Systems Research Platform

data-journalism
system-dynamics
bayesian
visualization
Published

February 1, 2026

Overview

Status: Active — first deep dive published Impact: Live published analysis with interactive D3 visualizations and system dynamics models Technologies: Python, Marimo, PySD, PyMC, DuckDB, D3.js, Observable Framework, dlt

A data journalism and systems analysis platform investigating the physical and economic consequences of AI capital flows. Combines system dynamics modeling, Bayesian inference, and multi-source data pipelines to produce publication-quality interactive analyses deployed as a static site.

Published Work

Deep Dive 001: AI Valuations vs. Infrastructure Spending

  • Analysis of $4T+ in market cap gains against $320B in 2025 capex across the Magnificent 7
  • 10 interactive D3 scrollytelling charts with scroll-synced narrative
  • Data sourced from SEC 10-K/10-Q filings, Yahoo Finance, FRED, company earnings calls
  • Tracks off-balance-sheet commitments (operating leases, SPVs) not visible in reported capex

Technical Implementation

Quantitative Modeling

  • PySD system dynamics models: grid capacity, interconnection queues, behind-the-meter bypass, renewable cost learning curves (Wright’s Law), regulatory feedback loops
  • PyMC Bayesian inference for probabilistic modeling of uncertain parameters (AI growth rate scenarios, regulatory favorability)
  • hmmlearn Markov regime switching for energy market state transitions
  • statsmodels time series econometrics

Multi-Source Data Pipeline

  • dlt (data load tool) for reproducible ETL from FRED, Yahoo Finance, SEC EDGAR, BLS, Census Bureau, World Bank, COMTRADE
  • DuckDB as the analytical store with pyarrow for columnar data
  • Python data loaders that execute at build time and output JSON for client-side rendering

Interactive Visualization

  • 10 custom D3.js chart modules with scroll-triggered animation
  • Observable Framework for the public site with custom editorial design system
  • Marimo reactive notebooks for analysis (pure Python, importable, testable, versionable)
  • Custom flowmpl library (published on PyPI) for flow diagrams and publication figures

Key Features

  • Bespoke design system: Cormorant Garamond + DM Sans, warm paper aesthetic
  • Company-specific color palette coordinated across all charts
  • Automated build and deploy to GitHub Pages
  • Planned deep dives: Grid Modernization, Stranded Capex, Extraction & Opportunity Cost
  • GeoPandas spatial analysis for mapping where capital lands physically

Results & Lessons Learned

  • System dynamics correctly captures feedback loops that static analysis misses (interconnection queue growth → behind-the-meter bypass → cost allocation shifts)
  • Marimo over Jupyter: reactive, pure Python, exports to clean HTML, version-controllable
  • Multi-source data pipelines require careful provenance tracking — every number must trace to a filing or dataset
  • The gap between reported capex and actual infrastructure commitment (including operating leases) is significant and consistently underreported

Documentation & Resources