Assessment Design

1. Design Principle — Output First

Every question exists because a specific output element needs the data. If a question doesn’t feed a visible output, it doesn’t belong.

FREE Tier Outputs → Data Needed

PAID Tier Outputs → Additional Data Needed

2. Assessment Flow

FREE Tier — Spark Profile (~7–8.5 min to result)

PAID Tier — Full Profile (~25–35 min)

3. Phase 0: AI Welcome Chat

Phase 0 combines conversational onboarding (Messages 1–4) with interactive mode profiling (Messages 5–6). It captures context, adoption level, mindset hint, AND full mode selection + depth — replacing both the old form fields and the old MODE1/MODE2 items.

Conversation Script (6–7 messages, ~2.5–3 min)

Mode Depth Cards (behavioral framing: “what do you DO?”)

User sees descriptive cards only — no numbers. Internal depth values are for scoring backend.

💬 Conversational

🎨 Creative

🔧 Builder

⚡ Orchestration

NLP Extraction Logic

Signals from Phase 0

Deferred Fields

4. Phase 2: Forced-Choice Item Bank (22 items)

Each item presents two positive statements. The user selects which resonates more. There is no wrong answer — the choice reveals which superpower is stronger. Items are tagged as Mindset (M) or Skill (S) to enable separate Mindset Index and Capability Index calculations.

Item Design Principles

1. Both options are positive. No “wrong” choice.
2. Each item contrasts two different dimensions.
3. Language is natural and work-contextual.
4. No tool names in items.
5. Mindset items = attitude/orientation. Skill items = capability/action.
6. Social desirability balance. Both options must sound equally “good” to a business professional. Validation required: present items to 20 people without context — if one option is chosen >65% as “more impressive,” rewrite it.

Perception Items (5)

Intelligence Items (4)

Knowledge Items (4)

Creation Items (4)

Integration Items (5)

Coverage Matrix

Balance note: Integration has the most appearances (11), Perception/Intelligence/Creation the fewest (8 each). Acceptable — Integration is the most cross-cutting dimension by design.

5. Phase 3: Anchoring, Agentic & Skill Items (12 items)

Dimension Anchoring Items (5 — Likert scale 1–5)

These items establish absolute levels for each dimension, solving the ipsative measurement trap. Rated 1–5 (“Not at all like me” to “Exactly like me”).

Dimension_L1_Final = (Ipsative_Rank_Normalized × 0.6) + (Anchor_Score_Normalized × 0.4)

Agentic Orientation (1 item — single select)

Skill Signal Items (3 — single select)

6. Layer 1 Scoring Model

Dimension Scoring

Each forced-choice item assigns +1 to the chosen dimension and 0 to the other. Raw dimension scores are the sum of all choices for that dimension across 22 items.

Dimension_Normalized = (Raw_Score / Max_Possible_Score) × 100

Mindset Index (for AI Potential)

Mindset_Index_L1 = average(
  Perception_Mindset_Items,   // AI Curiosity signals
  Intelligence_Mindset_Items, // Critical Trust signals
  Knowledge_Mindset_Items,    // Iterative Learning signals
  Creation_Mindset_Items,     // Creative Courage signals
  Integration_Mindset_Items   // Augmentation Vision signals
) × normalization_factor

AI Potential (Layer 1 estimate)

AI_Potential_L1 = min(100, weighted_average(
  Mindset_Index_L1 × 1.4,      // High mindset = high ceiling
  Peak_Dimension × 1.1,         // Strongest dimension → potential
  SKILL1_frequency × 10,        // Usage frequency as velocity proxy
  100 - |Mindset - Capability| × 0.5  // Gap = untapped potential
))

Archetype Assignment

Top 2 dimensions by normalized score → Archetype lookup:

7. Email Gate & Lead Capture

The Email Gate sits between Layer 1 scoring and result display — the moment of peak curiosity. The user has invested 6+ minutes, scoring is complete, and they’re about to see their profile. This is the optimal point for email capture.

UX Copy & Design

What Happens on Submit

1. Email validated (format + disposable email detection)
2. Spark Profile displayed immediately — zero delay
3. Welcome email sent within 60 seconds (profile card, top 3 superpowers, 3 growth tips, CTA to Full Profile)
4. Lead captured in CRM with tags: archetype, top superpower, industry, role, company type, AI adoption level

Email Follow-Up Sequence

Expected Metrics

8. Phase 4: Mini AI Chat (FREE Tier)

2–3 personalized follow-up questions based on Layer 1 results. Gives users a taste of the AI conversation and creates the upgrade trigger.

Q1: Strongest Superpower Probe (always asked first)

Q2: Growth Area Hint (asked second)

Upgrade Wall

Curiosity Gap Principle (expert panel)

After each mini-chat response, the AI references something it noticed but can’t explore:

• After Q1: “The way you described that tells me something about your [second_superpower] that I’d need more time to unpack properly.”
• After Q2: “I can see a gap between what you know is possible and what you’re doing. That gap is actually your biggest asset — but I’d need to go deeper to show you why.”

9. Phase 5: Full AI Conversation (PAID Tier)

Conversation Architecture

5a: Experience Exploration

Opening:

“Let’s go deeper. Tell me about a recent project where AI played a significant role. Walk me through it — what was the task, what did you do, and how did it turn out?”

The AI probes the 2 weakest dimensions and 1 strongest (to confirm) using these probes:

5b: Scenario Challenges

Selection logic: Focus on weakest 2 dimensions, include domain-specific scenario, include 1 outside primary mode, escalate complexity.

Universal Scenario Bank (select 2–3)

5c: Reflection & Closing

Three final questions that capture mindset signals:

1. Forward-looking: “Looking ahead 12 months — how do you expect your work with AI to change?”
2. Temperature check: “Gut feeling — how much of your work will involve AI two years from now?”
3. Identity: “If you had to describe your relationship with AI in one sentence?”

“Is there anything about how you use AI that we haven’t covered? Anything you’re particularly proud of, or struggling with?”

10. Layer 2 Scoring Model

Per-Response Scoring

Confidence Weighting

5 Observable Differentiators

Layer 2 Technical Architecture

// Single LLM call returns both:
{
  "conversation_response": "Natural language reply...",
  "scoring": {
    "sub_competencies": { "1.1_ai_curiosity": { "score": 3, "confidence": "high" } },
    "language_signals": ["exploration", "system_thinking"],
    "next_probe_priority": "knowledge"
  }
}

Observable Differentiator reliability: Score each differentiator at 3 separate points during conversation (not just once at end), average the 3 scores, calibrate with 50+ human-expert-scored transcripts before launch.

11. Combined Scoring (Layer 1 + Layer 2)

Sub-Competency Score

Sub_Competency_Final = (L1_Score × 0.30) + (L2_Score × 0.70)

Dimension Score

Dimension_Score =
  (Mindset_Sub × 0.30) +
  (Skill_Sub × 0.30) +
  (Additional_1 × 0.20) +
  (Additional_2 × 0.20)

Normalized to 0-100.

Superpowered Score

Superpowered_Score =
  (Mindset_Index × 0.35) +
  (Capability_Index × 0.30) +
  (Application_Index × 0.25) +
  (Technical_Depth × 0.10)

Range: 0-100

AI Potential (refined with Layer 2)

AI_Potential_raw = weighted_average(
  Mindset_Index × 1.4,
  Learning_Velocity × 1.2,
  Strongest_Dimension × 1.1,
  100 - Consistency_Gap × 0.5
)

// Sigmoid normalization to distribute across 40-100 range
AI_Potential_Final = sigmoid_map(AI_Potential_raw,
  pilot_mean, pilot_sd, target_range=[40, 100])

Self-Perception Gap

Consistency = 1 - |L1_normalized - L2_normalized| / 100

> 0.8: "Highly consistent self-awareness"
0.5-0.8: "Some gaps between self-perception and practice"
< 0.5: "Significant self-perception gap"

12. Data-to-Output Mapping

13. UX Design Principles

Progress & Pacing

• Show progress bar throughout (“Step 2 of 4” or percentage)
• Forced-choice: one per screen on mobile, 2–3 visible on desktop
• Estimated time remaining always visible
• Mode/Skill items feel like a break (different visual format)

Visual Design

• Clean, minimal interface — the assessment itself should feel “superpowered”
• Superpower colors: ■ Perception ■ Intelligence ■ Knowledge ■ Creation ■ Integration
• Forced-choice items: two cards side by side, tap to select
• Back button: Allowed within Phase 2 and Phase 3, not between phases. Phase 0 chat is forward-only. Going back is tracked as a signal
• Phase 0 → Phase 2 transition: Animated morph from chat bubbles to forced-choice cards (1–2 sec)
• Post-result → Phase 4 transition: Animated shift back to chat UI — already familiar from Phase 0

Tone

• Phase 0 welcome: Friendly, curious, human — “Hi! I’m here to map your AI superpowers.”
• Phase 0 → Phase 2: “Love it. Now let’s find out where your real superpowers are — this next part is quick.”
• Progress: “Great — you’re almost through the quick calibration”
• Results tease: “Your profile is ready. Let me show you what I found.”
• Phase 4 mini-chat opener: References Phase 0 context — “Earlier you mentioned [X]. That tells me something interesting…”

Mobile-First

• Full-width cards, large tap targets
• Chat: familiar messaging UI (bubbles, typing indicator)
• Results: swipeable card format
• Share: native share sheet integration

14. Anti-Gaming & Quality Controls

Time-Per-Item Tracking

If 22 items completed in <60 seconds (<3s per item), flag as “speed-through.” Don’t invalidate — apply confidence: low modifier to all L1 scores, increasing L2 weight.

Semantic Consistency Check

The Consistency Score (L1 vs L2 comparison) detects both self-perception gaps (coaching insight) and gaming (reliability concern):

• > 0.8: High confidence in all scores
• 0.5–0.8: Flag dimensions with biggest gaps for extra L2 probing
• < 0.5: Mark as “low confidence” — recommend re-assessment

No explicit attention checks (“select option B”). They break immersion and feel insulting to professionals.

Social Desirability Mitigation

1. Pre-launch validation: Present all 22 items to 20 people without context — if one option chosen >65%, rewrite it
2. Both-positive design: Both options must sound equally desirable
3. Ipsative format helps: Forced-choice is inherently more resistant to gaming than agree/disagree scales

15. Re-Assessment Strategy

Hybrid Item Rotation (60/40)

Expanded bank needed: 35 total items (13 permanent + 22 rotatable, drawing 9 per session). Target: Q3 2026.

16. Alternative Path — Conversation Upload Assessment

The Idea

Instead of answering questions about how you use AI, you show how you use AI. The user uploads 30–50 of their past AI conversations — from any tool (ChatGPT, Claude, Cursor, Copilot, Gemini, or any export) — and the system analyzes their actual behavior to generate a full SP™ AI Score profile.

This is a 100% behavioral assessment — no self-report, no forced-choice, no “how often do you…” questions. Just real evidence of how you work with AI.

Why This Path Exists

What the User Provides

Input: 30–50 past conversations with any AI tool.

Accepted formats:

• Cursor sessions — exported from session history (agent transcripts)
• ChatGPT exports — via Settings → Data Controls → Export Data (JSON)
• Claude exports — via conversation history export
• Plain text / Markdown — copy-pasted conversation logs
• Any AI tool — as long as the format shows user messages and AI responses

Minimum requirement: 30 conversations spanning at least 2 weeks. Recommended: 50+ conversations across 30+ days to avoid recency bias.

What the System Analyzes

1. AI Interaction Modes

2. Five Superpowers

3. AI Skills

4. Mindset — Observable Through Behavior

How the Output Looks

The output is a full SP™ AI Score profile — identical in structure to the Paid Full Profile:

1. SP™ AI Score (0–100) — composite score based on the four weighted components
2. Archetype — derived from top 2 superpowers
3. Narrative — 2–3 sentences describing who you are as an AI professional
4. Radar chart — 5 superpower dimensions
5. Mode profile — which modes you use and at what depth
6. Your Stack (Skills) — all 10 skills ranked and scored in human language
7. Behavioral Evidence — the strongest patterns observed, written as insights
8. What Would Make You Faster — 3–5 growth recommendations, each with a “Solve with AI” button that opens the prompt directly in Claude or ChatGPT

Scoring Methodology

Scoring follows the standard SP™ framework with one key difference: Layer 2 (behavioral evidence) carries 100% of the weight — there is no Layer 1 self-report.

SP™ AI Score = weighted_average(
  Mindset Index      × 35%   // observed via behavioral proxies
  Applied AI Skills  × 30%   // observed via tool usage and task types
  Domain Integration × 25%   // observed via workflow complexity
  Technical Depth    × 10%   // observed via builder-mode activity
)

Comparison to Standard Assessment

Target Audience

Ideal for:

• Power users who use AI daily and have rich conversation history
• Professionals who hate quizzes — “don’t ask me, look at what I do”
• Teams / enterprise — bulk upload from shared tools, no scheduling
• Developers and builders — Cursor/Copilot sessions contain the richest data
• Repeat assessment — upload latest 50 conversations quarterly

Not ideal for:

• AI beginners — fewer than 30 conversations won’t produce a meaningful profile
• Casual users — shallow conversations produce shallow profiles
• Privacy-sensitive users — requires sharing actual conversation content

Pilot Results

Tested with 50+ Cursor session transcripts (Jan–Feb 2026):

• 20 sessions → usable but recency-biased profile
• 50 sessions → stable, accurate, and rich profile
• Behavioral signals strong enough to score all 20 sub-competencies including mindset
• The “What Would Make You Faster” section had the highest perceived value

17. Open Questions

Resolved (from expert panel)

Still Open

1. Item validation (NON-NEGOTIABLE): 22 FC items + 5 anchoring items need validation with 50–100 pilot users. Required before public launch.
2. Phase 0 + Mini chat scoring weight: Phase 0 bonus signals (AI Adoption, Mindset Hint) are used for personalization and tiebreaking, NOT primary scoring. Mini chat (Phase 4) remains UX/conversion only. Revisit if pilot shows conversational responses are highly diagnostic.
3. Scenario bank size: Target 20 universal by Q2 2026, 35+ rotatable items by Q3 2026.
4. Layer 1 strategic accuracy: Business decision — make L1 directionally accurate (archetype correct 70%+) but numerically imprecise. Show archetype confidently; blur dimension scores/sub-competencies in free tier.
5. Chat technology: Start with text; add voice (ElevenLabs) as v2 feature.
6. Observable Differentiator calibration: 50+ human-expert-scored transcripts needed. Target: Cohen’s kappa > 0.7.
7. Enterprise path: Add “share with your team” prompt post-assessment.