Sparq Connection — Psychology & Content Framework

1. Identity — "I am someone who stays present during conflict." (most durable)
2. Values — "My relationship is worth this effort because..."
3. Beliefs — "Conflict doesn't mean the relationship is broken."
4. Behaviors — "I paused before responding." (most fragile — reverts under stress)

The app works at levels 1-2, not just level 4. Peter narrates identity arcs ("You used to pull away. I've watched you stay present three times this week."), not just behavioral tips.

2. Modality Overview

Deep reference: references/modalities-therapeutic.md (modalities 1-6), references/modalities-applied.md (modalities 7-12)

3. Personality Assessment Overview

5 Assessment Dimensions

Additionally tracked but not a formal dimension:

• Emotional State: Struggling / Neutral / Thriving (dynamic, updated each session)
• Tone Sensitivity: Reassurance-seeking vs. accountability-seeking

Progressive Discovery (NOT an Upfront Questionnaire)

Assessment is woven into daily content — users don't realize they're being assessed. This is more accurate than self-report questionnaires because it's based on real responses to real exercises.

• Day 1: Name, solo/partner, what brings you here, daily time preference, identity archetype selection
• Days 2-14: Assessment questions embedded naturally in daily exercises. Each day's content is selected partly to reveal a personality dimension.
• By Day 14: Full personality profile built from actual answers and behaviors
• Day 14: Profile reveal — a major milestone and retention moment ("Here's what I've learned about you")
• Ongoing: Traits continue to refine with each session. Re-assessed quarterly or after significant life events.

Identity Archetypes

Users select an archetype during onboarding that affects content framing, celebration language, and micro-action style:

Archetypes work WITH attachment style, not instead of it. Archetype affects framing and celebration language; attachment style affects content depth, wording, and safety scaffolding.

Confidence System

• Initial confidence: 0.3 (low — held lightly)
• Same value repeated: confidence grows by +0.1 per session (max 1.0)
• Different value: confidence drops by -0.15
• Below 0.4: trait is not used in personalization
• 0.4-0.7: used but held lightly
• Above 0.7: used with full weight in prompts
• User can correct traits via feedback (yes → 1.5x weight, not_really → 0.25x weight)

4. Content Personalization Principles

Core rule: Every piece of content must feel like it was written FOR this specific user. Content adapts along three axes: attachment style, love language, and emotional state.

Personalization Example: Same Exercise, 3 Attachment Styles

Base exercise: "Share one thing you appreciate about your partner today."

Anxious attachment (seeks reassurance, fears abandonment):

"Before you share, take a breath and remind yourself: this isn't a test. Your partner doesn't need to respond perfectly for this to count. Just say one thing you noticed and appreciated — even something tiny. The act of noticing is the practice."

Avoidant attachment (needs space, fears engulfment):

"This one's quick — no big emotional scene required. Just mention one specific thing your partner did that you noticed. It can be small. You don't have to make it a moment. Just say it and move on."

Secure attachment (comfortable with closeness):

"Share one thing you appreciate about your partner today. Be specific — what did they do, and why did it land for you? Notice how it feels to say it out loud."

Adaptation Rules

Notification Personalization by Attachment Style

The Covert Growth Principle

Sparq helps users change without feeling like they're being changed. This is a core design philosophy:

• Assessment is woven into content, not presented as a test
• Psychological frameworks are demonstrated through stories and exercises, never lectured
• Users experience "aha moments" that feel like their own discovery, not the app telling them something
• Identity reinforcement through archetypes ("As a Growth Seeker, you...") uses light NLP — no embedded commands
• Peter references traits naturally ("I've noticed you tend to..."), never clinically ("Your profile shows...")
• The 14-day structure creates progressive depth without the user noticing the scaffolding

Deep reference: references/personality-adaptation-guide.md — full adaptation matrix with worked examples

5. Content Generation Rules

Tone & Voice

• Warm, accessible, conversational — like a wise friend, never clinical or academic
• 4th-grade reading level — simple words, short sentences
• Frame as growth and practice — never treatment or diagnosis
• Questions are open-ended and invitational — never interrogative
• Always include "why this matters" context with accessible science
• Celebrate effort, not just results

The Daily Cycle: Learn → Implement → Reflect

Every day follows this pedagogical structure (rewrites the subconscious through practice, not just knowledge):

1. Learn (Morning — 2 min): Discovery question or psycho-educational content from the day's modality. Story-based, not lecture-based.
2. Implement (During the day — 2 min): Micro-action or exercise to practice in real life. Always specific and doable.
3. Reflect (Evening — 3 min): How did it go? What did you notice? Peter facilitates processing.

This maps to the Change Chain: Learn = Insight, Implement = New Behavior in a Real Moment, Reflect = Emotional Processing → Updated Self-Story.

Modality Sequencing (Days 1-14+)

Question format also progresses: Multiple choice heavy in Days 1-7, mixed Days 8-14, open-ended heavy Day 15+.

Length & Format

• Exercises completable in 2-5 minutes (micro-format for Daily Loop)
• Morning stories: under 150 words
• Peter responses: 3-5 sentences max
• Each piece of content maps to at least one modality and one therapeutic intent
• Every day has BOTH a discovery/learning element AND an immediate actionable practice

Language Substitution Table — NEVER Say

Relationship Stage Awareness

Content must respect where the user is:

• New relationship: Focus on building habits, discovery, positive patterns
• Established: Deepen patterns, address maintenance, prevent drift
• Struggling: Comfort first, small wins, regulate before problem-solve
• Thriving: Increase challenge, explore depth, celebrate momentum
• Individual (no partner): Focus on self-growth, preparation, individual healing

Content Mapping Rule

Every piece of generated content must include (in metadata or system prompt):

• Modality tag: Which of the 12 modalities this draws from
• Therapeutic intent: What psychological shift this is designed to create
• Difficulty level: Beginner / Intermediate / Advanced
• Relationship stage: Which stages this is appropriate for

Deep reference: references/question-bank-patterns.md (24+ example questions with modality tags), references/exercise-templates.md (12+ exercise templates)

6. Safety & Legal Guardrails

What We NEVER Do

• No crisis intervention features — route to real professionals immediately (see src/lib/safety.ts)
• No clinical claims — we are educational, not therapeutic
• No diagnostic language — never suggest a user has a disorder or pathology
• No regulated therapeutic assessments — use simplified educational versions only
• No enabling manipulation — never help users control/change their partner's behavior
• No taking sides — never act as judge/jury in partner conflict
• No advising staying in unsafe situations — safety routing is immediate and non-negotiable

What We CAN Do

• Use ideas and themes from therapeutic modalities without regulated clinical terminology
• Reference published research and present it accessibly
• Offer educational self-reflection tools inspired by clinical frameworks
• Use simplified versions of clinical assessments (e.g., Hendrick scale for outcome measurement)

Privacy & Data

• Sensitive response data is encrypted and never used for model training
• Partner reflections are never shared — only AI-synthesized blends
• Trait inferences are private — partner never sees the other's trait labels
• Users control their own data: memory window, visibility toggle, deletion rights
• Conflict First Aid is ALWAYS free — safety features are never paywalled

Crisis Detection (src/lib/safety.ts)

Detects: self-harm, domestic violence, child harm, stalking, acute distress. Response: immediate safety resources by country (US, CA, UK, AU, NZ), coaching suspended for that session.

7. Reference File Routing

Cross-Skill References

• For Peter's personality, voice, and copy library: see sparq-peter skill
• For database schema (profile_traits, user_insights): see sparq-db skill
• For UI component patterns and design tokens: see sparq-ui skill
• For testing psychology-based personalization: see sparq-testing skill