Phase 2: Rhetorical Architecture — Detail

The Joint Selection Process

This phase is a conversation, not a monologue. Use AskUserQuestion for each instrument selection. One decision per turn. Never combine multiple decisions into a single message — see the interaction-rules steering rule.

For each decision:

1. Extract the options from the vault summary (sections 2-13) and speaker profile (instrument_catalog). The vault is the living source — new instruments appear as more talks are parsed.
2. Present the options via AskUserQuestion with brief descriptions
3. Recommend based on the spec — put the recommended option first with "(Recommended)"
4. Wait for the author's choice before moving to the next decision

Mode Selection Logic

Read presentation_modes[] from the speaker profile. Each mode has a when_to_use field — use these to build a selection logic table dynamically. Present the modes with their descriptions and match signals from the spec.

Opening Pattern Selection Logic

Read instrument_catalog.opening_patterns[] from the speaker profile. Each pattern has a best_for field. Match to the spec's audience warmth, venue size, and context.

Narrative Arc Templates

Read instrument_catalog.narrative_structures[] from the speaker profile. Each has acts and time_allocation. Present the options with their time splits and best-for context.

Decision #10: Pattern Strategy

Read patterns/_index.md for the full taxonomy and profile → pattern_profile for the speaker's pattern history.

Present patterns in 4 tiers:

PATTERN STRATEGY for "{talk title}"
===================================
YOUR TOOLKIT (signature):
  ✓ Narrative Arc (22/24 talks) — recommended for this format
  ✓ Bookends (18/24) — strong with this audience
  ✓ Expansion Joints (20/24) — essential for 45→20 min adaptation

WORTH CONSIDERING (contextual):
  ○ Talklet (3/24) — good fit for the 20-min constraint
  ○ Foreshadowing (7/24) — pairs well with your arc style

NEW TO YOU:
  ★ [NEW] Preroll — display bio/topic on screen before you start
  ★ [NEW] Seeding the First Question — plant an easy Q for Q&A

SHAKE IT UP:
  ⚡ [WILD CARD] Red, Yellow, Green — audience voting with colored cards
  ⚡ [WILD CARD] Cave Painting — one giant canvas instead of slides

WARNINGS:
  ⚠ Shortchanged (8/24 detections) — plan cut lines for the 20-min slot
  ⚠ Dual-Headed Monster — co-presented talk, define handoff points
===================================

Tier logic:

1. Signature — mastery_level: signature patterns (80%+ usage), always shown
2. Contextual — patterns matching spec context that speaker uses occasionally (10-80%)
3. New to You — from never_used_patterns, filtered by spec relevance, marked [NEW]
4. Shake It Up — 1-2 random picks from never_used_patterns, NOT filtered by relevance. Provocations, not prescriptions.

Antipattern warnings — merge speaker's recurring antipatterns (from pattern_profile.antipattern_frequency) + contextual warnings derived from the spec (co-presented → Dual-Headed Monster, dense content → Bullet-Riddled Corpse, new format → Shortchanged, etc.)

Summary-only mode (no profile yet): Pattern taxonomy still works — patterns come from the reference files alone (no usage stats). All patterns presented as "new" (no tier separation, just a flat relevant-patterns list). Contextual antipattern warnings still apply.

Enhance decisions 2-9 with pattern cross-references as shared vocabulary: when recommending an opening pattern, reference the taxonomy ID; when selecting a narrative structure, note which Presentation Patterns it maps to (e.g., "problem-solution" = Narrative Arc + Triad).

Decision #11: Illustration Strategy (when applicable)

Not every talk needs generated illustrations — demo-heavy, data-heavy, or screenshot-driven talks may not. When the author wants AI-generated illustrations, this sub-decision walks through the visual identity collaboratively.

Step 1: Propose style ideas with sample prompts

Present 3-4 style options informed by three sources:

1. The talk's own concepts, metaphors, and narrative — the style should reinforce the thesis, not be decorative wallpaper
2. The vault's visual history — read speaker-profile.json → visual_style_history for the structured data: default_illustration_style, style_departures[] (what styles the speaker has used and what triggered them), mode_visual_profiles[] (which modes tend toward which aesthetics), and confirmed_visual_intents[] (hard rules about visual design). Also read rhetoric-style-summary.md (Section 13 cross-talk visual patterns), slide-design-spec.md, and design_rules. Know what the speaker's default looks like so you can propose informed departures
3. Historical precedent for this mode/context — read visual_style_history → mode_visual_profiles for the matching mode ID. If the vault shows the speaker uses a particular aesthetic for this talk type, surface that as a data point (e.g., "your vault shows you use terminal aesthetic for agent talks"). If this talk's mode/context has no visual precedent in style_departures, say so

Each option includes: a name, why it fits this talk's concepts, how it relates to the speaker's visual history (continuation vs. departure), and a sample prompt excerpt showing a specific slide from THIS talk rendered in the style.

ILLUSTRATION STYLE OPTIONS for "{talk title}"
=========================================================

A. [STYLE NAME]
   CONCEPT FIT: [Why this style reinforces the talk's thesis,
   metaphors, and narrative arc — not just what it looks like]

   VAULT CONTEXT: [How this relates to the speaker's visual
   history — continuation of default, intentional departure,
   or precedent from similar talk types]

   Sample prompt (Slide N — [slide title]):
   "[Complete prompt showing this specific slide rendered
   in the proposed style]"

B. [STYLE NAME]
   CONCEPT FIT: [...]
   VAULT CONTEXT: [...]
   Sample prompt (Slide N — [slide title]):
   "[...]"

C. [STYLE NAME]
   ...

RECOMMENDATION: [Which option and why — grounded in concept
fit and vault context, not just aesthetic preference]
=========================================================

The key: each style option explains WHY it fits this specific talk's concepts, not just what it looks like. The author picks one (or mixes elements), then they iterate on the anchor paragraph together.

Step 2: Define format vocabulary & aspect ratios

Once the style is chosen, define the slide format types for this talk:

SLIDE FORMAT VOCABULARY
========================
FULL     — full-bleed illustration, 1-2 sentences overlaid
           → Landscape 16:9 (1920×1080)
IMG+TXT  — illustration ~60% of slide, text beside/below
           → Portrait 2:3 (1024×1536)
EXCEPTION — real photo, data table, bio, or primary source
           → No generated illustration; uses [IMAGE NN] placeholder
========================

Format names and ratios are talk-specific — the author may use different names or add formats (e.g., DIAGRAM for technical slides, QUOTE for attributed quotations).

Step 3: Choose image generation model

Agree on the target model (affects prompt style and capabilities):

• Model name and API (e.g., gemini-3-pro-image-preview, dall-e-3, flux)
• Any model-specific prompt conventions to bake into the style anchor
• Use generate-illustrations.py --compare N to generate the same prompt across multiple models for visual comparison (see phase5-slides.md Image Generation Setup)

Step 4: Visual continuity devices

Define recurring elements that tie the deck together as a coherent visual artifact:

• Sequential numbering (e.g., "FIG. N" numbering) — ties the deck together as one coherent document. The generation model may render numbers imperfectly; that's acceptable
• Recurring characters/motifs in consistent style (same uniforms, same species, same rendering approach across all appearances)
• Checklist progression — a shared base image that gets edited to add checkmarks, fill-ins, or stamps across the talk. Use image editing (not regeneration) to preserve visual consistency. Track which slide is the base image
• Progressive visual elements with explicit base-image tracking: document which slide is the "source" image for each progression, so edits chain correctly
• Annotation style (callout labels, footnotes, stamps) — keep labels funny/deadpan if that's the tone; the gags in labels ARE the point

Gate: Author approves the style anchor paragraphs, format vocabulary, and model choice. These become the Illustration Style Anchor section in the outline header.

Slide Budget Calculation

Read guardrail_sources.slide_budgets[] from the speaker profile. Match the spec's duration to the closest budget entry. Read pacing for WPM and slides/min targets.