medsci-skills 4.1.0

package/skills/make-figures/references/pipeline_concepts_medical_ai.md

@@ -0,0 +1,240 @@
+# Pipeline & Architecture Diagrams — Medical AI / Engineering
+
+> **Triggered from**: SKILL.md Step 1 (Specify) when the figure is a
+> pipeline / architecture / workflow for a medical AI or imaging
+> engineering paper. Read alongside `design_principles.md` (key-message)
+> and `flow_diagram_lessons.md` (production lessons).
+
+This file covers the four most common diagram types in medical-imaging AI
+and engineering manuscripts that the standard reporting-guideline flows
+(PRISMA / CONSORT / STARD / STROBE) do not address: DICOM workflow,
+annotation pipeline, federated-learning topology, and model architecture.
+For each, this file gives the canonical layout, common pitfalls, and a
+preferred tool.
+
+---
+
+## 1. DICOM workflow diagram
+
+**Use when**: the manuscript describes how DICOM images flow from scanner
+→ PACS → research environment → preprocessing → model.
+
+**Canonical layout** (left-to-right):
+
+```
+[Modality / scanner]  →  [PACS]  →  [Research VNA / mirror]
+       ↓                                       ↓
+  acquisition                       de-identification
+  parameters                        + DICOM tag scrub
+                                              ↓
+                              [Preprocessed cohort store]
+                                              ↓
+                              [Train / tune / test split]
+                                              ↓
+                                     [Model input]
+```
+
+**Required annotations**:
+
+- Modality (CT, MR, US, X-ray, OCT) and key acquisition parameters
+  (kVp / mAs for CT, sequence and TE/TR for MR).
+- De-identification step explicitly named — DICOM tag list (e.g., remove
+  PatientName, PatientID, AccessionNumber, InstitutionName) or reference
+  to a published profile (DICOM PS3.15 Annex E "Basic Profile").
+- Whether pixel-data redaction (burned-in PHI) was performed.
+- Cohort filtering criteria with counts at each step (this is also a
+  STARD / TRIPOD requirement).
+
+**Common pitfalls**:
+
+- Drawing "PACS → model" with no de-identification box (privacy reviewers
+  reject).
+- Counts only at the final analytic cohort, not at each filter step
+  (reproducibility reviewers reject).
+- Using "DICOM" as a single block when the manuscript actually transforms
+  to NIfTI / NRRD partway through — show the conversion explicitly.
+
+**Preferred tool**: D2 with `--layout elk` for left-to-right; or Graphviz
+`rankdir=LR`. Avoid matplotlib `FancyBboxPatch` (manual coordinates break
+when text changes).
+
+---
+
+## 2. Annotation / labeling pipeline
+
+**Use when**: the manuscript reports a labeled dataset built by human
+annotators (segmentation, classification, bounding boxes).
+
+**Canonical layout**:
+
+```
+[Raw image cohort]
+       ↓
+[Annotation tool / platform]  ←  [Annotator pool: N readers, expertise]
+       ↓
+[Round 1 labels]  →  [QC / consensus]  →  [Adjudication: senior reader]
+                                                       ↓
+                                        [Final reference standard]
+                                                       ↓
+                                        [Inter-rater agreement: κ / Dice]
+```
+
+**Required annotations**:
+
+- Number and expertise level of annotators (e.g., "3 board-certified
+  radiologists, 5–18 years of post-board experience").
+- Annotation tool (e.g., 3D Slicer, ITK-SNAP, MD.ai, RIL-Contour, custom).
+- Consensus rule (majority vote / unanimous / senior arbitration).
+- Inter-rater agreement metric and value reported separately.
+- Whether annotators were blinded to model output / clinical history.
+
+**Common pitfalls**:
+
+- Drawing "labeled by experts" with no count / qualification (reviewers
+  ask for quantification).
+- Omitting the QC arrow (looks like single-pass labeling, low quality).
+- Conflating "reference standard" with "ground truth" — reference
+  standard is the imperfect human label; ground truth is what we wish
+  we had.
+
+**Preferred tool**: D2; nodes shaped as rectangles for steps, dashed
+border for human-in-the-loop steps to distinguish from automated.
+
+---
+
+## 3. Federated-learning topology
+
+**Use when**: the manuscript trains a model across multiple sites without
+centralizing raw data.
+
+**Canonical layout** (radial, central server in middle):
+
+```
+              [Site A: cohort A_n, scanner type]
+                              ↑↓
+                       (model weights only,
+                        not images)
+                              ↑↓
+              [Central aggregator / parameter server]
+                              ↑↓
+   [Site B: cohort B_n] ←→ [Site C: cohort C_n] ←→ [Site D: cohort D_n]
+```
+
+**Required annotations**:
+
+- Per-site cohort size, scanner / vendor, demographic summary.
+- What is exchanged (gradients / weights / encrypted updates) — and
+  what is **not** (raw images, intermediate features).
+- Aggregation algorithm (FedAvg / FedProx / FedBN / etc.).
+- Number of communication rounds and local epochs per round.
+- Privacy / security mechanism (differential privacy noise, secure
+  aggregation) if claimed.
+
+**Common pitfalls**:
+
+- Bidirectional arrows without labeling what flows each way.
+- No site-level demographic table — federated claims fall flat without
+  evidence of distribution shift across sites.
+- Labeling raw images crossing site boundaries — if that happens it is
+  not federated learning, it is centralized.
+
+**Preferred tool**: D2 with explicit `near: center` for the aggregator
+node, then sites at compass points. Or matplotlib polar layout for
+manuscripts that want radial symmetry.
+
+---
+
+## 4. Model architecture diagram
+
+**Use when**: the manuscript proposes or modifies a neural-network
+architecture and the message depends on the structural change.
+
+**Canonical layout** (left-to-right block diagram):
+
+```
+[Input: 3D volume H×W×D, 1 channel]
+                ↓
+[Backbone: ResNet-50 / ViT-B / nnU-Net]   (cite + checkpoint)
+                ↓
+[Neck / feature pyramid / skip connections]
+                ↓
+[Head: classification / segmentation / detection]
+                ↓
+[Output: class probability / mask / bounding boxes]
+```
+
+**Required annotations**:
+
+- Input shape (H × W × D × C) and modality.
+- Backbone family + variant + initialization (random / ImageNet /
+  domain-pretrained — cite the pretraining work).
+- Output shape and post-processing (softmax / argmax / non-max
+  suppression).
+- Loss function (cross-entropy / Dice / focal / compound).
+- Trainable parameter count.
+
+**Style conventions** (medical-AI / engineering):
+
+- Convolutional / linear blocks: rectangles with channel count below.
+- Feature maps: trapezoids that shrink/grow to convey resolution change
+  (optional but well-recognized).
+- Skip connections: dashed arrows.
+- Attention or transformer blocks: rectangles with internal split into
+  Q / K / V annotation only when the message depends on attention.
+
+**Common pitfalls**:
+
+- "We used ResNet-50" with no diagram — fine for a methods paper, not
+  for an architecture-contribution paper.
+- 3-D rendered "Convolution" pictograms with no dimension info — looks
+  decorative, says nothing.
+- Drawing every layer when only the modified blocks matter — the eye
+  cannot tell what changed. Show the modification, abstract the rest as
+  "ResNet-50 backbone (frozen)".
+
+**Preferred tools**:
+
+- **Diagrams.net (drawio)** — fastest for one-off architecture figures.
+- **NN-SVG** (https://alexlenail.me/NN-SVG/) — generates clean SVG for
+  fully-connected, LeNet, AlexNet variants.
+- **PlotNeuralNet** (LaTeX/TikZ) — used in the academic CV/ML community
+  for publication-quality 3-D block diagrams.
+- D2 with `shape: hexagon` / `shape: cylinder` if the architecture is
+  unusual enough that custom blocks are needed.
+
+For medical-AI papers targeting *Radiology AI*, *npj Digital Medicine*,
+or *Nature Medicine*, prefer NN-SVG or PlotNeuralNet output over
+hand-drawn blocks — reviewers in this venue are sensitive to alignment
+quality.
+
+---
+
+## When to use which (quick selector)
+
+| The figure is showing… | Use this section |
+|---|---|
+| How images move from scanner to model input | DICOM workflow |
+| How a labeled dataset was built | Annotation pipeline |
+| Multi-site training without raw-data sharing | Federated topology |
+| The structure of the proposed neural network | Model architecture |
+| Cohort filtering with counts at each stage | `flow_diagram_lessons.md` (PRISMA / STARD style) |
+| Training / tuning / test data splits | dataset-flow (see `reporting_guideline_figure_map.md` §AI-specific) |
+
+If the figure tries to do two of the above at once, split into two panels
+or two figures — combining DICOM workflow + architecture in one panel
+violates the cognitive-load budget (`design_principles.md` §4).
+
+---
+
+## Cross-references
+
+- `design_principles.md` — communication-first checks
+- `flow_diagram_lessons.md` — production lessons (template fidelity, PDF
+  export, version freeze)
+- `reporting_guideline_figure_map.md` — which figures CLAIM / TRIPOD+AI
+  / STARD-AI mandate
+- `critic_rubrics/data_plot.md` — calibration / fairness / colorblind
+  checks
+- `exemplar_diagrams/pipeline/pipeline_01.svg` … `pipeline_10.svg` —
+  worked exemplars (multimodal MLLM, contrastive learning, VQA, report
+  generation)

package/skills/make-figures/references/reporting_guideline_figure_map.md

@@ -0,0 +1,104 @@
+# Reporting Guideline → Figure Requirements Map
+
+> **Bridge**: this file connects `/make-figures` to `/check-reporting`
+> (32 reporting guidelines). Each row tells you which figures the guideline
+> **mandates** and how this skill currently supports them. Use during
+> Step 1 (Specify) once the study type is known.
+
+When `/check-reporting` flags missing figures for a target guideline, look
+up the row here and either render the supported template or fall back to
+the documented manual approach.
+
+---
+
+## Mandatory-figure map
+
+Legend for **Status** column:
+- ✅ Official template + R/Python generator shipped with this skill
+- ⚠️ Generic flow generator (`generate_flow_diagram.R`) covers the layout
+  but no official-template fidelity check
+- ❌ No template; user produces with D2 / Graphviz / Inkscape from the
+  guideline document, then runs critic_rubric manually
+
+| Guideline (year) | Study type | Mandatory figure(s) | Status | Source / where the official asset lives |
+|---|---|---|---|---|
+| **PRISMA 2020** | Systematic review | Flow diagram (4-phase: identification → screening → eligibility → included) | ✅ | `templates/official/prisma2020/`; R `PRISMA2020` package |
+| **PRISMA-DTA** | DTA systematic review | Modified PRISMA flow + DTA-specific exclusion reasons | ⚠️ | Salameh et al., *BMJ* 2020 (PMID 32312813); use generic flow + extra columns |
+| **PRISMA-NMA** | Network MA | PRISMA flow + network plot | ❌ | Hutton et al., *Ann Intern Med* 2015; network plot via R `netmeta::netgraph()` |
+| **PRISMA-P** | Protocol of SR | (none mandated; PRISMA-S search strategy figure recommended) | ❌ | Rethlefsen et al., *Syst Rev* 2021 |
+| **CONSORT 2025** | RCT | Participant-flow diagram (enrollment → allocation → follow-up → analysis) | ✅ | `templates/official/consort2010/` (now CONSORT 2025); R generator |
+| **CONSORT-AI 2020** | AI intervention RCT | CONSORT flow extended with AI training/validation/deployment dataset boxes | ❌ | Liu et al., *Nat Med* 2020 (PMID 32908283), Fig. 1; D2 / Graphviz custom |
+| **STARD 2015** | Diagnostic accuracy | Flow diagram (eligible → index test → reference standard → 2×2) + ROC | ✅ | `templates/official/stard2015/`; R generator |
+| **STARD-AI 2025** | AI diagnostic accuracy | STARD flow + dataset-flow (training / tuning / test) + subgroup-overlaid ROC/PR | ❌ | Sounderajah et al., *Nat Med* 2025 (PMID 40954311); produce manually, see `flow_diagram_lessons.md` |
+| **STROBE** | Observational cohort/case-control | (Flow diagram **recommended** but not strictly mandated) | ⚠️ | von Elm et al., *Ann Intern Med* 2007; use generic flow generator |
+| **TRIPOD 2015** | Prediction model | Calibration plot (mandatory) + discrimination (ROC, c-stat with CI) | ✅ (data plots) | Collins et al., *Ann Intern Med* 2015 |
+| **TRIPOD+AI 2024** | AI prediction model | TRIPOD figures + **fairness/subgroup panels** + **dataset-flow** + **decision-curve analysis** | ❌ (subgroup, DCA) | Collins et al., *BMJ* 2024 (PMID 38636956); produce manually |
+| **CLAIM 2024** | Medical imaging AI | Architecture diagram (model card style) + dataset-flow + calibration + per-subgroup performance + saliency/attention | ❌ | Tejani et al., *Radiology: AI* 2024 (PMID 38809149); 44 items total |
+| **DECIDE-AI 2022** | AI clinical eval (Stage 1–2) | Human-AI interaction diagram + safety-signal plot + override-rate over time | ❌ (uncertain — verify in full text) | Vasey et al., *Nat Med* 2022 (PMID 35585198) |
+| **CHEERS 2022** | Economic evaluation | Cost-effectiveness plane + cost-effectiveness acceptability curve | ❌ | Husereau et al., *BMJ* 2022 |
+| **SPIRIT 2025** | Trial protocol | Schedule-of-enrollment timeline (Figure 1 in published trials) | ✅ | `templates/official/spirit2013/` (updated to 2025); see also Robinson timeline figures |
+| **CARE 2016** | Case report | Timeline of patient course (recommended) | ⚠️ | Riley et al., *J Clin Epidemiol* 2017; use generic flow generator |
+| **SQUIRE 2.0** | Quality improvement | Run chart / SPC chart | ❌ | Ogrinc et al., *BMJ Qual Saf* 2016 |
+
+---
+
+## When the status is ⚠️ (generic flow only)
+
+The R `generate_flow_diagram.R` script handles the layout, but the layout
+will not match the canonical guideline document exactly. For circulation
+to senior co-authors, that is usually acceptable; for reviewer-facing
+journals where fidelity is checked (esp. in *BMJ*, *Lancet*, *Ann Intern
+Med*, *JAMA*), prefer the official template route documented in
+`flow_diagram_lessons.md` Lesson 1.
+
+## When the status is ❌
+
+For AI-extension guidelines (CONSORT-AI, STARD-AI, TRIPOD+AI, CLAIM 2024,
+DECIDE-AI), there is no shipped official template **as of 2026-05**.
+Production path:
+
+1. Read the original article to extract the canonical figure layout.
+2. Sketch in D2 (`flow.d2`) or Graphviz DOT, layout via ELK.
+3. Render to PDF with the platform-appropriate exporter
+   (`flow_diagram_lessons.md` Lesson 2).
+4. Apply the cognitive-load + key-message-visibility checks from
+   `critic_rubrics/flow_diagram.md` Section G.
+5. Cross-check against the corresponding `/check-reporting` checklist
+   (item-by-item).
+
+When 2 or more projects need the same custom guideline figure, propose
+adding a new template under `templates/official/{guideline}/` so the
+deterministic generator can support it.
+
+---
+
+## AI-specific figures most often missing (priority for new templates)
+
+These are the figures that medical AI manuscripts most often omit, ranked
+by how frequently a reviewer-facing checklist (CLAIM 2024 / TRIPOD+AI)
+flags them:
+
+1. **Dataset-flow diagram** — patient/image counts at each split
+   (training / tuning / internal test / external test). Required by
+   STARD-AI, CLAIM 2024, TRIPOD+AI.
+2. **Calibration plot** — supported by `critic_rubrics/data_plot.md` §C.
+3. **Fairness / subgroup panel** — performance by sex / race / device /
+   site. Required by TRIPOD+AI, CLAIM 2024.
+4. **Decision-curve analysis** — net benefit vs threshold. Required by
+   TRIPOD+AI; recommended by CLAIM 2024.
+5. **Architecture diagram** — input modality → preprocessing → backbone →
+   head → output. See `pipeline_concepts_medical_ai.md`.
+6. **Saliency / attention overlay** — qualitative panels showing model
+   attention. Required by CLAIM 2024 §4.
+
+---
+
+## Cross-references
+
+- `/check-reporting` skill — supports all 32 guidelines, item-level audit
+- `flow_diagram_lessons.md` — production lessons that apply across all flows
+- `pipeline_concepts_medical_ai.md` — DICOM / annotation / federated /
+  architecture diagram conventions
+- `design_principles.md` — communication-first design (Nat Hum Behav 2026)
+- `critic_rubrics/flow_diagram.md` — extended checklist (Sections A–G)
+- `critic_rubrics/data_plot.md` — calibration / fairness / colorblind checks

package/skills/make-figures/references/visual_abstract_templates/template_guide.md

@@ -0,0 +1,114 @@
+# Visual Abstract Template Guide
+
+How `generate_visual_abstract.py` maps content to PPTX template shapes.
+
+## Matching Rules
+
+The script iterates through all shapes on slide index 0 and matches by **text content**
+(case-insensitive substring match). Each shape is filled exactly once.
+
+| Content Field | Match Pattern (in shape.text) | CLI Flag |
+|---|---|---|
+| Article title | `ArticleTitle` | `--title` |
+| Hypothesis / research question | `Hypothesis` or `Question` | `--hypothesis` |
+| Methodology | `Methodology` or `flowchart` or `bullet` | `--methods` |
+| Visual element (image) | `Visual element` or `Image` or `Illustration` | `--visual` |
+| Main finding | `Main finding` or `relevance` | `--finding` |
+| Citation | `Eur Radiol` or `DOI` or `Author` | `--citation` |
+| Patient cohort badge | `Patient` or `cohort` | `--badges` (1st) |
+| Modality badge | `Modality` or `organ` | `--badges` (2nd) |
+| Center type badge | `Single` or `Multi-center` or `center` | `--badges` (3rd) |
+
+## European Radiology Template
+
+**Source:** EURA-GA-Jan2025.pptx (January 2025)
+**File:** `european_radiology.pptx`
+**Use slide:** Index 0 (slide 1 = blank template; slide 2 = filled example)
+
+### Shape Map (Slide 1)
+
+| Shape # | Name | Content Field | Position |
+|---------|------|---------------|----------|
+| 0 | Title 1 | Article title | Top, full width |
+| 5 | Abgerundetes Rechteck 7 | Hypothesis/Question | Below title, full width |
+| 3 | Abgerundetes Rechteck 7 | Methodology | Left panel, below hypothesis |
+| 4 | Abgerundetes Rechteck 7 | Visual element (image) | Right panel, large area |
+| 7 | Abgerundetes Rechteck 7 | Patient cohort badge | Left, below methodology |
+| 8 | Abgerundetes Rechteck 7 | Modality / organ badge | Center-left, below methodology |
+| 9 | Abgerundetes Rechteck 7 | Single / Multi-center badge | Center, below methodology |
+| 6 | Abgerundetes Rechteck 7 | Main finding | Bottom area, full width |
+| 2 | Abgerundetes Rechteck 2 | Citation line | Bottom bar |
+| 1 | Content Placeholder 4 | (Logo area — leave empty or add journal logo) |
+
+### Notes
+
+- All shape names in the EUR template are generic German ("Abgerundetes Rechteck" = rounded
+  rectangle). The script identifies shapes by their **text content**, not by name.
+- The Visual element shape (Shape 4) should have its text cleared and an image inserted.
+  The script places the image within the shape's bounding box, maintaining aspect ratio.
+- Badge shapes (7, 8, 9) have small icon images in the EUR example slides — the script
+  replaces text only. Icons can be added manually in PowerPoint after generation.
+
+## MedSci Default Template
+
+**File:** `medsci_default.pptx`
+**Use slide:** Index 0
+
+A journal-neutral template following the same structure as European Radiology but without
+journal-specific branding. Uses neutral colors (dark gray accent, white background).
+
+### Shape Map
+
+Same field mapping as EUR, with these differences:
+- No journal logo placeholder
+- Neutral accent color (#404040 dark gray)
+- Slightly wider visual element area
+
+## Adding a New Journal Template
+
+1. Obtain the journal's official visual abstract template (PPTX or PPT format).
+2. Copy to `visual_abstract_templates/{journal_name}.pptx`.
+3. Ensure the template slide has placeholder text matching the patterns in the
+   Matching Rules table above. If not, either:
+   - Manually edit the template to add matching placeholder text, OR
+   - Add custom matching rules to `generate_visual_abstract.py`
+4. Add a section to this guide documenting the shape map.
+5. Update the journal profile in `write-paper/references/journal_profiles/` with
+   the visual abstract requirement status and template name.
+
+## JACC Central Illustration Template
+
+`jacc_central_illustration.pptx` — for JACC family journals (`--type central-illustration`).
+
+Built reproducibly via `scripts/build_jacc_template.py` to match the official JACC submission PPTX layout (verified against doi:10.1016/j.jacc.2019.10.035 Figures 1–4).
+
+### Layout (10 × 7.5 in slide)
+
+| Slot | Type | Position (left, top) | Size (W × H) | Placeholder text | Filled by `--type central-illustration` |
+|---|---|---|---|---|---|
+| 1 | TEXT_BOX | (0.4, 5.3) | 9.4 × 0.5 in | `ARTICLECITATION` | `--citation` |
+| 2 | RECTANGLE → image | (3.0, 0.8) | 4.0 × 4.2 in | `VISUALELEMENT` | `--visual` (PNG/TIFF, ≥600 DPI) |
+| 3 | TEXT_BOX | (0.4, 7.0) | 4.1 × 0.5 in | `FOOTERNOTE` | optional, usually empty |
+| 4 | RECTANGLE | (7.3, 6.7) | 2.7 × 0.8 in | `JACCLOGO` | leave for editorial |
+
+### Why a separate template
+
+The JACC editorial team applies the red outer border and the blue "CENTRAL ILLUSTRATION:" header bar after acceptance. Author submissions must contain only:
+- The content figure (slot 2)
+- The citation line (slot 1)
+
+Pre-rendering JACC house elements (red border, blue header) is incorrect — JACC will replace them anyway, and pre-rendered versions clash with the family branding.
+
+### Validation rules
+
+The CI mode in `generate_visual_abstract.py` validates structural simplicity per Fuster-Mann 2019:
+- ≤ 3 visual zones in the content figure (`--ci-zones`)
+- ≤ 30 total label words (`--ci-label-words`)
+- ≤ 4 numerical highlights (`--ci-numerical-points`)
+- No methodology terms in `--ci-raw-text` (CI ≠ Visual Abstract)
+
+Override with `--ci-allow {zones|words|numerical|methods}` only when justified.
+
+### Future cardiology templates
+
+For Circulation, EHJ, JAHA, or other journals with similar CI requirements, copy this template and rename. The Fuster-Mann 5 rules and CI mode validation apply unchanged; only the slide dimensions and citation footer pattern may need adjustment.

package/skills/make-figures/scripts/build_jacc_template.py

@@ -0,0 +1,77 @@
+"""Build the JACC Central Illustration PPTX template.
+
+Layout matches official JACC submission templates (verified from
+doi:10.1016/j.jacc.2019.10.035 Figures 1–4 PPTX files).
+
+Run once. Output:
+  references/visual_abstract_templates/jacc_central_illustration.pptx
+"""
+from pathlib import Path
+from pptx import Presentation
+from pptx.util import Inches, Pt, Emu
+from pptx.dml.color import RGBColor
+from pptx.enum.shapes import MSO_SHAPE
+
+ROOT = Path(__file__).resolve().parents[1]
+OUT = ROOT / "references" / "visual_abstract_templates" / "jacc_central_illustration.pptx"
+
+# JACC submission slide is 10 x 7.5 in (4:3 standard PPT)
+prs = Presentation()
+prs.slide_width = Inches(10)
+prs.slide_height = Inches(7.5)
+
+# Blank layout
+blank = prs.slide_layouts[6]
+slide = prs.slides.add_slide(blank)
+
+# ---- Slot 1: citation text box ----
+cite = slide.shapes.add_textbox(Inches(0.4), Inches(5.3), Inches(9.4), Inches(0.5))
+tf = cite.text_frame
+tf.text = "ARTICLECITATION — replace with: FirstAuthor et al. JACC: Asia YYYY; vol(issue):pages."
+p = tf.paragraphs[0]
+p.font.size = Pt(11)
+p.font.bold = True
+p.font.color.rgb = RGBColor(0x1F, 0x4E, 0x79)
+
+# ---- Slot 2: content picture placeholder (rectangle as marker) ----
+ph = slide.shapes.add_shape(
+    MSO_SHAPE.RECTANGLE, Inches(3.0), Inches(0.8), Inches(4.0), Inches(4.2)
+)
+ph.fill.solid()
+ph.fill.fore_color.rgb = RGBColor(0xF2, 0xF2, 0xF2)
+ph.line.color.rgb = RGBColor(0xBF, 0xBF, 0xBF)
+ph.line.width = Pt(1.5)
+ph.text_frame.text = "VISUALELEMENT — replace this rectangle with the author content figure (PNG/TIFF, 600 DPI, ~4×4.2 in)."
+for para in ph.text_frame.paragraphs:
+    para.font.size = Pt(10)
+    para.font.color.rgb = RGBColor(0x59, 0x59, 0x59)
+    para.alignment = 1  # center
+
+# ---- Slot 3: footer text box (reserved; usually empty in JACC templates) ----
+foot = slide.shapes.add_textbox(Inches(0.4), Inches(7.0), Inches(4.1), Inches(0.5))
+foot.text_frame.text = "FOOTERNOTE — optional secondary caption (often empty)."
+for para in foot.text_frame.paragraphs:
+    para.font.size = Pt(9)
+    para.font.color.rgb = RGBColor(0x80, 0x80, 0x80)
+    para.font.italic = True
+
+# ---- Slot 4: logo placeholder (rectangle as marker) ----
+logo = slide.shapes.add_shape(
+    MSO_SHAPE.RECTANGLE, Inches(7.3), Inches(6.7), Inches(2.7), Inches(0.8)
+)
+logo.fill.solid()
+logo.fill.fore_color.rgb = RGBColor(0xFF, 0xFF, 0xFF)
+logo.line.color.rgb = RGBColor(0xBF, 0xBF, 0xBF)
+logo.line.width = Pt(0.75)
+logo.text_frame.text = "JACCLOGO — JACC family logo (provided by editorial)."
+for para in logo.text_frame.paragraphs:
+    para.font.size = Pt(8)
+    para.font.color.rgb = RGBColor(0x80, 0x80, 0x80)
+    para.alignment = 1
+
+# Save
+OUT.parent.mkdir(parents=True, exist_ok=True)
+prs.save(OUT)
+print(f"Saved: {OUT}")
+print(f"Slide size: {Emu(prs.slide_width).inches:.2f} x {Emu(prs.slide_height).inches:.2f} in")
+print("Placeholders: ARTICLECITATION, VISUALELEMENT, FOOTERNOTE, JACCLOGO")