aiox-core 5.0.5 → 5.0.7

package/pro/squads/aiox-sop/outputs/SOP-SLIDES-004.md

@@ -0,0 +1,975 @@
+# SOP-SLIDES-004 — Multi-Output Slide Pipeline
+
+| Field | Value |
+|---|---|
+| **SOP ID** | SOP-SLIDES-004 |
+| **Version** | 1.0.0 |
+| **Effective Date** | 2026-03-16 |
+| **Classification** | Public |
+| **Status** | DRAFT |
+| **Consumer** | LLM agents, slide-chief orchestrator |
+| **Companions** | SOP-SLIDES-001 (process), SOP-SLIDES-002 (AIOX TSX), SOP-SLIDES-003 (agnostic TSX) |
+| **Research Base** | `docs/research/2026-03-16-slide-generator-squad/` (6 reports, 36 sources) |
+| **Review Date** | 2026-09-16 |
+
+---
+
+## 1. PURPOSE
+
+This SOP defines the **multi-output slide pipeline**: from briefing to delivery across multiple rendering targets. It introduces a **canonical JSON interchange schema** that decouples content generation from rendering, enabling a single content pass to produce TSX Web, PPTX, and Markdown (Slidev/Marp) outputs.
+
+It also formalizes the **evaluation framework** combining PPTEval (3 dimensions) and GAD Score from the research literature.
+
+**This SOP does NOT duplicate** slide type patterns, brand config, or component APIs. Those live in SOP-002 (AIOX) and SOP-003 (agnostic). This SOP references them.
+
+---
+
+## 2. JSON INTERCHANGE SCHEMA
+
+The canonical JSON schema is the **single source of truth** between content generation (Step 3) and rendering (Step 4). Every renderer consumes this same JSON.
+
+### 2.1 Presentation Schema
+
+```json
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "type": "object",
+  "required": ["presentation"],
+  "properties": {
+    "presentation": {
+      "type": "object",
+      "required": ["title", "mode", "format", "aspect_ratio", "slides"],
+      "properties": {
+        "title": { "type": "string", "maxLength": 120 },
+        "subtitle": { "type": "string", "maxLength": 200 },
+        "author": { "type": "string" },
+        "date": { "type": "string", "format": "date" },
+        "mode": { "enum": ["palco", "live", "async"] },
+        "format": { "enum": ["ted_keynote", "pitch_deck", "sales_deck", "technical", "zoom_virtual", "carousel_stories"] },
+        "aspect_ratio": { "enum": ["16:9", "16:10", "9:16", "1:1"] },
+        "brand": { "$ref": "#/$defs/brand_config" },
+        "slides": {
+          "type": "array",
+          "minItems": 1,
+          "items": { "$ref": "#/$defs/slide" }
+        }
+      }
+    }
+  }
+}
+```
+
+### 2.2 Slide Schema
+
+```json
+{
+  "$defs": {
+    "slide": {
+      "type": "object",
+      "required": ["id", "type", "content"],
+      "properties": {
+        "id": { "type": "string", "pattern": "^slide-\\d{3}$" },
+        "type": {
+          "enum": [
+            "TITLE", "SECTION_BREAK", "STATEMENT", "QUOTE",
+            "CONTENT", "COMPARISON", "METRIC", "DATA_VIZ",
+            "IMAGE", "BUILD", "CODE", "CLOSING"
+          ]
+        },
+        "title": { "type": "string", "maxLength": 80 },
+        "content": { "type": "object", "description": "Type-specific fields (see 2.3)" },
+        "layout": {
+          "enum": [
+            "full_center", "left_aligned", "split_50_50",
+            "split_60_40", "grid_1x2", "grid_1x3"
+          ]
+        },
+        "decoratives": {
+          "type": "array",
+          "items": { "enum": ["CornerMarks", "MetaBar", "WatermarkNumber", "Watermark", "PageFooter", "SectionTag", "TechFrame"] }
+        },
+        "animation": {
+          "type": "object",
+          "properties": {
+            "preset": { "enum": ["fadeUp", "fadeIn", "scaleIn", "slideRight", "slideLeft", "growWidth", "growHeight"] },
+            "stagger": { "type": "object", "properties": { "base": { "type": "number", "default": 0.1 }, "increment": { "type": "number", "default": 0.08 } }, "description": "For lists/grids: delay = base + index * increment" },
+            "delay": { "type": "number", "minimum": 0, "maximum": 1.5 }
+          }
+        },
+        "speaker_notes": { "type": "string", "maxLength": 500 },
+        "duration_s": { "type": "integer", "minimum": 3, "maximum": 120 }
+      }
+    }
+  }
+}
+```
+
+### 2.3 Type-Specific Content Fields
+
+Each slide `type` has a specific `content` shape:
+
+#### TITLE
+
+```json
+{
+  "title": "string (max 5 words)",
+  "subtitle": "string",
+  "speaker": "string",
+  "date": "string",
+  "section_tag": "string"
+}
+```
+
+#### SECTION_BREAK
+
+```json
+{
+  "section_number": "string (e.g. '02')",
+  "section_label": "string (e.g. 'Parte 02')",
+  "title": "string (max 3 words)"
+}
+```
+
+#### STATEMENT
+
+```json
+{
+  "lines": ["string (each line, max 10 words total)"],
+  "highlight_indices": [0]
+}
+```
+
+`highlight_indices` marks which lines use accent color. Max 2.
+
+#### QUOTE
+
+```json
+{
+  "quote_text": "string",
+  "highlight_portion": "string (optional — portion to accent)",
+  "attribution_name": "string",
+  "attribution_company": "string (optional)"
+}
+```
+
+#### CONTENT
+
+```json
+{
+  "section_tag": "string",
+  "title": "string",
+  "subtitle": "string (optional)",
+  "items": [
+    {
+      "marker": "string (I, II, 01, etc.)",
+      "title": "string (keyword, not sentence)",
+      "description": "string (short fragment)"
+    }
+  ]
+}
+```
+
+Max 4 items in Palco/Live. Max 6 in Async.
+
+#### COMPARISON
+
+```json
+{
+  "section_tag": "string",
+  "title": "string",
+  "left": {
+    "label": "string (e.g. BEFORE)",
+    "style": "negative",
+    "items": ["string"]
+  },
+  "right": {
+    "label": "string (e.g. AFTER)",
+    "style": "positive",
+    "items": ["string"]
+  }
+}
+```
+
+#### METRIC
+
+```json
+{
+  "section_tag": "string (optional — label for SectionTag decorative)",
+  "metrics": [
+    {
+      "value": "string (e.g. '10x', '98%', '$2.4M')",
+      "label": "string (context above value)",
+      "unit": "string (context below value)"
+    }
+  ]
+}
+```
+
+Max 3 metrics per slide.
+
+#### DATA_VIZ
+
+```json
+{
+  "section_tag": "string",
+  "conclusion_title": "string (title IS the conclusion, not the description)",
+  "chart_type": "bar_vertical|bar_horizontal|line|donut|table",
+  "data": [
+    { "label": "string", "value": "number", "series": "string (optional)" }
+  ],
+  "source": "string (data attribution)"
+}
+```
+
+**Constraints (validation, NOT included in JSON output):**
+- Max elements per chart_type: bar_vertical=6, bar_horizontal=8, line=12, donut=4, table=4x6
+- Forbidden chart types: scatter, heatmap, treemap, sankey
+- Max 3 series per chart
+
+#### IMAGE
+
+```json
+{
+  "section_tag": "string (optional)",
+  "title": "string (max 8 words, optional)",
+  "image_url": "string (URL or asset path)",
+  "image_alt": "string (required, accessibility)",
+  "overlay": "gradient_bottom|gradient_left|none",
+  "caption": "string (max 8 words, optional)"
+}
+```
+
+Min 1920px wide. WebP preferred, max 500KB. No stock photos.
+
+#### BUILD
+
+```json
+{
+  "section_tag": "string",
+  "title": "string",
+  "orientation": "horizontal|vertical",
+  "steps": [
+    {
+      "number": "string (01, 02, ...)",
+      "title": "string",
+      "description": "string",
+      "time_label": "string (optional, e.g. 'Week 1-2')"
+    }
+  ]
+}
+```
+
+2-5 steps. In progressive mode, future steps render at 15% opacity.
+
+#### CODE
+
+```json
+{
+  "section_tag": "string (optional)",
+  "title": "string",
+  "language": "string (e.g. 'typescript')",
+  "code": "string (max 12 visible lines)",
+  "highlight_lines": [1, 3]
+}
+```
+
+Min font = body size. Never screenshot code.
+
+#### CLOSING
+
+```json
+{
+  "title": "string (max 5 words, e.g. 'THANK YOU.')",
+  "cta_text": "string (e.g. 'QUESTIONS?')",
+  "contact_lines": ["string (email, url, etc.)"]
+}
+```
+
+### 2.4 Brand Config Reference
+
+```json
+{
+  "$defs": {
+    "brand_config": {
+      "oneOf": [
+        { "type": "string", "description": "Reference to external brand config file path" },
+        {
+          "type": "object",
+          "description": "Inline brand config (SOP-003 schema)",
+          "required": ["name", "color", "font"],
+          "properties": {
+            "name": { "type": "string" },
+            "prefix": { "type": "string" },
+            "url": { "type": "string" },
+            "color": {
+              "type": "object",
+              "required": ["bg_primary", "bg_accent", "text_primary", "text_secondary", "accent"],
+              "properties": {
+                "bg_primary": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "bg_accent": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "surface": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "text_primary": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "text_secondary": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "accent": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "border": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" },
+                "warning": { "type": "string", "pattern": "^#[0-9A-Fa-f]{6}$" }
+              }
+            },
+            "font": {
+              "type": "object",
+              "required": ["sans", "mono"],
+              "properties": {
+                "sans": { "type": "string" },
+                "sans_fallback": { "type": "string" },
+                "mono": { "type": "string" },
+                "mono_fallback": { "type": "string" }
+              }
+            }
+          }
+        }
+      ]
+    }
+  }
+}
+```
+
+### 2.5 Complete Example — 5-Slide Pitch Deck
+
+```json
+{
+  "presentation": {
+    "title": "Acme AI Platform",
+    "subtitle": "Series A Pitch",
+    "author": "Jane Smith",
+    "date": "2026-03-16",
+    "mode": "palco",
+    "format": "pitch_deck",
+    "aspect_ratio": "16:9",
+    "brand": {
+      "name": "Acme",
+      "prefix": "ac",
+      "url": "acme.ai",
+      "color": {
+        "bg_primary": "#050505",
+        "bg_accent": "#3B82F6",
+        "surface": "#111111",
+        "text_primary": "#F8FAFC",
+        "text_secondary": "#94A3B8",
+        "accent": "#3B82F6",
+        "border": "#1E293B",
+        "warning": "#EF4444"
+      },
+      "font": {
+        "sans": "Inter",
+        "sans_fallback": "system-ui, sans-serif",
+        "mono": "JetBrains Mono",
+        "mono_fallback": "monospace"
+      }
+    },
+    "slides": [
+      {
+        "id": "slide-001",
+        "type": "TITLE",
+        "title": "Acme AI Platform",
+        "content": {
+          "title": "ACME AI",
+          "subtitle": "Series A Pitch",
+          "speaker": "Jane Smith, CEO",
+          "date": "March 2026",
+          "section_tag": "[01] - Introduction"
+        },
+        "layout": "full_center",
+        "decoratives": ["CornerMarks", "MetaBar", "WatermarkNumber"],
+        "animation": { "preset": "fadeUp", "delay": 0.3 },
+        "speaker_notes": "Welcome everyone. Today I'll present our vision for AI-powered automation.",
+        "duration_s": 15
+      },
+      {
+        "id": "slide-002",
+        "type": "STATEMENT",
+        "title": "The Problem",
+        "content": {
+          "lines": ["80% OF ENTERPRISE", "WORKFLOWS ARE", "STILL MANUAL"],
+          "highlight_indices": [2]
+        },
+        "layout": "full_center",
+        "decoratives": [],
+        "animation": { "preset": "scaleIn", "delay": 0.2 },
+        "speaker_notes": "The core problem we solve: most business processes haven't been touched by AI yet.",
+        "duration_s": 8
+      },
+      {
+        "id": "slide-003",
+        "type": "METRIC",
+        "title": "Traction",
+        "content": {
+          "metrics": [
+            { "value": "$2.4M", "label": "ARR", "unit": "Annual Recurring Revenue" },
+            { "value": "340%", "label": "YoY Growth", "unit": "Year over year" },
+            { "value": "98%", "label": "Retention", "unit": "Net Revenue Retention" }
+          ]
+        },
+        "layout": "grid_1x3",
+        "decoratives": ["SectionTag"],
+        "animation": { "preset": "scaleIn", "delay": 0.3 },
+        "speaker_notes": "Our numbers speak for themselves. Triple-digit growth with near-perfect retention.",
+        "duration_s": 15
+      },
+      {
+        "id": "slide-004",
+        "type": "COMPARISON",
+        "title": "Before & After",
+        "content": {
+          "section_tag": "[04] - Impact",
+          "title": "BEFORE & AFTER",
+          "left": {
+            "label": "WITHOUT ACME",
+            "style": "negative",
+            "items": ["3 weeks per workflow", "5 engineers required", "40% error rate"]
+          },
+          "right": {
+            "label": "WITH ACME",
+            "style": "positive",
+            "items": ["2 hours per workflow", "1 operator required", "0.5% error rate"]
+          }
+        },
+        "layout": "split_50_50",
+        "decoratives": ["SectionTag"],
+        "animation": { "preset": "slideRight", "delay": 0.3 },
+        "speaker_notes": "The transformation is dramatic. We reduce both time and error rates by orders of magnitude.",
+        "duration_s": 25
+      },
+      {
+        "id": "slide-005",
+        "type": "CLOSING",
+        "title": "Thank You",
+        "content": {
+          "title": "THANK YOU.",
+          "cta_text": "QUESTIONS?",
+          "contact_lines": ["jane@acme.ai", "acme.ai"]
+        },
+        "layout": "full_center",
+        "decoratives": ["CornerMarks"],
+        "animation": { "preset": "fadeUp", "delay": 0.2 },
+        "speaker_notes": "Thank you for your time. I'm happy to answer any questions.",
+        "duration_s": 10
+      }
+    ]
+  }
+}
+```
+
+---
+
+## 3. PIPELINE STEPS
+
+### Step 1: FORMAT SELECTION
+
+Select the presentation format. Format determines mode, constraints, and dominant slide types.
+
+```
+INPUT:  briefing (topic, audience, context, duration)
+OUTPUT: format, mode (override), slide count range, dominant types
+
+IF audience = investors AND purpose = fundraise:
+  format = "pitch_deck"
+ELSE IF audience = customers AND purpose = sell:
+  format = "sales_deck"
+ELSE IF context = conference AND duration >= 15min:
+  format = "ted_keynote"
+ELSE IF topic = technical AND audience = developers:
+  format = "technical"
+ELSE IF context in [zoom, meet, teams]:
+  format = "zoom_virtual"
+ELSE IF channel in [instagram, linkedin, stories]:
+  format = "carousel_stories"
+ELSE:
+  format = "pitch_deck" (safest default)
+```
+
+Format constraints (from SOP-002/003 section 4.0):
+
+| Format | Slides | Text Density | Bullets | Self-Explanatory |
+|---|---|---|---|---|
+| ted_keynote | 40-90 | Minimal | Never | Never |
+| pitch_deck | 10-15 | Moderate | Rare | Yes (leave-behind) |
+| sales_deck | 8-25 | Moderate | Allowed | Yes (champion) |
+| technical | 20-40 | Moderate+ | Allowed | Partial |
+| zoom_virtual | Variable | Reduced | Cautious | Depends |
+| carousel_stories | 5-10 | Minimal | Never | Always |
+
+### Step 2: MODE + ASPECT RATIO
+
+Mode is often determined by format (Step 1). If not locked, select by context:
+
+```
+IF context in [conference, keynote, workshop, presencial]:
+  mode = "palco"
+ELSE IF context in [youtube, twitch, screenshare, live_call]:
+  mode = "live"
+ELSE IF context in [post_event, course, onboarding, pdf]:
+  mode = "async"
+
+IF channel in [projector, monitor, stream]:    ratio = "16:9"
+ELSE IF channel in [macbook, mac_projector]:   ratio = "16:10"
+ELSE IF channel in [stories, reels, vertical]: ratio = "9:16"
+ELSE IF channel in [instagram, carousel]:      ratio = "1:1"
+ELSE:                                          ratio = "16:9"
+```
+
+Mode constraints:
+
+| Mode | Max Words | Min Font | Whitespace | Animations | Speaker Notes |
+|---|---|---|---|---|---|
+| palco | 15 | 2.2vw | >= 50% | Yes | Required |
+| live | 15 | 2.2vw | >= 45% | Yes (-30% delay) | Required |
+| async | 30 | 1.6vw | >= 40% | None | Not needed |
+
+### Step 3: CONTENT GENERATION -> JSON
+
+The LLM generates the canonical JSON from the briefing + format + mode + brand config.
+
+**Prompt template for LLM content generation:**
+
+```
+You are a presentation content architect. Generate a slide deck in the canonical JSON interchange format.
+
+INPUTS:
+- Topic: {{topic}}
+- Audience: {{audience}}
+- Format: {{format}} (see constraints below)
+- Mode: {{mode}}
+- Aspect Ratio: {{ratio}}
+- Duration: {{duration_minutes}} minutes
+- Brand Config: {{brand_config}}
+- Key Messages: {{key_messages}}
+- Source Material: {{source_material}}
+
+FORMAT CONSTRAINTS:
+{{format_constraints from Step 1 table}}
+
+MODE CONSTRAINTS:
+{{mode_constraints from Step 2 table}}
+
+SLIDE TYPE TAXONOMY:
+TITLE, SECTION_BREAK, STATEMENT, QUOTE, CONTENT, COMPARISON, METRIC, DATA_VIZ, IMAGE, BUILD, CODE, CLOSING
+
+SEQUENCE RULES:
+- FIRST slide = TITLE (always)
+- LAST slide = CLOSING (always)
+- SECTION_BREAK between thematic blocks
+- Max 3 consecutive same type
+- Max 120s per slide
+- Insert rhythm resets (SECTION_BREAK, IMAGE, STATEMENT) in long sequences
+
+RHYTHM PATTERNS:
+- STATEMENT -> CONTENT -> CONTENT = Impact -> detail -> detail
+- METRIC -> DATA_VIZ -> CONTENT = Number -> context -> explanation
+- IMAGE -> STATEMENT = Visual -> emotional
+- BUILD (3 steps) = Narrative progression
+- COMPARISON -> STATEMENT = Contrast -> conclusion
+
+OUTPUT: Valid JSON following the presentation schema (section 2.1).
+Every slide must have: id, type, title, content (type-specific), layout, speaker_notes, duration_s.
+```
+
+**Validation after generation:**
+
+```
+ASSERT slides[0].type == "TITLE"
+ASSERT slides[-1].type == "CLOSING"
+ASSERT all slide.type in TAXONOMY
+ASSERT sum(duration_s) approximately matches duration_minutes * 60
+ASSERT no more than 3 consecutive same type
+ASSERT word count per slide <= mode.max_words (for text-heavy types)
+ASSERT JSON validates against schema (section 2.1 + 2.2)
+```
+
+### Step 4: RENDERING FORK
+
+The canonical JSON is consumed by one or more renderers in parallel.
+
+```
+┌─────────────────────────────────────────────────┐
+│  INPUT: Canonical JSON (from Step 3)            │
+│                                                 │
+│  IF target includes "tsx_web":                  │
+│    → TSX Renderer (4.1)                         │
+│                                                 │
+│  IF target includes "pptx":                     │
+│    → PPTX Renderer (4.2)                        │
+│                                                 │
+│  IF target includes "markdown":                 │
+│    → Markdown Renderer (4.3)                    │
+│                                                 │
+│  Renderers run independently. Same JSON in,     │
+│  different format out.                          │
+└─────────────────────────────────────────────────┘
+```
+
+### Step 5: EVALUATION
+
+Apply the evaluation framework (section 5) to each rendered output.
+
+```
+IF composite_score >= 7.0:
+  verdict = "PASS"
+ELSE IF composite_score >= 5.0:
+  verdict = "REVISE" — return to Step 3 with feedback
+ELSE:
+  verdict = "FAIL" — return to Step 1 to reassess format/approach
+
+Max revision iterations = 3
+After 3 revisions with score < 7.0: HALT, escalate to human review
+```
+
+### Step 6: DELIVERY
+
+```
+OUTPUT = rendered file(s) in requested format(s)
+
+IF format requires leave-behind (pitch_deck, sales_deck):
+  Generate TWO versions:
+    1. "palco" version (sparse, narrator-dependent)
+    2. "async" version (self-explanatory, more text)
+  Both from same canonical JSON with mode override
+
+DELIVERY ARTIFACTS:
+  - Rendered file(s) (.tsx / .pptx / .md)
+  - Canonical JSON (for future re-rendering)
+  - Speaker notes export (if mode != async)
+  - Evaluation scorecard
+```
+
+---
+
+## 4. RENDERER SPECIFICATIONS
+
+### 4.1 TSX Web Renderer
+
+**Delegates to SOP-002 (AIOX brand) or SOP-003 (any brand).**
+
+JSON-to-TSX mapping:
+
+| JSON Field | TSX Output |
+|---|---|
+| `presentation.brand` | Brand config resolution (SOP-003 section 0) |
+| `slide.type` | Component selection (SOP-002/003 section 4/6) |
+| `slide.layout` | Layout class/structure selection |
+| `slide.decoratives[]` | `<CornerMarks>`, `<MetaBar>`, `<WatermarkNumber>`, etc. |
+| `slide.animation` | `{...fadeUp(delay)}`, `{...scaleIn(delay)}`, etc. |
+| `slide.content.*` | Props passed to type-specific component |
+| `slide.speaker_notes` | Exported as separate data structure (not rendered) |
+| `presentation.mode` | Controls animation behavior, word limits, safe zones |
+
+**Rules:**
+- IF `presentation.brand.name == "AIOX"` → use SOP-002 patterns (hardcoded AIOX tokens)
+- ELSE → use SOP-003 patterns (CSS variable-based tokens from brand config)
+- ALWAYS generate `SlideLayout` wrapper per slide
+- ALWAYS apply `style={FONT_SANS}` or `style={FONT_MONO}` to every text element
+- IF `mode == "async"` → strip all animation props
+- IF `mode == "live"` → reduce animation delays by 30%
+
+### 4.2 PPTX Renderer
+
+**Uses Office-PowerPoint-MCP-Server (32 tools, python-pptx based).**
+
+JSON-to-MCP mapping:
+
+| JSON Field | MCP Tool(s) |
+|---|---|
+| `presentation` (root) | `create_presentation` → set dimensions from `aspect_ratio` |
+| `presentation.brand.font` | `set_font` on slide master |
+| `presentation.brand.color` | Theme colors via `set_theme_colors` |
+| `slide` (each) | `add_slide` with layout selection |
+| `slide.content.title` | `add_text_box` or `set_title` with font/size |
+| `slide.content.items[]` | `add_text_box` per item, or `add_table` for structured data |
+| `slide.content.metrics[]` | `add_text_box` with large font for value |
+| `slide.content.chart_type` | `add_chart` (bar, line, pie, etc.) |
+| `slide.content.image_url` | `add_image` with positioning |
+| `slide.content.code` | `add_text_box` with mono font, syntax-highlighted |
+| `slide.speaker_notes` | `set_notes` per slide |
+| `slide.decoratives` | Manual positioning: corner marks as lines, watermark as low-opacity text |
+
+**PPTX-specific constraints:**
+- No CSS animations (PPTX has its own animation system — map `animation.preset` to PowerPoint entrance effects)
+- Font embedding required for non-system fonts
+- Image resolution: export at 96 DPI minimum for screen, 300 DPI for print
+- Aspect ratio: set via `presentation.slide_width` and `presentation.slide_height` in EMU
+
+**Animation mapping (PPTX entrance effects):**
+
+| JSON Preset | PPTX Effect |
+|---|---|
+| fadeUp | Fly In (from bottom) + Fade |
+| fadeIn | Fade |
+| scaleIn | Grow & Turn or Zoom |
+| slideRight | Fly In (from left) |
+| slideLeft | Fly In (from right) |
+| growWidth | Wipe (left to right) |
+| growHeight | Wipe (bottom to top) |
+| stagger | Apply entrance effect delays sequentially (base + index × 0.08s) |
+
+**Aspect Ratio → PPTX Dimensions (EMU):**
+
+| Ratio | Width (EMU) | Height (EMU) | Pixels |
+|---|---|---|---|
+| 16:9 | 12192000 | 6858000 | 1920×1080 |
+| 16:10 | 12192000 | 7620000 | 1920×1200 |
+| 9:16 | 6858000 | 12192000 | 1080×1920 |
+| 1:1 | 6858000 | 6858000 | 1080×1080 |
+
+### 4.3 Markdown Renderer
+
+**Targets Slidev (Vue-based) or Marp (pure Markdown).**
+
+JSON-to-Markdown mapping:
+
+```markdown
+---
+# Slidev frontmatter (from presentation metadata)
+theme: {{theme_from_brand}}
+title: {{presentation.title}}
+author: {{presentation.author}}
+date: {{presentation.date}}
+aspectRatio: {{presentation.aspect_ratio}}
+---
+
+# Slide content follows, separated by ---
+```
+
+| JSON Field | Slidev Markdown | Marp Markdown |
+|---|---|---|
+| Slide separator | `---` | `---` |
+| `slide.type: TITLE` | `# Title` + YAML `layout: cover` | `# Title` + `<!-- _class: lead -->` |
+| `slide.type: SECTION_BREAK` | `layout: section` + `# Section` | `<!-- _class: invert -->` + `# Section` |
+| `slide.type: STATEMENT` | `layout: statement` + `# Bold text` (centered, large) | `<!-- _class: lead -->` + `# Bold text` |
+| `slide.type: CONTENT` | `## Title` + `- item` list | `## Title` + `- item` list |
+| `slide.type: QUOTE` | `> quote` + `— attribution` | `> quote` + `— attribution` |
+| `slide.type: COMPARISON` | Two-column `::left::` / `::right::` | HTML table or `<div>` grid |
+| `slide.type: METRIC` | Custom Vue component or `<div>` | `# Value` (styled via CSS) |
+| `slide.type: DATA_VIZ` | Mermaid/Chart.js integration | Mermaid block or image |
+| `slide.type: IMAGE` | `![alt](url)` | `![alt](url)` + `<!-- _backgroundImage -->` |
+| `slide.type: BUILD` | `v-clicks` directive (Slidev) | Numbered list |
+| `slide.type: CODE` | ` ```lang ` fenced code block | ` ```lang ` fenced code block |
+| `slide.type: CLOSING` | `layout: end` + `# Thank You` + contact info | `<!-- _class: lead -->` + `# Thank You` + contact |
+| `slide.speaker_notes` | `<!-- speaker notes -->` block | `<!-- notes -->` block |
+
+**Slidev-specific features:**
+- Use `v-click` for progressive reveal (maps to BUILD steps)
+- Use `<style>` blocks for brand colors
+- Use Windi CSS / UnoCSS for utility classes
+
+**Marp-specific features:**
+- Use `<!-- _class: -->` directives for layout
+- Use `<!-- _backgroundColor: -->` for brand colors
+- Export: `marp --pdf` for PDF, `marp --pptx` for PPTX
+
+---
+
+## 5. EVALUATION FRAMEWORK
+
+Composite score from 4 dimensions, adapted from PPTEval (PPTAgent V2) and GAD Score (SlideGen).
+
+### 5.1 Content Score (weight: 0.30)
+
+| Criterion | Score Range | Evaluation |
+|---|---|---|
+| Factual accuracy | 0-10 | Claims supported by source material |
+| Topic relevance | 0-10 | Every slide serves the presentation goal |
+| Completeness | 0-10 | Key messages covered, no gaps |
+| Word economy | 0-10 | Respects mode word limits, no filler |
+
+`content_score = mean(factual, relevance, completeness, word_economy)`
+
+### 5.2 Design Score (weight: 0.30)
+
+| Criterion | Score Range | Evaluation |
+|---|---|---|
+| Visual consistency | 0-10 | Brand colors, fonts, spacing uniform |
+| Whitespace | 0-10 | Meets mode minimum (50%/45%/40%) |
+| Typography hierarchy | 0-10 | Clear title > subtitle > body > caption |
+| Legibility | 0-10 | Contrast >= WCAG AA, font >= minimum |
+| Decoration restraint | 0-10 | Max 2 CornerMarks/section, no clutter |
+
+`design_score = mean(consistency, whitespace, hierarchy, legibility, restraint)`
+
+### 5.3 Coherence Score (weight: 0.20)
+
+| Criterion | Score Range | Evaluation |
+|---|---|---|
+| Narrative flow | 0-10 | Logical progression slide-to-slide |
+| Sequence compliance | 0-10 | TITLE first, CLOSING last, breaks between sections |
+| Rhythm variety | 0-10 | No more than 3 consecutive same type |
+| Duration balance | 0-10 | No single slide > 120s, total matches target |
+
+`coherence_score = mean(flow, sequence, rhythm, duration)`
+
+### 5.4 GAD Score — Geometric Aesthetic Density (weight: 0.20)
+
+Measures spatial quality of slide layouts. Adapted from SlideGen's Geometry-Aware Density metric.
+
+**Note:** The original GAD Score in SlideGen is algorithmically computed from element bounding boxes and spatial coordinates. This adaptation converts it to LLM-judged criteria since rendered slides may not expose geometric metadata. For TSX renderer, a future enhancement could compute GAD programmatically from DOM element positions.
+
+| Criterion | Score Range | Evaluation |
+|---|---|---|
+| Element distribution | 0-10 | Elements use available space, not clustered in one area |
+| Alignment consistency | 0-10 | Elements follow a grid, not arbitrarily placed |
+| Density balance | 0-10 | Neither too sparse (empty) nor too dense (cluttered) |
+| Margin compliance | 0-10 | Content within safe margins per mode |
+
+`gad_score = mean(distribution, alignment, density, margins)`
+
+### 5.5 Composite Score
+
+```
+composite = (content_score * 0.30)
+          + (design_score  * 0.30)
+          + (coherence_score * 0.20)
+          + (gad_score * 0.20)
+```
+
+| Composite | Verdict | Action |
+|---|---|---|
+| >= 7.0 | PASS | Proceed to delivery |
+| 5.0 - 6.9 | REVISE | Return to Step 3 with dimensional feedback |
+| < 5.0 | FAIL | Return to Step 1, reassess approach |
+
+### 5.6 Rendered Output Capture
+
+The evaluator needs access to the rendered output, not just the JSON. Capture method per renderer:
+
+| Renderer | Capture Method |
+|---|---|
+| TSX Web | Screenshot each slide via Playwright/Puppeteer (1920×1080 viewport) |
+| PPTX | Convert each slide to PNG via `python-pptx` thumbnail or LibreOffice export |
+| Markdown | Render via `slidev export --format png` or `marp --images png` |
+
+For LLM-based evaluation, provide screenshots alongside the JSON. For programmatic evaluation (future), parse DOM/PPTX element positions directly.
+
+### 5.7 Evaluation Prompt Template
+
+```
+Evaluate this slide deck against 4 dimensions. Score each criterion 0-10.
+
+PRESENTATION JSON: {{canonical_json}}
+RENDERED OUTPUT: {{slide_screenshots_or_descriptions}}
+SOURCE MATERIAL: {{original_briefing}}
+
+DIMENSIONS:
+1. CONTENT (factual accuracy, relevance, completeness, word economy)
+2. DESIGN (consistency, whitespace, hierarchy, legibility, restraint)
+3. COHERENCE (narrative flow, sequence compliance, rhythm, duration)
+4. GAD (element distribution, alignment, density balance, margin compliance)
+
+For each dimension, provide:
+- Individual criterion scores (0-10)
+- Dimension average
+- Specific feedback for scores < 7
+
+OUTPUT FORMAT:
+{
+  "content": { "factual": N, "relevance": N, "completeness": N, "word_economy": N, "average": N, "feedback": "..." },
+  "design": { "consistency": N, "whitespace": N, "hierarchy": N, "legibility": N, "restraint": N, "average": N, "feedback": "..." },
+  "coherence": { "flow": N, "sequence": N, "rhythm": N, "duration": N, "average": N, "feedback": "..." },
+  "gad": { "distribution": N, "alignment": N, "density": N, "margins": N, "average": N, "feedback": "..." },
+  "composite": N,
+  "verdict": "PASS|REVISE|FAIL"
+}
+```
+
+---
+
+## 6. VALIDATION CHECKLIST
+
+### Per Renderer
+
+#### TSX Web
+
+- [ ] Every text element has `style={FONT_SANS}` or `style={FONT_MONO}`
+- [ ] `SlideLayout` wraps every slide
+- [ ] Contrast ratios pass WCAG AA (4.5:1 body, 3:1 large text)
+- [ ] Animations stripped in async mode
+- [ ] `prefers-reduced-motion` media query present
+- [ ] No hardcoded px outside 1920x1080 canvas
+- [ ] `npm run lint && npm run typecheck` = 0 errors
+
+#### PPTX
+
+- [ ] Slide dimensions match aspect_ratio (EMU values correct)
+- [ ] Fonts embedded or system-available
+- [ ] Speaker notes populated for palco/live modes
+- [ ] Image resolution >= 96 DPI
+- [ ] No overlapping elements
+- [ ] Text not clipped by shape boundaries
+- [ ] File opens correctly in PowerPoint, Keynote, and Google Slides
+
+#### Markdown (Slidev/Marp)
+
+- [ ] Slide separators (`---`) correct
+- [ ] Frontmatter valid YAML
+- [ ] Code fences use correct language identifiers
+- [ ] Images referenced with valid paths/URLs
+- [ ] Speaker notes in comment blocks
+- [ ] Renders correctly with `slidev` or `marp` CLI
+- [ ] Export to PDF produces correct page count
+
+### Cross-Renderer
+
+- [ ] All 12 slide types have a mapping in each target renderer
+- [ ] Slide count matches between JSON and rendered output
+- [ ] Speaker notes preserved across all formats
+- [ ] Brand colors applied consistently in all renderers
+
+---
+
+## 7. ANTI-PATTERNS
+
+### Pipeline Anti-Patterns
+
+| Anti-Pattern | Why It Fails | Correct Approach |
+|---|---|---|
+| Skip JSON schema, render directly from briefing | No interchange format = no multi-output, no evaluation | Always generate canonical JSON first (Step 3) |
+| Modify JSON per renderer | Schema drift, inconsistent content | One JSON, multiple renderers. Renderers adapt layout, not content |
+| Evaluate before rendering | Can't assess design/GAD without rendered output | Evaluate after rendering (Step 5) |
+| Use same word density for all modes | Palco slides unreadable, async slides empty | Respect mode constraints (Step 2 table) |
+| Generate PPTX by screenshot of TSX | Lossy, low resolution, no editable text | Each renderer reads JSON independently |
+| Hardcode brand in JSON schema | Schema becomes single-brand | Brand config is a reference or inline object, never embedded in schema definition |
+
+### Content Anti-Patterns
+
+| Anti-Pattern | Why It Fails | Correct Approach |
+|---|---|---|
+| Title describes the chart ("Revenue by Month") | Title should BE the insight | Title IS the conclusion ("Revenue Grew 40%") |
+| Same type 4+ times in a row | Kills rhythm, audience zones out | Max 3 consecutive, insert rhythm resets |
+| More than 15 words on palco/live slide | Illegible from distance, splits attention | Respect mode.max_words constraint |
+| Paragraph of text on any slide | Slides are not documents | Keywords and fragments only |
+| Missing speaker notes in palco/live | Narrator has no script | speaker_notes required for palco/live modes |
+| Self-explanatory slide in palco mode | Steals attention from narrator | Slide supports narrator, not replaces |
+
+### Rendering Anti-Patterns
+
+| Anti-Pattern | Renderer | Why It Fails |
+|---|---|---|
+| CSS animations in PPTX | PPTX | PPTX uses its own animation system |
+| `v-click` in Marp | Markdown | Marp has no progressive reveal; only Slidev supports `v-click` |
+| Stock photos | All | Destroys credibility |
+| Code as screenshot | All | Inaccessible, pixelates on resize |
+| Color-only data differentiation | All | Inaccessible to colorblind users |
+
+---
+
+## 8. COMPANION REFERENCES
+
+| SOP | Use For | Key Sections |
+|---|---|---|
+| SOP-SLIDES-001 | Human-readable process, decision criteria | Full workflow for designers/authors |
+| SOP-SLIDES-002 | AIOX-specific TSX code patterns | Section 2 (API), 3 (tokens), 6 (12 types) |
+| SOP-SLIDES-003 | Brand-agnostic TSX code patterns | Section 0 (brand config), 2 (shared module), 4 (12 types) |
+
+| Research | Use For | Key Data |
+|---|---|---|
+| `02-research-report.md` | Ecosystem overview, tool landscape | Sections 1 (libraries), 3.3 (architecture pattern) |
+| `03-recommendations.md` | Stack selection, MCP tools | Section 1.1 (MCP server), 2.3 (JSON schema) |
+| `04-pptagent-v2-domain-analysis.md` | PPTEval dimensions, business rules | Section 2 (taxonomy), 5 (evaluation) |
+| `05-competitive-analysis.md` | GAD Score, cross-project comparison | Section 5 (evaluation), 10 (diagnosis) |