
\colorbox{primary!8}{\parbox{...}{...}}

% CORRECT (visible, distinct):
\colorbox{primary!20}{\parbox{0.94\linewidth}{...}}
\colorbox{accent!20}{\parbox{0.94\linewidth}{...}}
\colorbox{secondary!20}{\parbox{0.94\linewidth}{...}}

Similarly, \rowcolor in tables should use 15% intensity: \rowcolor{primary!15}.

Font Size Rules (A0 at article class — NO scale factor)

⚠️ Critical: When using article class (not beamerposter), there is NO automatic scale factor. All font sizes are literal. A poster viewed from 1.5m needs much larger fonts than you think.

⚠️ Lesson learned from testing: Body text at 20pt on A0 is unreadable from more than 0.5m. 34pt is the minimum for comfortable reading at poster-session distance.

Content Budget

Total target: 300-500 words (excluding figure captions and stat callout numbers).

⚠️ The #1 content mistake is too much text. A poster is NOT a paper summary — it's a visual guide. Each bullet should be a key phrase (5-8 words), not a sentence. If you find yourself writing full sentences, you're putting too much text.

⚠️ Content density calibration: When in doubt, use LESS text. It's much easier to add a few words than to trim dense paragraphs. Target ~70% fill per card (some breathing room), NOT 100%.

Bullet point rules:

• Maximum 8 words per bullet when possible
• Use $\Rightarrow$ and $\to$ for causal arrows instead of words
• Numbers > words: "42% less memory" not "reduces memory usage by 42 percent"
• Colorbox labels: "vs. Depth: 4L CoE ≈ 12L MoE, 42% less memory" (one line)

Recommended 4-Column IMRAD Layout

┌──────────────────────────────────────────────────────────┐
│                    TITLE BAR (span=4)                     │
│     Title (90pt) + Authors (42pt) + Venue + GitHub        │
├──────────────────────────────────────────────────────────┤
│  Stat 1  │  Stat 2  │  Stat 3  │  Stat 4  │  STAT BANNER│
├──────────┼──────────┼──────────┼──────────┤             │
│Background│ Dataset  │Architectu│ Result 2 │             │
│    &     │    &     │   re     │  + Table │             │
│Motivation│Paradigms │ Overview │  + Stats │             │
│    +     │  + Fig   │  + Fig   │  + Fig   │             │
│Contributi│          │──────────│──────────│             │
│   ons    │──────────│ Result 1 │ Ablation │             │
│──────────│Computat. │  + Fig   │──────────│             │
│References│ Models   │  + Table │Conclusion│             │
│ + QR Code│+ Equations│ + Bullets│ + Future │             │
└──────────┴──────────┴──────────┴──────────┘

Workflow

Phase 0: Input Validation & Setup

1. Check prerequisites:

   which pdflatex && which latexmk
   

   If LaTeX is NOT installed, try in order:

   # Option 1: brew cask (requires sudo — may fail in non-interactive shells)
   brew install --cask mactex-no-gui
   
   # Option 2: BasicTeX (smaller, may still need sudo)
   brew install --cask basictex
   
   # Option 3: User-directory install (NO sudo needed — always works)
   curl -L https://mirror.ctan.org/systems/texlive/tlnet/install-tl-unx.tar.gz | tar xz
   cd install-tl-*
   cat > texlive.profile << 'PROF'
   selected_scheme scheme-basic
   TEXDIR ~/texlive/YYYY
   TEXMFLOCAL ~/texlive/texmf-local
   TEXMFSYSCONFIG ~/texlive/YYYY/texmf-config
   TEXMFSYSVAR ~/texlive/YYYY/texmf-var
   TEXMFHOME ~/texmf
   binary_x86_64-darwin 1
   instopt_adjustpath 0
   instopt_adjustrepo 1
   instopt_write18_restricted 1
   tlpdbopt_autobackup 1
   tlpdbopt_install_docfiles 0
   tlpdbopt_install_srcfiles 0
   PROF
   ./install-tl --profile=texlive.profile
   export PATH="$HOME/texlive/YYYY/bin/universal-darwin:$PATH"
   

   After installation, install required packages:

   tlmgr install tcolorbox pgf etoolbox environ trimspaces \
     type1cm pdfcol tikzfill latexmk lm enumitem geometry
   

   ⚠️ Lesson learned: brew install --cask mactex-no-gui often fails in non-interactive shells because the macOS installer requires sudo password. The user-directory TeX Live install (Option 3) always works without sudo.

   ⚠️ Do NOT install or use beamerposter. The article class approach does not need it.

2. Verify paper exists:

   ls $PAPER_DIR/main.tex || ls $PAPER_DIR/main.pdf
   ls $PAPER_DIR/sections/*.tex
   ls $PAPER_DIR/figures/
   

3. Backup existing poster: if poster/ exists, copy to poster-backup-{timestamp}/

4. Create output directory: mkdir -p poster/figures

5. Copy figures to poster directory:

   # IMPORTANT: Use cp, NOT ln -sf (symlinks)
   # pdflatex often fails to resolve symlinks across directories
   cp paper/figures/selected_figure.pdf poster/figures/
   

   ⚠️ Never use symlinks for poster figures. pdflatex cannot reliably follow symlinks across directories. Always cp the actual files.

6. Convert PDF figures to PNG for PPTX embedding:

   python3 -c "import pdf2image" 2>/dev/null || pip install pdf2image
   # For each figure:
   python3 -c "
   from pdf2image import convert_from_path
   for name in ['paradigm', 'architecture', 'results', 'hallucination']:
       imgs = convert_from_path(f'poster/figures/{name}.pdf', dpi=300)
       imgs[0].save(f'poster/figures/{name}.png', 'PNG')
   "
   

   ⚠️ python-pptx CANNOT embed PDF images. You MUST convert to PNG first. This is a hard limitation of the OOXML format. Always generate PNG copies at 300 DPI during setup.

7. Detect CJK: if paper contains Chinese/Japanese/Korean text, set ENGINE to xelatex

8. Check for resume: read poster/POSTER_STATE.json if it exists

Phase 1: Content Extraction

Read each section from paper/sections/*.tex and extract poster-appropriate content:

Extraction rules — a poster shows ~30-40% of the paper's content:

Total target: 400-700 words (excluding figure captions and stat callout numbers).

⚠️ No abstract paragraph on poster. Replace with a stat banner: 3-4 large-number callout boxes showing headline results. This is the single highest-impact change for 60-second comprehension.

Output: poster/POSTER_CONTENT_PLAN.md — structured markdown showing exactly what goes where, with word counts per box.

🚦 Checkpoint:

📋 Poster content plan ready:
- Title: [paper title]
- Venue: [VENUE] ([POSTER_SIZE] [ORIENTATION])
- Layout: [COLUMNS] columns, rows=20
- Figures selected: [N] figures
- Boxes per column: Col1=[N], Col2=[N], Col3=[N], Col4=[N]
- Estimated word count: [N] words

Proceed with this layout? Or adjust content selection?

⛔ STOP HERE and wait for user response.

State: Write POSTER_STATE.json with phase: 1.

Phase 2: Figure Selection & Layout

1. Inventory all figures in paper/figures/:

   ls -la paper/figures/*.{pdf,png,jpg,svg} 2>/dev/null
   

2. Rank by poster importance:

   • Tier 1 (must include): Architecture/method overview diagram, main results plot
   • Tier 2 (include if space): Ablation bar chart, qualitative examples, experimental paradigm
   • Tier 3 (skip): Appendix figures, supplementary plots, tables-as-figures

3. Select top 3-5 figures that fit the 4-column layout

4. Copy figures to poster directory (NOT symlinks) + convert PDF→PNG for PPTX

5. Design column layout — 4-column IMRAD:

   • Col 1: Background & Motivation + Contributions + References & QR
   • Col 2: Dataset & Paradigms (fig) + Computational Models (equations)
   • Col 3: Architecture (fig) + Result 1 (fig + stat table)
   • Col 4: Result 2 (fig + stat table) + Ablation + Conclusion

Phase 3: Generate Poster LaTeX

Create poster/main.tex using article class + geometry + tcbposter.

Template structure (validated through testing):

\documentclass{article}
\usepackage[paperwidth=1189mm,paperheight=841mm,margin=0mm]{geometry}
\usepackage{tcolorbox}
\tcbuselibrary{poster,skins,fitting}
\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{enumitem}
\usepackage[table]{xcolor}
\usepackage{lmodern}
\usepackage[T1]{fontenc}
\pagestyle{empty}

% ── Venue Color Theme ──
\definecolor{primary}{HTML}{VENUE_PRIMARY}     % deep, saturated
\definecolor{secondary}{HTML}{VENUE_SECONDARY} % medium
\definecolor{accent}{HTML}{VENUE_ACCENT}       % contrast
\definecolor{bgposter}{HTML}{VENUE_BG_DEEP}    % poster background (NOT white, use tinted)
\definecolor{redbg}{HTML}{FFF5F3}              % card backgrounds (tinted, NOT white)
\definecolor{bluebg}{HTML}{F0F4FF}
\definecolor{darkbg}{HTML}{FDF6F3}
\definecolor{redtitlebg}{HTML}{FDEAE8}         % card title bar backgrounds
\definecolor{bluetitlebg}{HTML}{E4ECFF}
\definecolor{darktitlebg}{HTML}{F5E8E2}
\definecolor{textdark}{HTML}{111827}
\definecolor{textgray}{HTML}{4B5563}
\definecolor{stathighlight}{HTML}{FEE8E8}

\pagecolor{bgposter}
\color{textdark}

% ── List styling ──
\setlist[itemize]{leftmargin=24pt, itemsep=6pt, parsep=2pt, topsep=2pt,
  label={\color{secondary}$\blacktriangleright$}}
\setlist[enumerate]{leftmargin=24pt, itemsep=6pt, parsep=2pt, topsep=2pt,
  label={\color{primary}\bfseries\arabic*.}}

% ── Figure+caption macro (ensures uniform spacing) ──
\newcommand{\posterfig}[3]{%
  \centering\includegraphics[width=#1\linewidth]{#2}\\[3mm]
  {\fontsize{26}{32}\selectfont\color{textgray}\textit{#3}}\vspace{2mm}%
}

% ── Card styles (left accent stripe design) ──
\tcbset{
  redcard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=redbg,
    borderline west={5pt}{0pt}{secondary},
    left=16pt, right=14pt, top=4pt, bottom=4pt,
    fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{secondary},
    coltitle=secondary, colbacktitle=redtitlebg,
    toptitle=6pt, bottomtitle=6pt,
    titlerule=2pt, titlerule style={secondary!50},
    valign=top, drop shadow={opacity=0.18},
  },
  bluecard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=bluebg,
    borderline west={5pt}{0pt}{accent},
    left=16pt, right=14pt, top=4pt, bottom=4pt,
    fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{accent},
    coltitle=accent, colbacktitle=bluetitlebg,
    toptitle=6pt, bottomtitle=6pt,
    titlerule=2pt, titlerule style={accent!50},
    valign=top, drop shadow={opacity=0.18},
  },
  darkcard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=darkbg,
    borderline west={5pt}{0pt}{primary},
    left=16pt, right=14pt, top=4pt, bottom=4pt,
    fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{primary},
    coltitle=primary, colbacktitle=darktitlebg,
    toptitle=6pt, bottomtitle=6pt,
    titlerule=2pt, titlerule style={primary!50},
    valign=top, drop shadow={opacity=0.18},
  },
  highlightcard/.style={
    enhanced, arc=0pt, boxrule=0pt, colback=primary!18!white,
    borderline west={6pt}{0pt}{primary},
    left=16pt, right=14pt, top=4pt, bottom=4pt,
    fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{white},
    coltitle=white, colbacktitle=primary,
    toptitle=6pt, bottomtitle=6pt,
    valign=top, drop shadow={opacity=0.22},
  },
}

\begin{document}
\begin{tcbposter}[
  coverage={spread},
  poster={columns=4, rows=20, spacing=0mm},  % Use columns=3 for portrait A0
]

% ══ TITLE BAR ══
\posterbox[
  enhanced, colback=primary, colframe=primary, colupper=white,
  arc=0pt, boxrule=0pt,
  left=40pt, right=40pt, top=12pt, bottom=8pt,
  halign=center, valign=center,
  drop shadow={opacity=0.3}
]{name=title, column=1, span=4, between=top and row4}{
  {\fontsize{84}{100}\selectfont\bfseries PAPER TITLE}\\[12pt]
  {\fontsize{36}{44}\selectfont Authors}\\[8pt]
  {\fontsize{30}{38}\selectfont\color{white!70} Affiliations | VENUE YEAR | github.com/...}
}

% ══ STATS BANNER ══
\posterbox[
  enhanced, colback=primary!15!white, boxrule=0pt, arc=0pt,
  left=12pt, right=12pt, top=6pt, bottom=6pt,
  valign=center, borderline south={3pt}{0pt}{primary!35},
]{name=stats, column=1, span=4, between=row4 and row6}{
  \centering
  \begin{minipage}[c]{0.235\linewidth}\centering
    \fcolorbox{primary!40}{stathighlight}{\parbox{0.88\linewidth}{%
      \centering\vspace{6pt}%
      {\fontsize{66}{80}\selectfont\bfseries\color{primary} STAT1}\\[4pt]
      {\fontsize{26}{32}\selectfont\color{textdark} Label 1}\vspace{6pt}%
    }}
  \end{minipage}\hfill
  % ... 3 more stat callouts in same pattern
}

% ══ CONTENT CARDS ══
% Use card styles: \posterbox[redcard, title={...}]{...}{...}
% Body text: \fontsize{34}{44}\selectfont
% Figures: \posterfig{0.96}{figures/name.png}{Caption.}
% Colorboxes: \colorbox{primary!20}{\parbox{0.94\linewidth}{...}}

\end{tcbposter}
\end{document}

Key formatting rules:

• Title: 84pt, bold, primary background, white text
• Author line: 36pt, white text
• Section headers: 40pt via fonttitle — colored text on tinted title background
• Body text: 34pt with 44pt leading — \fontsize{34}{44}\selectfont
• Figures: via \posterfig{0.96}{figures/name.png}{Caption} macro
• Stat callout numbers: 66pt in primary color on stathighlight background
• Tables: \renewcommand{\arraystretch}{1.6} with \rowcolor{primary!15} zebra striping
• Equations in colorboxes: use $\displaystyle ...$ (inline), NOT \[...\] (display math adds margins that cause overfull hbox)

Posterbox pattern (using card styles):

\posterbox[redcard, title={Section Title}
]{name=uniquename, column=N, between=rowA and rowB}{
  \fontsize{34}{44}\selectfont
  \begin{itemize}[itemsep=12pt]
    \item Key point one
    \item Key point two
  \end{itemize}
}

⚠️ Equations in narrow colorboxes: Display math \[...\] adds horizontal margins that cause overfull hbox errors inside \colorbox{\parbox{}}. Always use $\displaystyle ...$ with \centering instead. Reduce equation font to 26-28pt inside colorboxes.

🚦 Checkpoint:

🖼️ Poster LaTeX generated:
- Template: article + tcbposter (rows=20)
- Layout: [COLUMNS] columns, [ORIENTATION] [POSTER_SIZE]
- Colors: [VENUE] theme (primary: [HEX] / secondary: [HEX] / accent: [HEX])
- Figures: [N] embedded
- Font sizes: title=90pt, body=34pt, headers=42pt
- Word count: ~[N] words

Compile now?

⛔ STOP HERE and wait for user response.

State: Write POSTER_STATE.json with phase: 3.

Phase 4: Compile Poster

cd poster && latexmk -pdf -interaction=nonstopmode main.tex

⚠️ If using user-directory TeX Live, prepend PATH: export PATH="$HOME/texlive/YYYY/bin/universal-darwin:$PATH"

Error handling loop (max 3 attempts):

1. Parse error log for the first error
2. Fix the most likely cause:
   • grouping levels=255 → STOP. Switch from beamer to article class. This is not fixable by removing styles.
   • Missing package → tlmgr install <package>
   • File not found: adjustbox.sty → Remove \usepackage{adjustbox} and any max height options
   • File not found → verify poster/figures/ has the file (not a broken symlink)
   • Overfull boxes → reduce text or figure size
3. Recompile

Common missing packages (install proactively if not present):

tlmgr install type1cm pdfcol tikzfill

Verification:

pdfinfo poster/main.pdf
# Check: Pages: 1, Page size: ~3370.39 x 2383.94 pts (A0 landscape)

Visual inspection after compilation:

1. All 4 columns have content visible to the bottom — no silent clipping
2. No large whitespace gaps between title/stats and body content
3. Figures are fully visible, not cut off
4. Text is readable (zoom to 100% = actual A0 size)

Phase 5: Visual Review via Claude + Gemini (Iterative Refinement)

This phase uses Claude visual assessment on rendered poster images to iteratively refine layout, readability, and visual hierarchy — similar to the paper-illustration skill's review loop.

Step 1: Render poster to PNG preview

import fitz
doc = fitz.open('poster/main.pdf')
page = doc[0]
pix = page.get_pixmap(dpi=200)  # 200 DPI for visual review (higher than 150 preview)
pix.save('poster/poster_review.png')
doc.close()

Step 2: Claude visual assessment

Read the rendered poster/poster_review.png and perform a STRICT visual review with the following rubric (score 1-10):

Critical checks (must all pass, any failure = score ≤ 5):

1. Content accuracy — No fabricated data, all numbers match paper
2. Text readability — All text readable at simulated 1.5m distance (no text too small)
3. No clipping — All content visible, no cut-off figures or text
4. Column alignment — Row bands align across all columns

Secondary checks (affect score 6-10): 5. Visual hierarchy — Title → stat banner → body flow is immediately clear 6. Figure prominence — Figures occupy 40-50% of content area 7. Color coherence — Card tints, accent stripes, and venue colors work harmoniously 8. Whitespace balance — No large empty gaps, no overly cramped sections 9. Information density — Can understand the contribution in 60 seconds 10. Overall aesthetics — Would you be proud to present this at a top venue?

Scoring:

• 9-10: Print-ready, no changes needed
• 7-8: Minor tweaks (spacing, font size adjustments)
• 5-6: Needs revision (layout issues, readability problems)
• 1-4: Major issues (clipping, fabricated data, broken layout)

Step 3: Iterative refinement loop

MAX_ITERATIONS = 5
SCORE_THRESHOLD = 9

for iteration in 1..MAX_ITERATIONS:
    1. Render poster to poster/poster_v{iteration}.png (200 DPI)
    2. Claude reads the PNG and performs STRICT visual review
    3. Score the poster (1-10) with detailed feedback
    4. If score >= SCORE_THRESHOLD → PASS, proceed to Phase 6
    5. If score < SCORE_THRESHOLD:
       a. Identify top 3 issues (ranked by visual impact)
       b. Generate targeted LaTeX fixes for each issue
       c. Apply fixes to main.tex
       d. Recompile (Phase 4 error loop)
       e. Continue to next iteration
    6. Save all versions: poster/poster_v{iteration}.png

⚠️ All versions are preserved. Never overwrite previous renders. Save as poster_v1.png, poster_v2.png, etc. This allows comparison and rollback.

⚠️ Targeted fixes only. Each iteration should fix at most 3 specific issues. Do NOT rewrite the entire LaTeX — small, focused edits prevent regression.

Optional: Gemini visual generation (if mcp__illustrator__run is available):

For poster elements that need custom illustrations (e.g., hero architecture diagram, method workflow), use the Gemini illustration pipeline:

1. Write a detailed specification for the illustration
2. Call mcp__illustrator__run with the specification
3. Claude reviews the generated image for accuracy
4. Iterate until score ≥ 9 or max 3 attempts
5. Save final illustration to poster/figures/ and embed in LaTeX

Step 4: Save visual review log

Append all iteration scores and feedback to poster/POSTER_VISUAL_REVIEW.md:

# Visual Review Log

## Iteration 1 — Score: 7/10
- Issue 1: Title font too small (72pt → should be 84pt+)
- Issue 2: Results figure clipped at bottom
- Issue 3: Stat banner numbers not prominent enough
- Fixes applied: [list of changes]

## Iteration 2 — Score: 9/10
- All critical checks pass
- Minor: References column slightly shorter than others
- Decision: PASS — print-ready

Phase 6: Codex MCP Review

Send the poster content plan + key LaTeX sections to GPT-5.4 xhigh for review.

mcp__codex__codex:
  config: {"model_reasoning_effort": "xhigh"}
  prompt: |
    Review this academic conference poster for [VENUE].

    Evaluate using these criteria (score 1-5 each):

    1. **Information hierarchy** — Can someone understand the contribution in 60 seconds?
    2. **Text density** — Is it concise enough? (Target: 400-700 words total, bullet points only, NO abstract paragraph)
    3. **Figure prominence** — Are key results visually dominant? (Target: figures occupy 40-50% of area)
    4. **Column balance** — Are columns roughly equal height?
    5. **Readability** — Font sizes appropriate for 1.5m distance? (Title ≥90pt, body ≥34pt)
    6. **Narrative flow** — Does the poster tell a left-to-right story?
    7. **Whitespace** — Is content filling the space well? No large empty gaps?

    Poster content:
    [PASTE POSTER_CONTENT_PLAN.md]

    LaTeX source:
    [PASTE key sections of main.tex]

    Provide:
    - Score for each criterion
    - Top 3 actionable fixes (ranked by impact)
    - Overall: Ready to print? (Yes / Needs revision / Major issues)

Apply CRITICAL and MAJOR fixes to poster/main.tex. Recompile if changes were made.

Save review to poster/POSTER_REVIEW.md.

⚠️ Important: After applying review fixes, proceed to Phase 6 only when the poster is finalized. PPTX and SVG must be generated from the final LaTeX/PDF — never from an intermediate version.

Phase 7: Editable Format Export

⚠️ Generate PPTX and SVG only AFTER all revisions are complete. This phase runs last (after review fixes) to ensure all formats contain identical content.

6.1 PowerPoint (.pptx)

Generate a native PPTX using python-pptx (not pandoc — pandoc conversion is lossy):

python3 -c "import pptx" 2>/dev/null || pip install python-pptx

Write a Python script poster/generate_pptx.py that:

1. Creates a single-slide PPTX with poster dimensions (A0 landscape: 1189mm x 841mm)
2. Replicates the 4-column layout using positioned text boxes
3. Embeds PNG figures (from poster/figures/*.png — NOT PDFs, python-pptx cannot embed PDFs)
4. Applies venue color scheme (primary/secondary/accent) to title bar and section headers
5. Keeps all text editable (not images of text)
6. Uses large font sizes matching the PDF (title 86pt, body 34pt, headers 42pt, stats 68pt)
7. Reads content from the FINAL main.tex — do NOT hardcode content separately

⚠️ PPTX font sizes must also be large. A common mistake is using small fonts (17-24pt) in the PPTX while the PDF has 34pt+. The PPTX is A0-sized so needs identical large fonts.

PPTX helper pattern:

def add_image(left, top, w, filename):
    """Add PNG image, auto-calculate height from aspect ratio."""
    path = os.path.join(FIG_DIR, filename)
    if not os.path.exists(path):
        txt(left, top, w, 60, f"[Image: {filename}]", ...)
        return top + 60
    pic = slide.shapes.add_picture(path, Mm(left), Mm(top), Mm(w))
    h_mm = pic.height / Mm(1)
    return top + h_mm

cd poster && python3 generate_pptx.py
# Output: poster/poster.pptx

6.2 SVG (for Adobe Illustrator)

Convert the compiled PDF to editable SVG. Preferred method: PyMuPDF (always available via pip, no brew/system install needed):

# Preferred: PyMuPDF (pip install pymupdf) — always works, no system deps
python3 -c "import fitz" 2>/dev/null || pip install pymupdf
python3 -c "
import fitz
doc = fitz.open('poster/main.pdf')
page = doc[0]
svg = page.get_svg_image()
with open('poster/poster.svg', 'w') as f:
    f.write(svg)
doc.close()
print('SVG saved')
"

# Fallback 1: pdf2svg (if installed)
which pdf2svg && pdf2svg poster/main.pdf poster/poster.svg

# Fallback 2: inkscape
which inkscape && inkscape poster/main.pdf --export-type=svg --export-filename=poster/poster.svg

⚠️ SVG inherits all layout issues from PDF. If the PDF has whitespace gaps or clipped figures, the SVG will too. Always fix the PDF first.

💡 **PyMuPDF bonus

…(truncated)