Skip to main content
Skip to content
Forensic Methodology

How We Follow the Money

Every dollar traced in our database is backed by a Bates-referenced court document. This page documents our sources, verification standards, and the methodology behind the forensic financial analysis of Jeffrey Epstein's financial network.

Last updated: March 27, 2026

2,594
VERIFIED FLOWS
1,699
ENTITIES TRACKED
17,108
BANK TRANSACTIONS
7
BANKS COVERED

Overview

The Epstein Exposed financial forensics database traces money movement through Jeffrey Epstein's network of shell companies, trusts, and personal accounts across seven banking institutions. Our data comes from two primary sources: court-filed exhibits from federal proceedings, and independent community research by volunteer analysts who transcribed bank statements from the EFTA document releases.

Unlike previous attempts to quantify Epstein's financial activity, we prioritize verifiability over volume. Every transaction in our verified dataset links to a specific Bates-referenced court document. Claims that lack documentary support are quarantined in a separate “Claims” tab with clear “UNVERIFIED” labeling, preserving them for future research without contaminating the verified record.

On March 26, 2026, we rebuilt the entire financial database from scratch after community researchers identified significant data integrity issues including double-counting, phantom NLP transactions, and inflated narrative summaries in the previous version.

Data Sources & Provenance

Five independent sources contribute to the database. Each source's methodology, coverage period, and confidence tier is documented below.

Master Wire Ledger

by R.S. TaylorIndependent researcher (not affiliated with Epstein Exposed)

T1–T2
Records
481 verified wire transfers
Volume
$973.4M
Period
2008–2019
Bank
Multi-bank (Deutsche Bank, BoA, JPMorgan, Citibank, Wells Fargo, Barclays)

Court-exhibit authenticated wires extracted from DOJ EFTA releases. 25-phase audit pipeline with amount dedup, date-aware census, and entity-pair resolution. Published as open data under CC BY 4.0.

Deutsche Bank & FirstBank USVI Transactions

by TS WhaleEpstein Exposed community researcher

T3
Records
17,613 transactions (10,426 DB + 7,187 FB)
Volume
$2.6B absolute volume
Period
2012–2020
Bank
Deutsche Bank, FirstBank USVI

Direct transcription from Deutsche Bank and FirstBank (USVI) statements filed as EFTA court exhibits. Every transaction has an EFTA Bates reference. Deutsche Bank data categorized by entity group (New York, Trusts, Aircraft, Florida/VI, Zorro Ranch) with 34 transaction categories. Also compiled 307 counterparty identifications with investigative context notes.

JPMorgan Transactions

by Anonymous ResearcherEpstein Exposed community researcher

T3
Records
784 transactions
Volume
$156.2M absolute volume
Period
2013–2014
Bank
JPMorgan Chase

Extracted from JPMorgan Private Bank and Classic Business Checking statements. Covers Jeffrey Epstein personal accounts, Darren K. Indyke PLLC, Southern Trust Company, and NES LLC.

Verification & Confidence Tiers

Every financial flow is assigned a confidence tier based on the strength of its documentary evidence. Higher tiers require stronger provenance.

T1Court-Verified190 flows

Transaction appears in a court exhibit with a Bates reference number. Verified by judicial proceedings.

Example: Alfano exhibit wires with DB-SDNY Bates stamps

T2Bank-Confirmed262 flows

FinCEN SAR amount match, bank statement correlation, or multi-source cross-reference confirmation.

Example: FinCEN amount matches, audited fund flows with PROVEN/STRONG rating

T3Community-Verified2,113 flows

Independently extracted from bank statements by community researchers with EFTA document references.

Example: Deutsche Bank wire transfers transcribed by TS Whale with EFTA citations

T4AI-Extracted0 flows

Automatically extracted from OCR text by the finance-extractor agent. Requires admin review before promotion.

Example: Pending extractions from 45,304 financial documents

Bank Coverage

The database covers transactions across seven banking institutions, spanning from 2008 to 2020. Coverage depth varies by bank — Deutsche Bank has the most comprehensive transaction-level data.

Deutsche Bank

2013–201935 entities

FirstBank USVI

2012–202010 entities

JPMorgan Chase

2013–20145 entities

Bank of America

2008–20198 entities

Citibank

2013–20186 entities

Wells Fargo

2014–20173 entities

Barclays

2016–20171 entities

Coverage gaps: Pre-2012 records are limited to court exhibits only. Some banks may have purged records under retention policies. Sealed court documents (estimated $40–60M in additional flows) remain inaccessible.

Deduplication Methodology

Preventing double-counting is the single most critical integrity challenge in financial forensics. The same transaction can appear in multiple source documents, court exhibits, and researcher datasets.

Composite Dedup Key

Every flow has a dedup_key computed as MD5(source_entity + target_entity + amount + date + bates_ref). This is enforced as a UNIQUE database constraint — it is physically impossible to insert the same transaction twice.

Cross-Source Detection

When community bank transaction data overlaps with the master wire ledger, the system detects the match via dedup key comparison and links the records rather than duplicating them. The bank transaction record gets a promoted_flow_id reference to the existing verified flow.

Entity-Level Aggregation

Entity totals (inflows/outflows per entity) are computed via a PostgreSQL materialized view that aggregates only from the verified flows table — never from raw bank transaction records. This prevents the double-counting bug identified in the original database where entity totals counted both sides of each flow.

What We Excluded & Why

The following categories of data were identified as unreliable and removed from the verified dataset. Some are preserved in the Claims tab for research purposes.

Narrative Summaries

$4.52B(58 records)

Aggregate claims without individual transaction documentation (e.g., '$1.08B JPMorgan suspicious transactions'). These were editorial assertions, not verified transactions. Quarantined to the Claims tab for transparency.

NLP Phantom Transactions

$3.34B (pre-filter)(~5,600 records)

Automated NLP extraction from news articles and emails incorrectly identified dollar amounts near person names as transactions. Examples: '$64.7M attributed to Donald Trump' from news coverage, '$36.6M to Bill Gates' from a journalist's mention. All $0 verified.

Possessive Grammar Artifacts

$23.9M(24 records)

NLP misinterpreted possessive constructions like 'Donald Trump\'s $3.4M settlement' as self-transfers. Identified and removed during Phase 22 audit.

Chain-Hop Inflation

$311M(Variable records)

Internal→Internal transfer chains created 4x multiplier effects where the same money was counted at each hop through shell companies. Identified in Randall Scott\'s Phase 22 audit.

Brokerage Noise

$155M(Variable records)

Automated brokerage account movements (margin calls, rebalancing) miscategorized as intentional transfers. Filtered via BROKERAGE_NOISE regex patterns.

ICIJ Offshore Leaks Cross-Reference

We cross-reference all persons in the database against the International Consortium of Investigative Journalists (ICIJ) Offshore Leaks database, which includes the Panama Papers, Paradise Papers, Pandora Papers, and Bahamas Leaks.

103
Panama Papers
692
Paradise Papers
32
Pandora Papers
30
Bahamas Leaks
98
Offshore Leaks

955 total matches across 342 unique persons, with 7,273 relationship chains connecting officers to offshore entities. Matches use a minimum fuzzy threshold of 85% and are available via the Offshore API.

Automated Discovery Pipeline

A dedicated AI agent continuously scans court documents for undiscovered financial transactions, feeding a human-review pipeline before any data enters the verified database.

Agent
Finance Extractor
Claude Sonnet 4.6, runs 2x daily (4AM/4PM UTC)
Document Pool
45,304 financial documents
Account records, wire transfers, bank statements, trust distributions, SARs, tax documents
Review Gate
Human approval required
Extractions become pending leads. Admin reviews before promotion to T4 verified status.

Community Contributions

Independent researchers can submit structured financial transaction data for review and inclusion in the database. All submissions go through admin review before promotion.

Submission Format

POST /api/financial/submit
Content-Type: application/json

{
  "contributor": "Your Name / Discord Handle",
  "source": "Bank statement transcription",
  "transactions": [
    {
      "from_entity": "Jeffrey Epstein",
      "to_entity": "Southern Trust Company",
      "amount": 50000,
      "date": "2015-03-15",
      "bank": "Deutsche Bank",
      "transaction_type": "wire",
      "efta_ref": "EFTA01234567",
      "description": "Wire transfer"
    }
  ]
}

Required fields: from_entity, to_entity, amount (min $100)

Optional: date (YYYY-MM-DD), bank, transaction_type, efta_ref, description

Max 500 transactions per submission. All submissions require admin approval.

Attribution & Licensing

This forensic financial analysis is built on the work of independent researchers who donated their time and expertise to public accountability. Proper credit matters.

Eric Keller

Creator — Epstein Exposed Forensic Finance System

Built the 14-tab forensic financial analysis system, designed the v2 database architecture with dedup-key integrity, imported and verified all community-contributed data, created the automated discovery pipeline, and developed the community submission API. Founder and lead developer of Epstein Exposed.

epsteinexposed.com

R.S. Taylor

Independent Researcher — Master Wire Ledger (not affiliated with Epstein Exposed)

CC BY 4.0

Created the foundational 481-wire master wire ledger through a 25-phase audit pipeline processing 1.48 million DOJ documents. Published as open data. Self-described as a data scientist whose "day job is multi-affiliate financial reconciliation at institutional scale." Explicitly notes: "I am not a forensic accountant. I am not a prosecutor. I don't hold a CPA or a CFE."

github.com/randallscott25-star/epstein-forensic-finance

TS Whale

Community Researcher — Deutsche Bank, FirstBank & Counterparty Data

Independently transcribed 17,613 bank transactions from Deutsche Bank and FirstBank USVI court exhibits, covering 2012 through 2020. Compiled 307 counterparty identifications with investigative context notes. Identified critical data integrity issues in the original database that prompted the full rebuild.

Anonymous Researcher

Community Researcher — JPMorgan Data

Extracted 784 JPMorgan Chase transactions from Private Bank and Classic Business Checking statements, plus monthly financial summaries for the 2013–2014 period.

Disclaimer: This analysis does not constitute an audit, examination, or review performed in accordance with GAAS, GAGAS, or AICPA SSFS No. 1. All financial amounts are sourced from court-filed documents and community research. Entity classifications are analytical, not legal determinations. Appearance in this database does not imply wrongdoing.

The database excludes verified victims and survivors in accordance with our data policy. For questions about the methodology, contact the research team via Discord.

Back to Follow the Money

Data built with transparency. Every claim is verifiable.