/echo

Cognitive Persona Model (CPM)

A 6-dimension framework for deep persona simulation, grounded in peer-reviewed research on LLM-based persona modeling.

Academic Foundation

Core finding: Structured, multi-dimensional persona profiles produce significantly more faithful simulations than flat descriptions. The CPM operationalizes this finding for UX walkthroughs.

Six Dimensions

1. Beliefs

What the persona holds to be true about the world, technology, and themselves.

UX Walkthrough Impact: Beliefs determine what the persona expects to see. Mental model gaps (a core Echo analysis) originate here. A persona with self_efficacy: low blames themselves for errors; self_efficacy: high blames the interface.

Cross-dimension interactions:

• Beliefs → Goals: constrain which goals feel achievable
• Beliefs → Stance: shape default reactions to features (e.g., trust_disposition: skeptical → negative stance on data collection)

2. Goals

Hierarchical motivation structure extending JTBD.

UX Walkthrough Impact: Hierarchical goals reveal why a persona abandons or persists. Flat JTBD misses goal conflicts (e.g., wanting both privacy and personalization). Priority triggers simulate real decision points.

Cross-dimension interactions:

• Values → Goals: prioritize which goals matter most
• Goals → Emotions: unmet goals activate frustration; exceeded goals activate delight

3. Emotions

Persona-specific emotional profile extending Echo's emotion scoring.

UX Walkthrough Impact: baseline_mood offsets all emotion scores (a persona starting anxious scores lower). frustration_threshold determines when -2 becomes -3 (abandonment). recovery_pattern guides whether a good step after a bad one actually helps.

Cross-dimension interactions:

• Goals → Emotions: goal achievement/failure directly modulates emotional state
• Emotions → Communication Style: high arousal increases complaint intensity

4. Values

What the persona cares about beyond the immediate task.

UX Walkthrough Impact: non_negotiables create hard exit conditions — if violated, the persona leaves regardless of other scores. value_axes determine how tradeoffs are perceived (a privacy-leaning persona sees "personalized recommendations" as threatening, not helpful).

Cross-dimension interactions:

• Values → Goals: prioritize goal hierarchy
• Values → Stance: generate specific feature reactions

5. Stance

Pre-formed opinions about specific UX patterns and features.

UX Walkthrough Impact: Stance creates immediate reactions to features before the persona even uses them. A persona with chatbot: "waste of time" will dismiss a chatbot CTA with -1 before testing it. decision_style determines how long they spend on choice screens.

Cross-dimension interactions:

• Beliefs → Stance: beliefs shape default stances
• Values → Stance: values generate stances on value-loaded features
• Stance + risk_appetite → bias vulnerability (risk-averse personas are more susceptible to status quo bias)

6. Communication Style

How the persona expresses their experience during simulation.

UX Walkthrough Impact: Communication Style governs the SPEAK step output. A self-blame persona says "I'm probably doing this wrong" (-1) while a system-blame persona says "This is broken" (-2) for the same friction point. reference_frame provides comparison anchors for expectation gaps.

Cross-dimension interactions:

• Emotions → Communication Style: high emotional arousal increases expression intensity
• Beliefs.self_efficacy → Communication Style.complaint_pattern: low self-efficacy correlates with self-blame

Cross-Dimension Interaction Rules

Beliefs → Goals          # Beliefs constrain which goals feel achievable
Beliefs → Stance         # Beliefs shape default feature reactions
Values  → Goals          # Values prioritize goal hierarchy
Values  → Stance         # Values generate stances on value-loaded features
Goals   → Emotions       # Goal achievement/failure activates emotions
Emotions → Communication Style  # Emotional state modulates expression intensity
Beliefs.self_efficacy → Communication Style.complaint_pattern  # Self-efficacy drives blame attribution
Stance.risk_appetite → Bias Vulnerability  # Risk aversion increases susceptibility to status quo bias

Application during WALK step:

1. Load Beliefs → set expectations for what the persona thinks they'll see
2. Check Goals hierarchy → identify what matters most right now
3. At each interaction point: evaluate against Stance → generate initial reaction
4. Modulate reaction through Emotions → apply baseline_mood offset and reactivity
5. Check Values.non_negotiables → trigger hard exit if violated
6. Express through Communication Style → generate persona-voiced feedback

CPM Consistency Verification

Adapted from Eval4Sim's 3-axis evaluation framework. Run during the ANALYZE step when a Cognitive Profile was active.

Axis 1: Adherence

Did the simulation match the defined profile?

Axis 2: Consistency

Were dimensions internally coherent throughout the walkthrough?

Axis 3: Naturalness

Did the simulation feel like a real person, not a checklist?

Fidelity Score

Score = (items passed / 15) * 100%

Rating:
  >= 80%  → High Fidelity    (reliable simulation)
  50-79%  → Moderate Fidelity (usable with caveats)
  < 50%   → Low Fidelity      (profile needs revision)

Example CPM Profiles

The Newbie — Cognitive Profile

beliefs:
  technology_beliefs:
    - "Good apps don't need instructions"
    - "If I can't find it in 10 seconds, it's not there"
  self_efficacy: low
  trust_disposition: neutral
  mental_models:
    - "Navigation = top menu bar"
    - "Red = error, Green = success"
    - "X button = close/cancel"
  assumptions:
    - "There's a free version"
    - "I can undo anything"

goals:
  immediate_goals: ["Figure out what this app does"]
  journey_goals: ["Complete one successful task"]
  life_goals: ["Feel competent with technology"]
  goal_conflicts: ["Want to explore but afraid of breaking things"]
  goal_priority_triggers:
    - trigger: "Error message appears"
      shift: "life_goals.feel_competent dominates → considers abandoning"

emotions:
  baseline_mood: {V: 0.1, A: 0.3, D: -0.5}  # Slightly positive, slightly anxious, low control
  emotional_reactivity: high
  frustration_threshold: 3  # Abandons after 3 friction points
  recovery_pattern: "Needs hand-holding; one success resets frustration partially"
  delight_sensitivity: high  # Small wins feel big

values:
  value_axes:
    simplicity_vs_power: -0.9  # Strongly prefers simplicity
    speed_vs_thoroughness: -0.6  # Prefers quick results
  non_negotiables: ["No account required to browse", "Visible undo option"]
  willingness_to_pay: "Won't pay until value is proven through free use"
  effort_tolerance: low

stance:
  feature_stances:
    onboarding_tour: "helpful if short"
    advanced_settings: "scary, avoid"
    chatbot: "might try if stuck"
  ux_pattern_preferences:
    navigation: "simple top menu"
    forms: "one field at a time"
  risk_appetite: averse
  decision_style: delegating  # Looks for recommendations

communication_style:
  vocabulary_level: basic
  expression_style: passive
  complaint_pattern: self-blame  # "I'm probably doing something wrong"
  question_style: "Asks a friend or searches YouTube before reading docs"
  reference_frame: ["Their phone's built-in apps", "Google"]

The Skeptic — Cognitive Profile

beliefs:
  technology_beliefs:
    - "Companies always collect more data than they admit"
    - "Free products means I'm the product"
    - "Default settings are optimized for the company, not me"
  self_efficacy: high
  trust_disposition: skeptical
  mental_models:
    - "Privacy settings exist but are deliberately hard to find"
    - "Unsubscribe links sometimes don't work"
    - "Cookie banners are designed to trick you into accepting"
  assumptions:
    - "There are hidden fees"
    - "They'll sell my email"

goals:
  immediate_goals: ["Find the privacy policy", "Check what data is collected"]
  journey_goals: ["Determine if this service is trustworthy"]
  life_goals: ["Protect personal information", "Not get scammed"]
  goal_conflicts: ["Wants the service's benefits but doesn't trust it"]
  goal_priority_triggers:
    - trigger: "Requests phone number or credit card"
      shift: "life_goals.protect_information dominates → high abandonment risk"

emotions:
  baseline_mood: {V: -0.2, A: 0.4, D: 0.3}  # Slightly negative, alert, moderate control
  emotional_reactivity: medium
  frustration_threshold: 5  # Patient but principled
  recovery_pattern: "Transparency restores trust; one dark pattern erases all progress"
  delight_sensitivity: low  # Hard to impress

values:
  value_axes:
    privacy_vs_convenience: -0.9  # Strongly privacy-oriented
    transparency_vs_simplicity: -0.8  # Wants full disclosure
  non_negotiables:
    - "Clear privacy policy accessible from every page"
    - "No pre-checked consent boxes"
    - "Account deletion option visible"
  willingness_to_pay: "Prefers paid if it means no data monetization"
  effort_tolerance: high  # Will dig through settings

stance:
  feature_stances:
    social_login: "tracking vector — avoid"
    personalized_recommendations: "evidence of surveillance"
    cookie_banner: "judge entire site by this"
  ux_pattern_preferences:
    navigation: "doesn't matter if privacy controls are accessible"
    forms: "minimal fields only"
  risk_appetite: averse  # But for different reasons than Newbie
  decision_style: deliberate  # Reads everything

communication_style:
  vocabulary_level: advanced
  expression_style: blunt
  complaint_pattern: system-blame  # "This is a dark pattern"
  question_style: "Searches terms of service and privacy policy directly"
  reference_frame: ["Signal", "DuckDuckGo", "EU GDPR standards"]

The Power User — Cognitive Profile

beliefs:
  technology_beliefs:
    - "Every app should have keyboard shortcuts"
    - "Good software is fast software"
    - "If the UI is slow, the backend is worse"
  self_efficacy: high
  trust_disposition: neutral
  mental_models:
    - "Cmd+K = command palette"
    - "Settings should have a search function"
    - "Bulk operations should exist for any list"
  assumptions:
    - "API access is available"
    - "Data export in standard formats"
    - "Undo/Redo exists everywhere"

goals:
  immediate_goals: ["Complete the task in minimum clicks"]
  journey_goals: ["Set up workflow automation", "Customize to my preferences"]
  life_goals: ["Maximize productivity", "Master tools I use daily"]
  goal_conflicts: ["Wants feature density but also clean UI"]
  goal_priority_triggers:
    - trigger: "Task requires > 5 clicks"
      shift: "immediate_goals.minimum_clicks → searches for shortcut or automation"

emotions:
  baseline_mood: {V: 0.2, A: 0.5, D: 0.7}  # Slightly positive, alert, high control
  emotional_reactivity: medium
  frustration_threshold: 7  # Very patient with complex tools
  recovery_pattern: "Finds workaround independently; stays if power features exist"
  delight_sensitivity: medium  # Appreciates clever shortcuts

values:
  value_axes:
    power_vs_simplicity: 0.9  # Strongly prefers power
    speed_vs_thoroughness: 0.7  # Wants speed
    customization_vs_convention: 0.6  # Prefers customizable
  non_negotiables:
    - "Keyboard navigation works"
    - "No feature regression in updates"
    - "Bulk operations available"
  willingness_to_pay: "Pays for pro/power tier without hesitation if features justify it"
  effort_tolerance: high

stance:
  feature_stances:
    onboarding_tour: "skip immediately"
    advanced_settings: "first place I visit"
    keyboard_shortcuts: "essential — judge app by this"
    API: "must exist"
  ux_pattern_preferences:
    navigation: "command palette + keyboard shortcuts"
    search: "instant, fuzzy matching"
    forms: "bulk input, paste support"
  risk_appetite: seeking  # Tries beta features
  decision_style: impulsive  # Decides fast, iterates

communication_style:
  vocabulary_level: advanced
  expression_style: blunt
  complaint_pattern: system-blame  # "This is inefficient"
  question_style: "Reads docs, searches changelog, files feature requests"
  reference_frame: ["VS Code", "Linear", "Raycast", "Notion"]