Entities most influential in the network's topology. PageRank approximates each node's structural importance based on incoming connections from other influential nodes.

Which patterns are active in each operational layer? L17 (Ideological Architecture) has the highest pattern density at 6 patterns active.

Bayesian analysis of four operational questions about the trafficking network's documented end-state and possible continuation.

Simulating 200 perturbations of the current conflict trajectories. Each bar shows the probability that the conflict ends in that state, with the 90% confidence interval shaded after the fill.

Two-stage matching: topic relevance gate (entity/actor mentions) + criterion-keyword scoring (specificity-weighted). Surfaces evidence for human review at threshold 0.30.

Reconciles intermediary-layer entities against ICIJ's leaked corporate registry data. Strong match (score ≥ 50) warrants further investigation; common surnames generate false positives.

Sourcing. Every node and edge in the corpus is sourced from at least one of: court filings, congressional records, investigative journalism (named outlet, named reporter where available), corporate registrations, financial disclosures, declassified documents, or peer-reviewed research. Inferences are tagged separately from documented connections.

Layers. The 19 operational layers are not categorical — most prominent entities span 8-18 layers. Layer assignment reflects an entity's documented operational center of gravity, not its sole domain.

Patterns. Eight structural patterns are detected by name + tag matching across all 20,008 derived events. Pattern instances do NOT establish causation or guilt — they identify recurring operational signatures whose density and correlation can be measured. False positives are expected; the falsification criteria for each pattern are made explicit.

Predictions. Each prediction logs explicit confirmation criteria, explicit falsification criteria, an evidence basis, and a probability. The engine cannot log a prediction without falsification criteria — this is enforced architecturally. Predictions are scored by Brier score on resolution; the auto-scorer never resolves a prediction without human confirmation.

Primary source categories represented in the corpus, in approximate order of weight in evidence basis:

The corpus and predictions update on a daily refresh cycle. Pattern detection and Markov sensitivity recompute on every refresh. The auto-scorer ingests news feeds at each refresh and surfaces flags for review. ICIJ reconciliation runs nightly.