ATTRIBUTION MATRIX

BEHAVIORAL INTELLIGENCE SYSTEM

SYSTEM ONLINE
9/9 VALIDATIONS PASSED

TACTICAL ATTRIBUTION
ANALYSIS PLATFORM

Transform your behavioral data into actionable intelligence using rigorous mathematical models. Unlike black-box analytics, every calculation is transparent, defensible, and grounded in first principles.

Deploy advanced Markov-Shapley algorithms combined with Kelley's Covariation Model to extract behavioral intelligence from digital footprints.

THE PROBLEM WITH STANDARD ANALYTICS

✗ Last-Click Attribution

Ignores 90% of the customer journey. A user sees 5 touchpoints but only the last gets credit.

✗ Black-Box ML Models

Can't explain why. When the CFO asks "how did you calculate this?" you have no answer.

✗ Cloud-Dependent Tools

Your personal data travels to third-party servers. Privacy policies change. Data gets leaked.

PRECISION
99.9%
Attribution Accuracy
Validated against synthetic ground truth
SECURITY
LOCAL
Zero Cloud Storage
Your data never leaves your machine
SPEED
<60s
Analysis Runtime
Full Shapley enumeration included
TRANSPARENCY
100%
Explainable Results
Every calculation is auditable

HOW IT WORKS

1. UPLOADYour Data
2. PARSECanonicalize
3. MARKOVCausality
4. SHAPLEYFairness
5. BLENDHybrid α
6. REPORTIntelligence

DEPLOY TARGET FILE

Upload your data export to begin analysis

DEPLOY TARGET FILE

DROP FILE OR CLICK TO SELECT FROM SYSTEM

JSONCSVZIPTXT

SUPPORTED DATA SOURCES

Google Takeout • Facebook Export • Apple Archives • Browser History

OR TRY WITH SAMPLE DATA

SUPPORTED DATA SOURCES

G
Google Takeout
Chrome, YouTube, Search
f
Facebook Export
Activity, Ads, Pages
Apple Archives
Safari, Screen Time
CSV
Custom CSV
Any journey data

Accepts: JSON, CSV, ZIP, TXT • Max 50MB • All processing happens locally in your browser

THE THREE PILLARS

MARKOV-SHAPLEY

Hybrid Attribution

Combines Markov chain removal effects (causality) with Shapley value allocation (fairness). The α parameter lets you tune the balance.

Output: Channel attribution % with 95% confidence intervals

COVARIATION

Kelley's Model

Tags behaviors as dispositional (personality-driven) or situational (context-driven) using Consistency, Distinctiveness, and Consensus.

Output: Behavioral tags with psychological grounding

AI ANALYSIS

LLM Reports

Generates three reports: Executive Summary (insights), Technical Analysis (methodology), and Risk Assessment (limitations).

Output: Actionable recommendations + caveats

ETHICAL ARCHITECTURE

This is a Personal Epistemic Instrument—designed for self-reflection, not surveillance. We analyze what you did, never what you felt.

Aggregate patterns
Declared context
No mental state inference
LIVE DEMO

INTERACTIVE HYBRID
ATTRIBUTION ANALYTICS

Experience the power of Markov-Shapley hybrid attribution with real-time visualization and advanced uncertainty quantification. Explore channel performance and journey patterns.

🧠

First-Principles Attribution Engine

Personal Epistemic Instrument v1.0.0

A thinking instrument that transforms your behavioral data into actionable insights using rigorous mathematical models. Unlike black-box analytics, every calculation is transparent, defensible, and grounded in first principles.

α Blend ParameterCurrent: 0.5

Control the balance between causality and fairness in attribution

Formula: Hybrid = α·Markov + (1-α)·Shapley
⚖️ Shapley
(Pure Fairness)
⛓️ Markov
(Pure Causality)
🎯 Balanced: Equal weight to causal impact and fair distribution (recommended for most analyses)
5
Total Journeys
$425
Total Revenue
80%
Conversion Rate
5
Active Channels

📊 Journey State Machine3D Interactive

Each node is a Markov state. Drag to rotate. Connections show transition probabilities between touchpoints.

💡 Reading this: START → channel states → CONVERSION. The path shows how users flow through your funnel.

🌳 Channel HierarchyAnimated

Center: All journeys → Inner ring: Channels → Outer ring: Individual journeys

Search Direct Email Paid
💡 Hover over nodes to see attributed values. Node size reflects touchpoint frequency.

📈 Multi-Dimensional Journey ViewParallel Coordinates

Each line = one touchpoint. See how journeys traverse: Channel → Position → Touchpoints → Duration → Value → Conversion

💡 Reading this: Lines going to "Converted: Yes" (top right) show successful paths. Hover to highlight entire journey.

🌊 Transition Flow RiverLive Animation

Animated particles show transition probability flow. Wider streams = more frequent transitions.

💡 This is your Markov matrix visualized. Each particle represents probability mass flowing between states.

📋 Attribution Resultsα = 0.5

Final credit allocation with 95% confidence intervals from Bootstrap + Dirichlet UQ

ChannelHybrid ShareAttributed Value95% CIConfidence
💡 CI = Confidence Interval. Narrower range = more certainty. "High" confidence means the interval is <10% of the value.

⚙️ Model Configuration

IR Version: 1.0.0
Markov Order: First-order
Shapley Mode: Exact enumeration
Coalitions: 2^n = 32
UQ Method: Bootstrap + Dirichlet
Psychographic: Context-weighted
Key Invariant: All channel shares sum to exactly 1.0 (Efficiency Axiom enforced)

🛡️ Ethical Boundary

This is a Personal Epistemic Instrument—designed for reflection, not surveillance. It helps you understand your own behavioral patterns.

Allowed: Aggregate behavior, transition patterns, declared context
Prohibited: Mental state inference, predictive profiling, surveillance

"Layer 4 (Psychographic Inference) is architecturally prohibited. We analyze what you did, never what you felt."

Common Questions

Why not just use last-click attribution?

Last-click ignores all earlier touchpoints. A user might see 5 ads before converting—last-click gives 100% credit to the final one. Our hybrid model credits the entire journey fairly.

What makes this "first principles"?

Every calculation is derivable from axioms. Shapley values satisfy Efficiency, Symmetry, Dummy, and Additivity. Markov chains are provably row-stochastic. No black boxes.

What do the confidence intervals mean?

We resample your data 10,000 times (Bootstrap) and add Bayesian smoothing (Dirichlet). The 95% CI shows the range where the true value likely falls.

Can I trust these numbers for decisions?

Yes, within the model's assumptions. Check the CI width—narrow = high confidence. Compare against holdout tests. This is a scientific instrument, not an oracle.

CLASSIFIED - PRIVACY-FIRST - LOCAL PROCESSING - SCHEMA-VALIDATED

2024 ATTRIBUTION MATRIX - ALL SYSTEMS OPERATIONAL