The Forsythe Attribution and Measurement Framework

A Global First-Principles Architecture for Enterprise Marketing Science

An end-to-end causal inference infrastructure bridging the gap between media spend and revenue reality.

About the Framework

With the deprecation of third-party cookies and the degradation of pixel tracking, traditional last-click attribution is mathematically obsolete. Most "modern" attribution is simply weighted correlation disguised as data science.

This repository serves as the central hub for the Forsythe Attribution Framework -- a comprehensive, peer-verifiable technical architecture combining Bayesian Marketing Mix Modeling, Causal Inference, and Real-Time Streaming Identity Resolution. It is designed for data teams at brands spending $1M+/month on media who require measurement infrastructure they can mathematically defend.

The framework spans 10 published technical whitepapers, each addressing a specific failure point in modern marketing analytics, and 11 production dashboards demonstrating working implementations.

System Architecture

The theoretical frameworks are operationalized via a Kafka-native streaming pipeline designed for sub-100ms real-time attribution, identity resolution, and budget optimization.

Engineering Principles

Epistemic Bound Measurement. Every model outputs a confidence interval. If we cannot bound the error, we do not report the ROAS.

First-Principles Causality. We move past last-touch correlation by utilizing Markov Chain state modeling for temporal causality and Shapley value decomposition for game-theoretic fairness in credit allocation.

Bayesian Uncertainty Quantification. Bounding epistemic vs. aleatoric error so media buyers understand the actual confidence interval of the reported return, not just a point estimate.

Privacy-First Identity Resolution. All identity resolution is handled via probabilistic clustering (Gaussian Mixture Models), removing reliance on deprecating third-party cookies and maintaining GDPR/CCPA compliance by design.

The 10-Paper Measurement Stack

Each paper is published on Zenodo with a permanent DOI and compiled in LaTeX. Full HTML versions are hosted via GitHub Pages.

Pillar	Research Paper	Live Dashboard
Foundation	A First-Principles Hybrid Attribution Framework	Streaming Engine
Optimization	Bayesian Media Mix Modeling	MMM Optimizer
Psychographics	Behavioral Profiling and Causal Uplift	Profiling Suite
Calibration	The Causal Calibration System	Causal Inference Suite
Identity	Probabilistic Identity Resolution	Identity Graph Demo
Broadcast	Live Event Attribution	Live Event Dashboard
Pipelines	Real-Time Streaming Attribution	Streaming Engine
Geo-Testing	Incrementality Testing at Scale	Incrementality Lab
Data Engineering	Marketing Data Connectors	Connector Hub
Reconciliation	The MMM-Incrementality Bridge	MMM Bridge

Additional Production Systems

System	Dashboard	Description
Experimentation Platform	A/B and Bandit Testing	Multi-armed bandit experiment management with sequential testing
Demand Forecasting	Forecast System	Time-series forecasting with hierarchical reconciliation
Portfolio Hub	Command Center	Central navigation across all deployed systems

Technical Stack

Layer	Technology
Attribution Engine	Markov-Shapley hybrid with cooperative game-theoretic credit allocation
Marketing Mix Modeling	Bayesian hierarchical models (PyMC, PyStan) with adstock and Hill saturation
Identity Resolution	Probabilistic graph with S_ij scoring, Gaussian Mixture clustering, device fingerprinting
Real-Time Processing	Apache Kafka, Apache Flink SQL, sub-100ms latency, exactly-once semantics
Causal Inference	DAG-based modeling, propensity scoring, geo-lift experiments, synthetic control
Incrementality Testing	Randomized holdout design, Difference-in-Differences, Bayesian Structural Time Series
Feature Store	Delta Lake with versioned feature pipelines
Frontends	Next.js, deployed on Vercel

About the Architect

Michael Forsythe Robinson is a Marketing Science Engineer and AI Systems Architect.

He specializes in the design and engineering of production-grade measurement systems that transform complex behavioral signals into defensible causal estimates. His work sits at the intersection of Bayesian statistics, streaming data infrastructure, and cooperative game theory -- applied to the specific problem of marketing budget allocation under uncertainty.

As the founder of Forsythe Publishing and Marketing, he provides fractional technical advisory for enterprise brands and high-growth agencies building in-house measurement capabilities.

Portfolio: Command Center
LinkedIn: Michael Forsythe Robinson
Email: Forsythepublishing@gmail.com
Newsletter: The Measurement Standard
Orchid: Open Researcher

Citation

@techreport{robinson2026attribution,
  author       = {Robinson, Michael Forsythe},
  title        = {The Forsythe Attribution and Measurement Framework},
  year         = {2026},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.18557680},
  url          = {https://doi.org/10.5281/zenodo.18557680},
  note         = {10-paper technical portfolio covering the full attribution stack}
}

License

All research papers are published under CC BY 4.0. Implementation code is MIT licensed.

Michaelrobins938/attribution-assets