Methodology

How it is built

Each finding is a structured research/*.json (entities, directed edges, amounts, status, and source URLs) plus a .md write-up. models/graph/build_graph.py consolidates the edges into data/graph.json and runs the Tarjan SCC. The Z3, TLA+, and Alloy models read that data. The dashboard, this site, and the reports are generated from the same data.

Graph layers

Every edge is tagged financial (capital/credit/compute flows — 117) or structural (governance/legal/security/ownership/statistics — 117). The SCC over the financial layer alone equals the SCC over all edges (structural_edges_add_no_cycle = True): the circular core rests on capital flows.

Grading

Claims outside the proven financial core are graded fact · contested · weak · unsupported and excluded from the Z3/TLA+/Alloy proofs. Intent is not inferred from adjacency.

Consistency checks

models/audit.py (cross-document number/coverage checks) and models/cross_review.py (edge-amount reconciliation, connectors, under-connection) run via scripts/new-research.sh. Current audit: 0 flags.

Primary sources

Load-bearing claims cite primary government and court records (SCOTUS, DOJ/SDNY, Treasury/OFAC, the Fifth Circuit, SEC EDGAR, Congress.gov, FDIC, BLS, ICIJ, NIST, exchanges). See the dashboard for a sample and the research index for per-block sources.

Data pipeline

The charts and the cross-sectional analysis are built from keyless/public feeds, cached under data/: fetch_fred.py (FRED CSV — rates, the ICE BofA OAS rating ladder, sovereign 10Ys), fetch_yahoo.py (ETF distribution yields — per-state munis, corporates by maturity), and fetch_tape.py (the FINRA TRACE corporate trade tape via the OAuth Query API; credentials read from the environment only, never stored). cross_section.py then computes dispersion, relative-value z-scores, snapshot rankings, and the PCA common-factor (PC1) share across the credit, sovereign, muni, and bank cross-sections.

Reproduce

bash run_all.sh runs the Z3 engines (circularity, reflexive marks, the self-marked-value theorem, the depreciation/duration-mismatch trap, the Fed trap, the defense/power chokepoints, age-verification futility), TLA+, Alloy, and the graph/bank/temporal/gold/defense/energy/contagion/cross-section models, then regenerates this site (including an on-site page for every research block). Python + Z3; Java for TLA+/Alloy (jars auto-fetched).

Open source · no backend · updated 2026-06-11 · contact resistant@tuta.com