Visual Analogy Creation: Cognitive Protocol

Visual analogies succeed when they map relational structure from a familiar domain onto an unfamiliar domain, using spatial relationships to make abstract patterns concrete.

Core Principle

Add elements when they reveal structure; remove them when they obscure it. The question is: "Does this complexity serve understanding?"

---

Part 1: The Diagnostic Phase

Step 1.1: Extract the Core Relational Structure

Identify the pattern of relationships, not the surface features.

Procedure:

1. Write down the concept/system you need to visualize
2. Ask: "What relates to what here?" List all relationships
3. Ask: "What affects what?" List all causal chains
4. Ask: "What changes together?" List all correlations
5. Identify the single most important relationship - this is your core structure

Example - "Customer loyalty programs":

• Relationships: Customer ↔ Company, Purchase ↔ Points, Points ↔ Rewards
• Causal chains: Purchase → Points accumulate → Rewards unlock → More purchases
• Changes together: Time invested ↔ Switching cost increases
• Core structure: Incremental investment creates increasing exit barriers

Why this works (Gentner's Structure-Mapping Theory): Analogies succeed by aligning systems of relations, not by matching surface attributes.

Step 1.2: Classify the Question Type (6W Framework)

Determine what KIND of understanding you're creating, which constrains the visual form.

Decision tree:

IF asking "what kind of thing is this?" or "how do these categories relate?"
  → WHO/WHAT (Qualitative/Portrait)
  → Use: Venn diagrams, attribute lists, categorical groupings

ELSE IF asking "how much/many?" or "compare quantities"
  → HOW MUCH (Quantitative/Charts)
  → Use: Bar charts, line graphs, proportional areas

ELSE IF asking "where is this in relation to other things?"
  → WHERE (Positional/Maps)
  → Use: 2x2 matrices, positioning maps, spatial layouts

ELSE IF asking "in what order?" or "when does this happen?"
  → WHEN (Temporal/Timelines)
  → Use: Sequential flows, timelines, stages

ELSE IF asking "how does this work?" or "what's the mechanism?"
  → HOW (Process/Flowcharts)
  → Use: Flow diagrams, cycle diagrams, system maps

ELSE IF asking "why does this happen?" or "what causes what?"
  → WHY (Causation/Multi-variable)
  → Use: Causal diagrams, feedback loops, correlation plots

Why this works (Tversky's Mind in Motion): Different cognitive tasks require different spatial structures.

Step 1.3: Identify the Image Schema

Find the pre-conceptual spatial pattern that matches your relational structure.

Universal image schemas:

CONTAINER - Inclusion/exclusion/boundaries

• Geometric: Circles, boxes, nested shapes
• Relations: Inside/outside, member/non-member, bounded/unbounded
• Example: "Users in our ecosystem" → Nested circles

PATH - Progression/journey/sequence

• Geometric: Lines with direction, connected segments
• Relations: Start/end, before/after, progress/regress
• Example: "Customer journey" → Path with waypoints

FORCE - Causation/influence/pressure

• Geometric: Arrows, vectors, pushing/pulling
• Relations: Cause/effect, strong/weak, resist/yield
• Example: "Market forces" → Opposing arrows

BALANCE - Trade-offs/equilibrium/reciprocity

• Geometric: Scales, see-saws, balanced arrangements
• Relations: Equal/unequal, stable/unstable, compensate
• Example: "Work-life balance" → Scale with weights

LINK - Connection/relationship/dependency

• Geometric: Lines connecting nodes
• Relations: Connected/disconnected, strong/weak ties, central/peripheral
• Example: "Social network" → Graph with nodes and edges

CYCLE - Repetition/feedback/recursion

• Geometric: Loops, circles, spirals
• Relations: Reinforce/balance, growth/decay, iterate
• Example: "Feedback loop" → Circular arrows

SCALE - Magnitude/hierarchy/levels

• Geometric: Vertical position, size differences, nested levels
• Relations: More/less, superior/subordinate, macro/micro
• Example: "Organizational hierarchy" → Pyramid or tree

Why this works (Lakoff & Johnson's Conceptual Metaphor Theory): Abstract reasoning is grounded in bodily experience of space.

---

Part 2: The Generation Phase

Step 2.1: Source Domain Selection via Structural Matching

Find a familiar domain that has the SAME relational structure, even if surface features differ completely.

Procedure:

1. List your target domain's relational properties:

   • From Step 1.1: "X relates to Y by [relationship type]"
   • From Step 1.3: "The geometric essence is [image schema]"

2. Generate source domain candidates:

   • Query: "What familiar things have this same pattern of relationships?"
   • Consider: Physical objects, natural phenomena, games, human relationships, machines, economics, biology
   • List 5-10 candidates

3. Test structural alignment: Create explicit mapping table:

   Target Element → Source Element
   Relation 1     → Equivalent Relation
   Relation 2     → Equivalent Relation
   

4. Evaluate mapping quality:

   • ✓ Do ALL key relations have equivalents?
   • ✓ Are the relations the SAME TYPE (causal→causal, hierarchical→hierarchical)?
   • ✓ Does the source domain add insight, not just restate?
   • ✓ Is the source domain more familiar than the target?

Why this works (Hofstadter & Sander's Analogy Theory): Analogy is about finding abstract sameness beneath surface differences. Quality depends on structural isomorphism.

Step 2.2: Refine Through Conceptual Blending

Create emergent meaning by selectively combining elements from source and target.

Procedure:

1. Create the mapping table:

   Target Domain | Source Domain | Blended Concept | Emergent Insight
   

2. Identify emergent structure:

   • What does the blend reveal that neither domain alone shows?
   • What new inferences does the mapping enable?

3. Test inferential productivity:

   • Can you make predictions in the target domain using source domain logic?
   • Do these predictions reveal non-obvious insights?

Why this works (Fauconnier & Turner's Conceptual Blending): Meaning emerges from the blend. Good analogies create insights that exist in neither source nor target alone.

Step 2.3: Determine SQVID Settings

Configure the visual style to match audience and purpose.

S - STRUCTURAL DETAIL LEVEL

What this controls: How much internal structure to reveal

IF concept's insight requires seeing internal mechanisms/interactions
  → ELABORATE (show subsystems, intermediate steps, feedback loops)
ELSE IF concept's insight is in overall pattern/relationship
  → SIMPLE (show high-level structure only)

Test: Does showing internal detail change the conclusion?
  YES → Elaborate reveals necessary structure
  NO → Simple reveals essential pattern

Q - QUALITY vs. QUANTITY

What this controls: Categorical or numerical information

IF insight is about characteristics/attributes/types/categories
  → QUALITY (labels, distinctions, feature descriptions)
ELSE IF insight is about amounts/measurements/comparisons/magnitudes
  → QUANTITY (numbers, scales, proportions, data encodings)

V - VISION vs. EXECUTION

What this controls: Temporal reference frame

IF showing ideal state OR future goal OR possibility space OR aspirational
  → VISION (how it should/could be, potential)
ELSE IF showing current state OR actual data OR constraints OR reality
  → EXECUTION (how it is, actuality)

I - INDIVIDUAL vs. COMPARISON

What this controls: Number of entities and focus

IF showing one entity's internal structure/behavior/properties
  → INDIVIDUAL (deep dive into single case)
ELSE IF showing relationships between entities OR contrasting alternatives
  → COMPARISON (multiple entities, relative positioning)

D - DELTA vs. STATUS QUO

What this controls: Temporal dynamics

IF emphasizing change/transformation/trend/evolution
  → DELTA (before/after, trajectories, rates of change)
ELSE IF emphasizing current configuration/snapshot/state
  → STATUS QUO (present moment, what exists now)

Why this works (Tufte's Visual Display Principles): Every visual encoding choice should match the type of information being conveyed.

---

Part 3: The Construction Phase

Step 3.1: Translate Relational Structure to Spatial Structure

Convert abstract relationships into concrete spatial relationships.

Spatial encoding rules:

1. Causation → Arrows

   • A causes B: Arrow from A to B
   • Mutual causation: Arrows both directions (cycle)
   • Strong/weak: Thick/thin arrows

2. Magnitude → Size or Position

   • More/greater: Larger size OR higher vertical position
   • Less/smaller: Smaller size OR lower vertical position

3. Time → Horizontal or Vertical Progression

   • Earlier → Left OR bottom
   • Later → Right OR top

4. Similarity → Proximity

   • Related concepts: Close together
   • Different concepts: Far apart

5. Hierarchy → Vertical Position or Nesting

   • Superordinate: Above OR containing
   • Subordinate: Below OR contained

6. Process Flow → Connected Path

   • Sequential: Linked chain
   • Parallel: Side-by-side paths
   • Branching: Path splits

7. Intensity/Degree → Visual Variable

   • More intense: Darker, thicker, larger
   • Less intense: Lighter, thinner, smaller

Why this works (Tversky's Cognitive Design Principles): When spatial structure matches conceptual structure, comprehension is immediate.

Step 3.2: Apply Progressive Structure Revelation

Build from core structure outward, testing at each layer whether additional structure adds insight.

Construction procedure:

Layer 0 - Core Relationship Hypothesis State in one sentence what relationship/pattern/mechanism you're revealing. This is your north star.

Layer 1 - Essential Structure

1. Identify minimum elements needed to show the relationship
2. Draw the relationship using appropriate spatial encoding
3. STOP. Test: Does this make the stated relationship visible?
   • If YES → Proceed to Layer 2
   • If NO → Revise core relationship or element selection

Layer 2 - Structural Context

1. Identify what modulates/constrains/amplifies the core relationship
2. Add these elements and their relationships
3. STOP. Test: Does this reveal how the system behaves under different conditions?
   • If YES → Context adds genuine insight
   • If NO → Remove context, it's not structural

Layer 3 - Structural Detail

1. Identify whether understanding requires seeing HOW the relationships work internally
2. Add mechanism details ONLY if they change conclusions
3. STOP. Test: Does this detail reveal non-obvious behavior?
   • If YES → Keep, you're showing necessary complexity
   • If NO → Remove, it's decorative

Layer 4 - Exceptional Structure

1. Add ONLY if interesting behavior happens at extremes or boundaries
2. Show what breaks, saturates, or reverses
3. Test: Does this explain real-world observations that normal case doesn't?

Decision rule: Include layers 0-1 always. Include 2 if behavior is context-dependent. Include 3-4 only if explaining observed phenomena requires them.

Key insight: Each layer should add genuine structural insight, not decorative detail.

Why this works (Tufte's Data-Ink Ratio): Maximize the proportion of ink devoted to data.

Step 3.3: Validate Through Mental Simulation

Test if someone can "run" your visual like a simulation in their mind.

Validation procedure:

1. Trace test: Can you follow a path through the visual?

   • Start at beginning, follow connections, reach conclusion or cycle

2. Prediction test: Can you predict what happens next?

   • Given initial state, following the relationships, infer next state

3. Manipulation test: Can you mentally change one variable and see effects?

   • If X increases, what happens to Y?

4. Explanation test: Can someone unfamiliar explain it back?

   • If explanation ≠ intended insight, visual fails

Why this works (Hegarty's Mental Animation Theory): Effective diagrams support mental simulation.

---

Part 4: Refinement Protocols

Anti-Pattern Detection

Run this checklist on every visual analogy:

□ DECORATION TRAP For each visual element: "Does this encode information about relationships?"

• If NO → REMOVE IT
• Note: Styling that encodes information (color for category, depth for hierarchy) is NOT decoration

□ STRUCTURAL MISMATCH Does the analogy break down at boundaries or in important cases?

• Test: Push to extremes. Does it still work?
• Test: Apply to relevant edge cases. Does it break?

□ SURFACE SIMILARITY TRAP Are you matching on appearance rather than relationships?

• Create explicit mapping table from Step 2.1
• Verify RELATIONSHIP TYPE matches (causal→causal, part-whole→part-whole)

□ MULTIPLE INSIGHTS Can you state the insight in one sentence?

• If NO → Create multiple visuals, each with single clear insight
• Note: Complex visuals showing ONE multilayered structure are acceptable

□ PRECISION THEATER Are you showing false precision?

• Test: Is the data/relationship actually that precise?
• If NO → Use approximation that honestly reflects uncertainty

□ COGNITIVE LOAD Does the visual require excessive working memory?

• Test: How many distinct relationships must be held simultaneously?
• If >7 → Consider breaking into multiple linked visuals or hierarchical organization

Structural Validity Test

TEST 1: Structural Completeness Does the visual show all relationships necessary for the insight?

TEST 2: Mapping Fidelity For analogies: Does every mapped relationship preserve its type?

TEST 3: Cognitive Simulation Can someone mentally "run" the system using your visual?

TEST 4: Insight Necessity Does the visual reveal something non-obvious?

TEST 5: Boundary Testing Does the analogy break at extremes you care about?

TEST 6: Emergent Insight Check Does the mapping reveal implications you didn't initially consider?

TEST 7: Clarity Under Modification If you change one element, do the consequences propagate correctly?

---

Quick Reference

Image Schema Quick Match

YOUR CONCEPT INVOLVES...     → USE THIS SCHEMA → VISUAL FORM

Categories/membership        → CONTAINER       → Circles, Venn diagrams
Sequence/journey             → PATH            → Linear flow, timeline
Influence/causation          → FORCE           → Arrows, vectors
Trade-offs/reciprocity       → BALANCE         → Scales, see-saw
Relationships/links          → LINK            → Network graph
Repetition/feedback          → CYCLE           → Loops, circular arrows
Magnitude/hierarchy          → SCALE           → Vertical position, size

6W Quick Decision

QUESTION TYPE → VISUAL FORM     → SHOWS

Who/What?     → Portrait        → Categories, attributes
How much?     → Chart           → Quantities, comparisons
Where?        → Map             → Position, relationships
When?         → Timeline        → Sequence, duration
How?          → Flowchart       → Process, mechanism
Why?          → Causal diagram  → Causation, correlation

Spatial Encoding Rules

RELATIONSHIP → ENCODE AS

Causation    → Arrow
Magnitude    → Size or vertical position
Time         → Horizontal axis or sequence
Similarity   → Proximity
Hierarchy    → Nesting or vertical stack
Flow         → Connected path
Intensity    → Darkness, thickness, size

---

Summary: The Complete Protocol

Phase 1: Diagnose (What am I showing?)

1. Extract core relational structure
2. Classify question type (6W)
3. Identify image schema

Phase 2: Generate (What's it like?) 4. Find source domain with structural match 5. Create explicit mapping table 6. Test for emergent insights (blending) 7. Set SQVID parameters

Phase 3: Construct (How do I show it?) 8. Translate relations to spatial structure 9. Build progressive layers (0-4) 10. Validate through mental simulation

Phase 4: Refine (What can I remove?) 11. Run anti-pattern checks 12. Apply Structural Validity Tests 13. Iterate until structural clarity achieved

Throughout: Remember the Why

• Gentner: Match relational structure, not surface features
• Lakoff & Johnson: Ground abstract concepts in spatial experience
• Tversky: Spatial relationships carry conceptual relationships
• Tufte: Maximize information, minimize ink
• Hofstadter: Analogy creates insight through novel mapping

The Meta-Principle: Visual analogies succeed by making the invisible visible through structural correspondence. Every element must earn its place by revealing structure.

---

Part 5: Animated Visual Storytelling

When the medium is video or interactive composition (not a static image), time becomes a structural dimension. Animation controls attention, reveals relationships sequentially, and creates emotional response through pacing.

5.1: Narrative Arc for Technical Explainers

Every animated visualization needs a story structure with clear beats. Dead time kills engagement.

The 5-beat arc:

1. Hero moment (0-15% of duration): The subject fills the frame. Close camera. The viewer understands WHAT they're looking at before anything happens. Never start zoomed out — earn the wide shot.

2. The question (15-25%): Pose the problem or choice. Text overlay: "How do you render it?" This creates tension that the rest of the animation resolves.

3. The mechanism (25-55%): Show HOW the system works. This is where the structural analogy plays out spatially. Side-by-side comparisons are powerful here — the viewer's eye races between the two approaches.

4. The punchline (55-70%): The payoff. The parallel side finishes while the sequential side is still going. The "4× faster" moment. This should arrive when the viewer is primed but not bored.

5. The hold (70-100%): Let the conclusion breathe. Camera orbits toward the winner. But don't hold too long — 3-4 seconds max, not 30% of the video.

Common timing mistake: The punchline arrives too late. If the video is 14 seconds, the punchline should hit by second 10, not second 13.

5.2: Camera as Attention Director

Camera position is not neutral — it IS the viewer's focus.

Camera rules:

• Start close, earn the wide shot: Begin with the subject filling the frame. Pull back only when you need to reveal context (new elements, comparison). The pull-back itself is a narrative beat.

• Pull back BEFORE new elements appear: If you're going to add worker nodes below the timeline, pull the camera back first so there's room. Never let elements appear off-screen.

• Orbit toward the winner: In a comparison, the camera should drift toward the winning side in the final act. The loser recedes. The viewer's perspective literally shifts to favor the answer.

• Snap zoom for emphasis: When the key metric appears ("4× faster"), a quick 200-300ms zoom snap toward it, then ease back. This is the visual equivalent of bold text.

• Subtle shake during activity: Very light camera vibration (0.01-0.02 units) during processing phases makes the scene feel alive. Too much is nauseating.

5.3: Comparison Visualization (The Race Pattern)

When showing "our way vs the old way," a side-by-side race is far more compelling than sequential before/after.

The race pattern:

1. Start unified: Show one object (the composition, the data, the process)
2. Dramatic split: A laser cut, a flash, a duplication event. This is a story beat, not a quiet transition.
3. Label immediately: "Traditional" (dim) vs "Editframe" (bright). Use brand names, not abstract CS terms.
4. Race simultaneously: Both sides process at the same time. The viewer's eye bounces between them. Progress indicators make the speed difference visible.
5. Winner finishes first: The fast side completes while the slow side is visibly behind. Hold this moment — it's the proof.

Color coding emotion:

• Slow/old/bad: warm tones (orange, amber), dimmer, lower opacity
• Fast/new/good: cool tones (blue, teal), brighter, higher emissive

5.4: Concrete > Abstract

Every element must represent something the viewer recognizes.

The "box + label" trap: The default failure mode is abstract boxes with labels ("SERIALIZE", "INITIALIZE", "F₁"). These communicate nothing — they're just a flowchart with shadows.

Better: Show a timeline that looks like a real editor timeline (multi-track, varied clip lengths). Show worker nodes that pulse and glow during processing. Show progress bars that fill at different rates. Show particles streaming through active nodes. The viewer should understand what's happening even without the text labels.

Metrics close the sale: "4× faster" is concrete. "Parallel processing" is abstract. Always end with a measurable claim.

5.5: Sizing for the Medium

Objects need to be significantly larger than you think for rendered video output.

Rules of thumb:

• If an element is smaller than 5% of the frame width, it's invisible in the rendered output
• Progress bars need to be at least 3-4% of frame height to be visible
• Text overlays need 50% larger font than you'd use in a web page (the video has lower effective resolution)
• Particle effects need to be 2-3× larger than looks good in live preview (they compress poorly)
• Test by rendering and watching the output, not by looking at the live preview

5.6: 3D as a Structural Tool

3D perspective adds depth hierarchy and professional polish, but it must serve the structure.

When 3D helps: Showing parallelism (side-by-side lanes), hierarchy (above/below), progression (near/far from camera), emphasis (closer = more important)

When 3D hurts: When the concept is inherently 2D (flowcharts, timelines). When the perspective distortion obscures the comparison. When it's purely decorative.

3D lighting communicates: Brighter = active/important. Shadows = grounding/reality. Emissive glow = energy/processing. Specular highlights = quality/polish.

---

Additional Resources

For detailed worked examples (technical debt, network effects, compound interest, etc.) and domain-specific pattern libraries, see examples.md.

For advanced techniques (progressive disclosure, constraint highlighting, diachronic comparison) and common failure modes with corrections, see advanced.md.