medsci-skills 4.1.0

package/skills/present-paper/references/generate_pptx_templates.py

@@ -0,0 +1,604 @@
+#!/usr/bin/env python3
+"""Reference template library for editable academic PPTX generation.
+
+Design tokens (NAVY/TEAL/ORANGE/GRAY, Apple SD Gothic Neo) and 16:9 fixed
+coordinate zones are defaults — adapt to journal/institution conventions
+as needed. NO markdown parsing, NO `cur_top` cumulative tracking (both
+produce drift artifacts).
+
+Each slide is defined as structured data → rendered by a type-specific
+template function (T_lead / T_text / T_table / T_image_right / T_two_col
+/ T_quote_slide / T_highlight_slide / T_metaphor_body / T_two_col_with_box
+/ T_table_two_col).
+
+`build_demo_slides()` at the bottom demonstrates every template with
+generic academic placeholder content; copy and replace with your own
+content function for a real deck.
+
+Speaker-note injection: `parse_notes()` reads a markdown script with
+`## Slide N:` headers and maps sections back to slide indices. For an
+existing deck where you only want to add notes (no design regeneration),
+use the sibling `inject_speaker_notes.py` instead.
+
+Mac PowerPoint compatibility — verify before delivery:
+  1. TIFF images: `find ppt/media -iname '*.tif*'` → convert to PNG
+  2. `<a:sp3d>` 3D bevels in rPr render as red outlines on Mac → strip
+  3. `docProps/app.xml` Slides/Notes/HeadingPairs/TitlesOfParts must
+     match actual slide count
+  4. `<a:srcRect>` values are 1/1000-percent (cap 100000) — never EMU
+
+See `references/workflow-checklist.md` and the user-level rule
+`~/.claude/rules/pptx-mac-compatibility.md` for the full checklist.
+"""
+
+import os
+import re
+from pptx import Presentation
+from pptx.util import Inches, Pt, Emu
+from pptx.dml.color import RGBColor
+from pptx.enum.text import PP_ALIGN, MSO_ANCHOR, MSO_AUTO_SIZE
+from pptx.enum.shapes import MSO_SHAPE
+from pptx.oxml.ns import qn
+
+# ── Design tokens ──────────────────────────────────────────────
+NAVY    = RGBColor(0x1B, 0x2A, 0x4A)
+TEAL    = RGBColor(0x00, 0x72, 0xB2)
+ORANGE  = RGBColor(0xD5, 0x5E, 0x00)
+GRAY    = RGBColor(0x33, 0x33, 0x33)
+WHITE   = RGBColor(0xFF, 0xFF, 0xFF)
+LGRAY   = RGBColor(0x99, 0x99, 0x99)
+BG_BLUE = RGBColor(0xF0, 0xF7, 0xFF)
+BG_YELLOW = RGBColor(0xFF, 0xF8, 0xE1)
+SUB_BLUE = RGBColor(0x7B, 0xB8, 0xE0)
+
+FONT = 'Apple SD Gothic Neo'
+SW = Inches(13.333)   # slide width
+SH = Inches(7.5)      # slide height
+ML = Inches(0.8)      # margin left
+MR = Inches(0.8)      # margin right
+MT = Inches(0.5)      # margin top
+CW = SW - ML - MR     # content width
+
+# Fixed vertical zones
+TITLE_Y = MT
+TITLE_H = Inches(0.8)
+SUB_Y   = TITLE_Y + TITLE_H
+SUB_H   = Inches(0.5)
+BODY_Y  = SUB_Y + SUB_H + Inches(0.1)   # ~1.9in from top
+BODY_H  = SH - BODY_Y - Inches(0.5)     # fill to bottom margin
+
+
+# ── Low-level helpers ──────────────────────────────────────────
+def _bg(slide, color):
+    slide.background.fill.solid()
+    slide.background.fill.fore_color.rgb = color
+
+def _text(slide, x, y, w, h, txt, sz=20, color=GRAY, bold=False,
+          align=PP_ALIGN.LEFT, auto_fit=False):
+    """Add a text box. Returns the shape."""
+    tb = slide.shapes.add_textbox(x, y, w, h)
+    tf = tb.text_frame
+    tf.word_wrap = True
+    if auto_fit:
+        tf.auto_size = MSO_AUTO_SIZE.TEXT_TO_FIT_SHAPE
+    p = tf.paragraphs[0]
+    p.alignment = align
+    _runs(p, txt, sz, color, bold)
+    return tb
+
+def _runs(para, txt, sz, color, bold):
+    """Add text with **bold** inline markup support."""
+    parts = re.split(r'(\*\*.*?\*\*)', txt)
+    first = True
+    for part in parts:
+        if not part:
+            continue
+        is_b = part.startswith('**') and part.endswith('**')
+        clean = part.strip('*') if is_b else part
+        run = para.runs[0] if first and para.runs else para.add_run()
+        first = False
+        run.text = clean
+        run.font.size = Pt(sz)
+        run.font.color.rgb = NAVY if is_b else color
+        run.font.bold = bold or is_b
+        run.font.name = FONT
+
+def _multiline(slide, x, y, w, h, lines, sz=20, color=GRAY, auto_fit=True):
+    """Add multi-line text box with bullet support."""
+    tb = slide.shapes.add_textbox(x, y, w, h)
+    tf = tb.text_frame
+    tf.word_wrap = True
+    if auto_fit:
+        tf.auto_size = MSO_AUTO_SIZE.TEXT_TO_FIT_SHAPE
+
+    first = True
+    for line in lines:
+        if not line.strip():
+            continue
+        if first:
+            p = tf.paragraphs[0]
+            first = False
+        else:
+            p = tf.add_paragraph()
+            p.space_before = Pt(6)
+
+        is_bullet = line.startswith('- ') or line.startswith('✓ ')
+        txt = line.lstrip('-✓ ').strip() if is_bullet else line.strip()
+
+        if line.strip().startswith('### '):
+            _runs(p, line.strip()[4:], sz + 2, TEAL, True)
+        else:
+            if is_bullet:
+                p.level = 0
+            _runs(p, txt, sz, color, False)
+    return tb
+
+def _title_block(slide, title, subtitle=None, width=None):
+    """Standard title + teal underline + optional subtitle."""
+    w = width or CW
+    _text(slide, ML, TITLE_Y, w, TITLE_H, title, sz=28, color=NAVY, bold=True)
+    # Teal underline
+    bar = slide.shapes.add_shape(MSO_SHAPE.RECTANGLE,
+                                  ML, TITLE_Y + Inches(0.7), w, Inches(0.05))
+    bar.fill.solid()
+    bar.fill.fore_color.rgb = TEAL
+    bar.line.fill.background()
+
+    if subtitle:
+        _text(slide, ML, SUB_Y, w, SUB_H, subtitle, sz=22, color=TEAL)
+
+def _table(slide, headers, rows, x, y, w, row_h=Inches(0.45)):
+    """Add a styled table."""
+    nr = len(rows) + 1
+    nc = len(headers)
+    if nc == 0:
+        return
+    shape = slide.shapes.add_table(nr, nc, x, y, w, row_h * nr)
+    tbl = shape.table
+
+    for j, h in enumerate(headers):
+        c = tbl.cell(0, j)
+        c.text = h.strip('*')
+        _cell_style(c, 15, WHITE, True, TEAL)
+
+    for i, row in enumerate(rows):
+        for j, val in enumerate(row):
+            if j >= nc:
+                break
+            c = tbl.cell(i + 1, j)
+            c.text = val.strip('*')
+            bg = RGBColor(0xF5, 0xF5, 0xF5) if i % 2 == 0 else WHITE
+            bld = val.strip().startswith('**')
+            _cell_style(c, 14, GRAY, bld, bg)
+
+def _cell_style(cell, sz, color, bold, bg):
+    for p in cell.text_frame.paragraphs:
+        for r in p.runs:
+            r.font.size = Pt(sz)
+            r.font.color.rgb = color
+            r.font.bold = bold
+            r.font.name = FONT
+    cell.vertical_anchor = MSO_ANCHOR.MIDDLE
+    tcPr = cell._tc.get_or_add_tcPr()
+    sf = tcPr.makeelement(qn('a:solidFill'), {})
+    sf.append(sf.makeelement(qn('a:srgbClr'), {'val': str(bg)}))
+    tcPr.append(sf)
+
+def _quote(slide, txt, x, y, w, h, sz=20):
+    """Blockquote with teal left bar + blue background."""
+    # Bar
+    bar = slide.shapes.add_shape(MSO_SHAPE.RECTANGLE, x, y, Inches(0.08), h)
+    bar.fill.solid(); bar.fill.fore_color.rgb = TEAL
+    bar.line.fill.background()
+    # Background
+    bg = slide.shapes.add_shape(MSO_SHAPE.RECTANGLE,
+                                 x + Inches(0.08), y, w - Inches(0.08), h)
+    bg.fill.solid(); bg.fill.fore_color.rgb = BG_BLUE
+    bg.line.fill.background()
+    # Text
+    _text(slide, x + Inches(0.3), y + Inches(0.1),
+          w - Inches(0.5), h - Inches(0.2), txt, sz=sz, auto_fit=True)
+
+def _highlight(slide, lines, x, y, w, h):
+    """Yellow highlight box with orange border."""
+    box = slide.shapes.add_shape(MSO_SHAPE.ROUNDED_RECTANGLE, x, y, w, h)
+    box.fill.solid(); box.fill.fore_color.rgb = BG_YELLOW
+    box.line.color.rgb = ORANGE; box.line.width = Pt(2)
+    _multiline(slide, x + Inches(0.3), y + Inches(0.2),
+               w - Inches(0.6), h - Inches(0.4), lines, sz=18, auto_fit=True)
+
+def _metaphor(slide, lines, x, y, w, h):
+    """Light-blue metaphor box with teal left bar."""
+    bar = slide.shapes.add_shape(MSO_SHAPE.RECTANGLE, x, y, Inches(0.08), h)
+    bar.fill.solid(); bar.fill.fore_color.rgb = TEAL
+    bar.line.fill.background()
+    bg = slide.shapes.add_shape(MSO_SHAPE.RECTANGLE,
+                                 x + Inches(0.08), y, w - Inches(0.08), h)
+    bg.fill.solid(); bg.fill.fore_color.rgb = BG_BLUE
+    bg.line.fill.background()
+    _multiline(slide, x + Inches(0.3), y + Inches(0.15),
+               w - Inches(0.6), h - Inches(0.3), lines, sz=17, auto_fit=True)
+
+def _image(slide, path, x, y, w, h):
+    """Add image preserving aspect ratio, centered in box."""
+    if not os.path.exists(path):
+        print(f'  WARNING: {path} not found')
+        return
+    from PIL import Image as PILImage
+    with PILImage.open(path) as img:
+        iw, ih = img.size
+    aspect = iw / ih
+    tw, th = int(w), int(h)
+    if tw / th > aspect:
+        fw, fh = int(th * aspect), th
+    else:
+        fw, fh = tw, int(tw / aspect)
+    ax = x + (tw - fw) // 2
+    ay = y + (th - fh) // 2
+    slide.shapes.add_picture(path, ax, ay, fw, fh)
+
+def _slidenum(slide, n, total):
+    _text(slide, SW - Inches(1.2), SH - Inches(0.45), Inches(1.0), Inches(0.3),
+          f'{n} / {total}', sz=10, color=LGRAY, align=PP_ALIGN.RIGHT)
+
+
+# ── Slide type templates ───────────────────────────────────────
+def T_lead(prs, title, subtitle='', extra=''):
+    """Section divider: navy bg, centered white text."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _bg(s, NAVY)
+    _text(s, ML, Inches(2.2), CW, Inches(1.2), title,
+          sz=40, color=WHITE, bold=True, align=PP_ALIGN.CENTER)
+    if subtitle:
+        _text(s, ML, Inches(3.6), CW, Inches(0.8), subtitle,
+              sz=24, color=SUB_BLUE, align=PP_ALIGN.CENTER)
+    if extra:
+        _text(s, ML, Inches(4.6), CW, Inches(1.0), extra,
+              sz=18, color=LGRAY, align=PP_ALIGN.CENTER)
+    return s
+
+def T_text(prs, title, body_lines, subtitle=None):
+    """Title + bullet/text body. Most common type."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    _multiline(s, ML, BODY_Y, CW, BODY_H, body_lines, sz=20)
+    return s
+
+def T_table(prs, title, headers, rows, subtitle=None, body_before=None):
+    """Title + optional body text + table."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    tbl_y = BODY_Y
+    if body_before:
+        _multiline(s, ML, BODY_Y, CW, Inches(1.5), body_before, sz=20)
+        tbl_y = BODY_Y + Inches(1.6)
+    _table(s, headers, rows, ML, tbl_y, CW)
+    return s
+
+def T_image_right(prs, title, body_lines, img_path, img_pct=30, subtitle=None):
+    """Title + body on left, image on right."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    frac = img_pct / 100
+    tw = int(SW * (1 - frac) - ML - Inches(0.3))
+    iw = int(SW * frac)
+    ix = SW - iw
+
+    _title_block(s, title, subtitle, width=tw)
+    _multiline(s, ML, BODY_Y, tw, BODY_H, body_lines, sz=20)
+    _image(s, img_path, ix, Inches(0.2), iw, SH - Inches(0.4))
+    return s
+
+def T_quote_slide(prs, title, quotes, body_after=None, img_path=None,
+                  img_pct=30, subtitle=None):
+    """Title + blockquote(s). Optionally with image."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+
+    if img_path:
+        frac = img_pct / 100
+        tw = int(SW * (1 - frac) - ML - Inches(0.3))
+        iw = int(SW * frac)
+        ix = SW - iw
+        _title_block(s, title, subtitle, width=tw)
+        _image(s, img_path, ix, Inches(0.2), iw, SH - Inches(0.4))
+    else:
+        tw = CW
+        _title_block(s, title, subtitle)
+
+    # Distribute quotes evenly in body area
+    n = len(quotes)
+    gap = Inches(0.15)
+    q_h = int((BODY_H - gap * (n - 1)) / n) if n else BODY_H
+
+    for i, q in enumerate(quotes):
+        qy = BODY_Y + (q_h + gap) * i
+        _quote(s, q, ML, qy, tw, q_h, sz=18)
+
+    if body_after:
+        _text(s, ML, SH - Inches(1.2), tw, Inches(0.6),
+              body_after, sz=16, color=GRAY)
+    return s
+
+def T_two_col(prs, title, left_lines, right_lines, subtitle=None):
+    """Two-column layout."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    cw = int((CW - Inches(0.5)) / 2)
+    gap = Inches(0.5)
+    _multiline(s, ML, BODY_Y, cw, BODY_H, left_lines, sz=18)
+    _multiline(s, ML + cw + gap, BODY_Y, cw, BODY_H, right_lines, sz=18)
+    return s
+
+def T_two_col_with_box(prs, title, left_lines, right_lines, subtitle=None,
+                       metaphor_col=None, metaphor_lines=None):
+    """Two-column with metaphor/highlight box in one column."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    cw = int((CW - Inches(0.5)) / 2)
+    gap = Inches(0.5)
+
+    # Left column
+    if metaphor_col == 'left' and metaphor_lines:
+        _metaphor(s, metaphor_lines, ML, BODY_Y, cw, Inches(2.5))
+        _multiline(s, ML, BODY_Y + Inches(2.7), cw, BODY_H - Inches(2.7),
+                   left_lines, sz=18)
+    else:
+        _multiline(s, ML, BODY_Y, cw, BODY_H, left_lines, sz=18)
+
+    # Right column
+    rx = ML + cw + gap
+    if metaphor_col == 'right' and metaphor_lines:
+        _metaphor(s, metaphor_lines, rx, BODY_Y, cw, Inches(2.5))
+        _multiline(s, rx, BODY_Y + Inches(2.7), cw, BODY_H - Inches(2.7),
+                   right_lines, sz=18)
+    else:
+        _multiline(s, rx, BODY_Y, cw, BODY_H, right_lines, sz=18)
+    return s
+
+def T_highlight_slide(prs, title, highlight_lines, subtitle=None,
+                      body_before=None):
+    """Slide with prominent highlight box."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    hy = BODY_Y
+    if body_before:
+        _multiline(s, ML, BODY_Y, CW, Inches(1.0), body_before, sz=20)
+        hy = BODY_Y + Inches(1.2)
+    _highlight(s, highlight_lines, ML, hy, CW, SH - hy - Inches(0.5))
+    return s
+
+def T_metaphor_body(prs, title, body_lines, metaphor_lines, subtitle=None):
+    """Body text + metaphor box below."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    mid = BODY_Y + BODY_H * 0.55
+    _multiline(s, ML, BODY_Y, CW, mid - BODY_Y - Inches(0.1),
+               body_lines, sz=20)
+    _metaphor(s, metaphor_lines, ML, mid, CW, SH - mid - Inches(0.5))
+    return s
+
+def T_table_two_col(prs, title, left_lines, headers, rows, subtitle=None):
+    """Left text + right table."""
+    s = prs.slides.add_slide(prs.slide_layouts[6])
+    _title_block(s, title, subtitle)
+    cw = int((CW - Inches(0.5)) / 2)
+    gap = Inches(0.5)
+    _multiline(s, ML, BODY_Y, cw, BODY_H, left_lines, sz=18)
+    _table(s, headers, rows, ML + cw + gap, BODY_Y, cw)
+    return s
+
+
+# ── Speaker notes ──────────────────────────────────────────────
+def parse_notes(path):
+    with open(path, 'r', encoding='utf-8') as f:
+        content = f.read()
+    # Match either `## Slide N:` (English) or legacy `--- 슬라이드 N: ... ---`
+    pattern = r'(?:^##\s*Slide\s+(\d+)\s*:|---\s*슬라이드\s*(\d+)\s*:.*?---)'
+    matches = list(re.finditer(pattern, content, flags=re.MULTILINE))
+    notes = {}
+    for i, m in enumerate(matches):
+        num = int(m.group(1) or m.group(2))
+        start = m.end()
+        end = matches[i + 1].start() if i + 1 < len(matches) else len(content)
+        raw = content[start:end].strip()
+        # Clean
+        lines = []
+        for line in raw.split('\n'):
+            s = line.strip()
+            if s.startswith('## Part ') or s == '---' or s.startswith('```'):
+                continue
+            if s.startswith('*(') and s.endswith(')*'):
+                lines.append(f'[{s[2:-2].strip()}]')
+            else:
+                lines.append(s.replace('**', '').replace('*', ''))
+        text = re.sub(r'\n{3,}', '\n\n', '\n'.join(lines).strip())
+        if text:
+            notes[num] = text
+    return notes
+
+
+# ── Demo: showcase every template ───────────────────────────────
+def build_demo_slides(prs):
+    """Generic academic demo — one slide per template type.
+
+    Replace this function with your own content builder. Keep all content
+    as inline structured data (lists, dicts) — never parse markdown.
+    """
+    slides = []
+
+    # 1. Title (T_lead)
+    slides.append(T_lead(
+        prs,
+        'Editable Academic PPTX — Template Showcase',
+        'A python-pptx reference deck',
+        'Speaker · Affiliation · Date'))
+
+    # 2. Section divider (T_lead)
+    slides.append(T_lead(
+        prs,
+        'Section 1: Background',
+        'Setting up the question'))
+
+    # 3. Bullet body (T_text)
+    slides.append(T_text(
+        prs,
+        'Background',
+        ['- Disease X affects an estimated **N patients** worldwide.',
+         '- Standard imaging modality: **modality Y**.',
+         '- Existing limitations:',
+         '  - Inter-reader variability',
+         '  - Long acquisition time',
+         '  - Reader fatigue on volume studies'],
+        subtitle='Why this study matters'))
+
+    # 4. Two-column compare (T_two_col)
+    slides.append(T_two_col(
+        prs,
+        'Conventional vs. Proposed Approach',
+        ['### Conventional',
+         '- Manual segmentation',
+         '- ~30 min per case',
+         '- Reader-dependent'],
+        ['### Proposed',
+         '- Deep-learning assisted',
+         '- ~3 min per case',
+         '- Standardized output'],
+        subtitle='Workflow comparison'))
+
+    # 5. Quote slide (T_quote_slide)
+    slides.append(T_quote_slide(
+        prs,
+        'Prior Evidence',
+        ['"AI-assisted reading reduced inter-reader variability '
+         'by 40% in a multi-center study." — Author et al., Journal Y, 2024',
+         '"External validation remains the principal gap in clinical '
+         'translation of imaging AI." — Reviewer A, 2025']))
+
+    # 6. Table (T_table)
+    slides.append(T_table(
+        prs,
+        'Cohort Characteristics',
+        ['Variable', 'Training (n=800)', 'Internal test (n=200)', 'External (n=300)'],
+        [['Age, mean (SD)', '62 (11)', '63 (10)', '60 (12)'],
+         ['Female, n (%)', '420 (52.5)', '105 (52.5)', '160 (53.3)'],
+         ['Disease prevalence', '38%', '40%', '35%'],
+         ['**Imaging vendor mix**', 'A/B/C', 'A/B/C', 'D/E']],
+        subtitle='Demographics across splits'))
+
+    # 7. Image-right (T_image_right) — pass empty path to skip image safely
+    slides.append(T_image_right(
+        prs,
+        'Methods Pipeline',
+        ['- DICOM ingest with anonymization',
+         '- Preprocessing: resampling, normalization',
+         '- Model: 3D U-Net (or transformer variant)',
+         '- Output: probability map + binary mask',
+         '- QA: radiologist override loop'],
+        img_path='figures/pipeline.png',  # replace with real asset
+        img_pct=35,
+        subtitle='From DICOM to decision'))
+
+    # 8. Highlight slide (T_highlight_slide)
+    slides.append(T_highlight_slide(
+        prs,
+        'Primary Result',
+        ['**AUC = 0.92 (95% CI 0.89–0.95)** on external test set',
+         'Sensitivity 0.88 · Specificity 0.91',
+         'Non-inferior to senior reader (Δ AUC = 0.01, p = 0.78)'],
+        subtitle='Diagnostic performance',
+        body_before=['On the prospectively collected external cohort '
+                     '(n = 300, 5 sites):']))
+
+    # 9. Metaphor body (T_metaphor_body)
+    slides.append(T_metaphor_body(
+        prs,
+        'Interpretation',
+        ['- Performance preserved across vendor and protocol shifts.',
+         '- Calibration check: ECE = 0.04 → clinically usable.',
+         '- Failure modes concentrated in motion-degraded studies.'],
+        ['Think of the model as a **second reader** — fast, consistent, '
+         'never tired — but still requiring a final human sign-off on '
+         'edge cases.'],
+        subtitle='What it means in practice'))
+
+    # 10. Two-col with metaphor box (T_two_col_with_box)
+    slides.append(T_two_col_with_box(
+        prs,
+        'Limitations and Mitigations',
+        ['### Limitations',
+         '- Single-modality input',
+         '- Retrospective external set',
+         '- No outcome data yet'],
+        ['### Planned next steps',
+         '- Multi-modal fusion',
+         '- Prospective registration',
+         '- 12-month outcome follow-up'],
+        metaphor_col='right',
+        metaphor_lines=['**Open question** — is diagnostic accuracy a '
+                        'sufficient endpoint, or do we need clinical '
+                        'utility evidence?'],
+        subtitle='Honest accounting'))
+
+    # 11. Table-two-col (T_table_two_col)
+    slides.append(T_table_two_col(
+        prs,
+        'Comparison with Published Models',
+        ['### Take-aways',
+         '- Comparable AUC at lower compute',
+         '- Better calibration on minority subgroup',
+         '- Released checkpoint + DICOM-SR exporter'],
+        ['Model', 'AUC', 'Params'],
+        [['Baseline', '0.85', '11M'],
+         ['Prior SOTA', '0.91', '78M'],
+         ['**Ours**', '**0.92**', '**24M**']],
+        subtitle='Where this work sits'))
+
+    # 12. Take-home (T_text)
+    slides.append(T_text(
+        prs,
+        'Take-Home Messages',
+        ['1. The model generalized to a **5-site external cohort** '
+         '(AUC 0.92).',
+         '2. Performance was **non-inferior to a senior reader**.',
+         '3. **Calibration and failure-mode transparency** were the '
+         'features that earned clinician trust — not raw AUC.',
+         '4. Next: prospective validation tied to clinical outcomes.'],
+        subtitle='Three points to remember'))
+
+    return slides
+
+
+# ── Main ────────────────────────────────────────────────────────
+def main():
+    prs = Presentation()
+    prs.slide_width = SW
+    prs.slide_height = SH
+
+    print('Building demo slides...')
+    slides = build_demo_slides(prs)
+    total = len(slides)
+    print(f'  {total} slides created')
+
+    # Slide numbers (skip lead/section-divider slides with solid bg)
+    for i, slide in enumerate(slides):
+        bg_type = slide.background.fill.type
+        is_lead = bg_type is not None and 'SOLID' in str(bg_type)
+        if not is_lead:
+            _slidenum(slide, i + 1, total)
+
+    # Optional speaker-note injection — uncomment if a script file exists
+    # notes = parse_notes('speaker_script.md')
+    # for i, slide in enumerate(slides):
+    #     num = i + 1
+    #     if num in notes:
+    #         slide.notes_slide.notes_text_frame.text = notes[num]
+
+    out = 'demo_slide_deck.pptx'
+    prs.save(out)
+    sz = os.path.getsize(out) / 1024
+    print(f'\nDone: {out}')
+    print(f'  Slides: {total}')
+    print(f'  Size: {sz:.0f} KB, Editable: YES')
+
+
+if __name__ == '__main__':
+    main()