Methodology

How the composite implication score is computed — and how to interrogate it yourself.

What is the composite score?

The composite implication score ranks entities by depth of documented involvement across multiple public data sources. Higher scores indicate more frequent appearances across documents, flights, emails, and connections, adjusted by legal status. The score is not a measure of guilt — it is a measure of documentation density.

Scores are computed client-side, in your browser, using data we provide. You can adjust the weight of each signal below and see how rankings change in real time. This transparency is intentional: the methodology is the product.

Score Formula

// Composite score formula

Score = LegalMultiplier × Σ(wᵢ × log(1 + rawᵢ) / log(1 + maxᵢ)) / Σ(wᵢ) × 100

Where wᵢ is the weight for signal i, rawᵢ is the raw count for that signal, and maxᵢ is the maximum value of that signal across all entities. The result is multiplied by 100 to yield a 0–100 scale.

Log Normalization

Raw counts are log-normalized to prevent outliers from dominating. An entity with 1,000 document mentions scores roughly twice as high as one with 30 mentions — not 33× as high. This ensures the ranking reflects breadth of involvement across multiple signals, rather than a single extreme metric.

This compression is a deliberate editorial choice. Jeffrey Epstein’s document count (3,140) is roughly 10× the median. Without log normalization, he would visually dwarf every other entity and the treemap would be meaningless. The log scale reveals the structure of the network rather than just the top outlier.

Input Signals

SignalSourceWhat it measuresDefault weight
Document FrequencyEpstein ExposedCourt documents and FOIA releases1.0
Flight FrequencyEE / rhowardstone (max)Documented flights on Epstein aircraft1.0
Email FrequencyEpstein ExposedEmail mentions in document releases1.0
Connection CountEpstein ExposedDocumented connections to other persons1.0
KG WeightrhowardstoneProminence in curated knowledge graph1.0

All default weights are 1.0 — equal weighting across all signals. This is the only defensible editorial stance for a public-interest tool. Adjust them below to explore how different weightings change the rankings.

Legal Severity Multipliers

The legal status multiplier scales the weighted signal sum. Persons with stronger legal outcomes appear higher in the rankings relative to their raw documentation density. These multipliers reflect formal legal proceedings only — they are not editorial judgments.

Legal StatusSourceMultiplier
ConvictedCourt records / overrides2.0×
ChargedCourt records / overrides1.8×
Immunity DealCourt records / overrides1.6×
Settled (Civil)Court records / overrides1.4×
Accused (Civil)Court records / overrides1.3×
InvestigatedCourt records / overrides1.2×
TestifiedCourt records / overrides1.1×
Not ChargedCourt records / overrides1.0×

Adjust Weights — Live Preview

Move the sliders to change the relative weight of each signal. The ranking updates in real time. This is not a feature — it is the point. Rankings are a function of methodology, and methodology is a choice.

Data Sources

Epstein Exposed

Person profiles, flight counts, document counts, email counts, connection counts, bios

AccessOpen API (no auth required)
Entities~1,463 persons

rhowardstone Research Data

Knowledge graph entities, typed relationships, mention counts, legal status

AccessPublic GitHub repository (static JSON)
Entities489 curated persons

Epstein Investigation Archive

Role descriptions, flight records, entity cross-reference

AccessOpen API (rate limited ~100 req/min)
Entities23,000+ entities (filtered to persons)

Disclaimer

This application visualizes publicly available court records and FOIA documents. Inclusion does not imply guilt or wrongdoing. Legal status labels reflect court records only. “Not charged” is the default status for any person without a confirmed legal proceeding on file. This tool does not editorialize — it presents public data and lets users draw their own conclusions. The composite score is a mathematical function of document counts — it is not a determination of culpability.