Epstein Financial Forensics GitHub Repository

PostedAugust 6, 2026

UpdatedSeptember 6, 2026

ByEpsteinWiki

The Epstein Financial Forensics repository is a large scale effort to reconstruct financial activity found across the Department of Justice Epstein Files Transparency Act releases.

The project uses automated extraction, forensic accounting methods, entity resolution, confidence tiers, audit trails, narratives, and network visualizations to organize transactions connected to Epstein related people, companies, trusts, banks, and accounts.

Its published figures are project findings, not government totals or judicial conclusions. Researchers must verify individual transactions through the cited bank record, court exhibit, or EFTA document before publication.

Snapshot

Project name: Epstein Financial Forensics

Repository owner: randallscott25 star

Named author: R. S. Taylor

Primary function: Automated financial reconstruction from released Epstein records

Documents indexed as reported on April 16, 2026: 1,476,437

Media items cataloged as reported: 503,517

Publication ledger records as reported: 6,397

Total financial volume as reported: $2.308 billion

Verified master wires as reported: 481

Integrated datasets as reported: DS1 through DS12 and DS98 through DS104

What Is Epstein Financial Forensics?

Epstein Financial Forensics is a research repository focused on extracting and reconciling financial records from the released Epstein document corpus.

The project states that its team wrote the extraction code, designed the database schema, built the classification pipeline, and conducted the analysis. Artificial intelligence tools assisted with code development, while the project attributes the analytical decisions to its research team.

The repository is not simply a document search engine. Its stated goal is to model transactions, fund flows, entity relationships, shell companies, trusts, banks, operators, and terminal accounts.

Repository Structure

The repository contains sections for data, documentation, narratives, payment extraction, tools, and visualizations.

The documentation directory contains materials covering compliance, findings, methodology, network analysis, and database schema.

The payment extraction directory includes programs designed to process book transfers, checks, CHIPS records, Fedwire material, MT103 messages, and JPMorgan statements.

The narratives directory contains the project’s larger analytical reports, correction logs, and quality assurance material. The visualizations present relationships and fund flows in a more accessible format.

The Publication Ledger

The repository reported 6,397 publication ledger records representing $2.308 billion in total financial volume as of April 16, 2026.

This total should not be described as stolen money, trafficking proceeds, suspicious activity, or criminal revenue. The repository explains that its ledger includes both suspicious and apparently legitimate transactions.

The total represents activity reconstructed by the project under its current methodology. It may change as documents are added, duplicates are removed, entities are corrected, and transactions are reclassified.

The Four Evidence Layers

The repository organizes its canonical evidence model into four layers.

Layer One contains the master wire ledger. The project reports 481 individual wire transfers totaling $973.4 million and describes them as fully reconciled.

Layer Two contains extracted payments. The project reports 3,876 records totaling $529.2 million.

Layer Three contains bank statements and court verified records. The project reports 1,267 records totaling $512 million.

Layer Four contains audited fund flows curated from the automated extraction pipeline. The project reports 773 records totaling $293.4 million.

These are project defined evidence layers. Researchers should inspect the source and confidence assigned to each transaction rather than assuming that all layers carry equal evidentiary weight.

The Confidence Tier System

The repository uses four confidence tiers to separate stronger evidence from unresolved automated findings.

Tier One is described as court verified. It contains records linked to Bates stamped bank documents or court exhibits.

Tier Two is described as bank verified. It includes suspicious activity report narratives and bank statement line items.

Tier Three requires two or more independent sources.

Tier Four contains corpus extracted records where entity resolution remains pending.

The distinction is essential. A Tier Four record should not be reported with the same certainty as a transaction verified through a bank document and court exhibit.

The Verification Wall

The project refers to its separation between verified and unresolved data as the verification wall.

The repository states that Tier One through Tier Three figures are anchored to court exhibits, bank records, or several sources. Tier Four contains records that still require complete entity resolution.

This structure allows researchers to see where a number comes from and how much confidence the project assigns to it. The original evidence must still be reviewed independently.

Automated Payment Extraction

Financial extraction programs search documents for fields such as dates, amounts, senders, recipients, accounts, banks, references, and transaction types.

The repository includes specialized extractors for checks, bank statements, book transfers, CHIPS messages, Fedwire records, and MT103 payment messages.

Automated extraction can process far more documents than a person could review manually. It can also misread statement totals, dates, account balances, quoted amounts, repeated attachments, and numbers appearing in news stories or legal arguments.

The Phantom Transaction Problem

The project openly documents the danger of automated financial extraction. An earlier natural language processing pass produced 7,355 records, with 94.8 percent classified as low confidence and $3.34 billion described as phantom volume.

The project explains that many false hits came from emails containing forwarded financial news, Bloomberg material, legal memoranda, and names appearing near dollar signs.

This is one of the repository’s most important lessons. Finding a person and an amount in the same document does not prove that the person sent, received, controlled, or benefited from that money.

The April 2026 Integrity Audit

The repository reports that an April 14 integrity audit removed approximately $70 million from the ledger.

The removals included 21 statement summary rows totaling $53.9 million that optical character recognition had misread as individual transactions. The audit also reclassified 341 interest payments totaling $4 million and removed eight duplicates totaling $12.4 million.

The reported ledger changed from 6,767 records and $2.378 billion to 6,397 records and $2.308 billion.

The correction demonstrates that the project maintains an audit process. It also confirms that large automated totals can contain substantial errors before reconciliation.

Shell Companies And Trust Hierarchies

The repository organizes entities into a hierarchy of holding trusts, distribution trusts, operating companies, and personal accounts.

Named entities include Southern Trust Company, the 2017 Caterpillar Trust, the Haze Trust, Southern Financial, Jeepers Inc., Plan D LLC, Gratitude America, NES LLC, and accounts connected to Jeffrey Epstein and Darren Indyke.

The hierarchy is an analytical model created by the project. Each claimed relationship should be verified through account records, wire instructions, corporate filings, trust documents, or court exhibits.

For broader context, see the EpsteinWiki guide to Epstein’s companies, trusts, and financial infrastructure.

Suspicious Activity Reports And Total Financial Activity

The repository compares its reconstructed ledger with a suspicious activity report benchmark of approximately $1.878 billion.

The project explains that suspicious activity reports cover transactions flagged by financial institutions, while its larger ledger also includes activity that may be lawful or routine.

A transaction appearing in a suspicious activity report is not itself proof of a crime. Banks file these reports when activity meets regulatory reporting criteria. The reports are investigative leads rather than judicial findings.

Representative Financial Evidence

The released EFTA corpus contains bank communications, transfer instructions, statements, invoices, account material, and transaction records that can be used to test financial claims.

EFTA00166211 and EFTA00166317 contain records referencing transfers to Société Générale.

EFTA00166341 and EFTA00166541 contain records referencing transfers involving BNP Paribas.

EFTA01367420 contains a euro denominated payment record connected to a French vendor. These examples demonstrate the type of evidence required to support an individual ledger entry.

Bates Stamps And Source Traceability

The repository states that every Tier One through Tier Three figure traces to a Bates stamped source document.

A Bates stamp allows a researcher to locate the exact document or page supporting a transaction. This is more reliable than citing only a chart, database row, or aggregate total.

When publishing a finding, record the EFTA number, page, account, date, amount, sender, recipient, transaction type, and any ambiguity in the source.

Entity Resolution

Entity resolution is the process of determining whether different names, abbreviations, accounts, and companies refer to the same real person or legal entity.

This is difficult in the Epstein records because one entity may appear under several names. Accounts may be identified only by partial numbers. A person may act personally, as a trustee, through a company, or as an authorized signer.

An unresolved entity match can redirect a transaction to the wrong person or company. Tier Four records require particular caution because the repository states that their entity resolution remains pending.

What The Repository Can Establish

The repository can show how its research team extracted, classified, reconciled, and modeled financial records from the released corpus.

It can identify candidate transactions and provide source references for verification.

It can reveal patterns involving accounts, trusts, companies, banks, counterparties, dates, and amounts.

It can document the project’s confidence level, corrections, and analytical methodology.

The underlying bank record or court exhibit establishes the transaction itself.

What The Repository Does Not Prove

An extracted transaction does not automatically prove criminal activity.

A payment does not prove the purpose claimed by an analyst without supporting evidence.

A shared bank, trust, or company does not prove that every connected person participated in Epstein’s crimes.

A suspicious activity report does not establish guilt.

A Tier Four match does not establish a resolved sender or recipient.

The $2.308 billion total does not represent a proven total of trafficking proceeds.

The project’s aggregate figures are not court judgments or official government findings.

How To Verify A Financial Record

Record the ledger identifier, evidence layer, confidence tier, date, amount, sender, recipient, account, bank, and source citation.
Locate the Bates stamp or EFTA number attached to the record.
Open the original document through Epstein Data.
Confirm that the amount is an individual transaction rather than a balance, statement total, quoted figure, or summary line.
Verify the sender and recipient from the original fields.
Check whether the account holder and beneficial owner are the same person.
Look for duplicate copies of the same transfer.
Distinguish principal, interest, fees, reversals, credits, and internal transfers.
Compare the entry with court exhibits, statements, corporate records, and related correspondence.
Publish the confidence tier and unresolved questions with the finding.

Versioning And Changing Totals

The repository describes itself as an active forensic investigation. Figures can change when records are revalidated, entities are reclassified, or new source material is processed.

The main repository was last validated on April 16, 2026. Older narratives may display earlier validation dates or figures.

Researchers should record the repository version, access date, and live statistics used for a claim. Cached copies should not be treated as current without comparison.

License And Reuse Restrictions

The repository states that it became all rights reserved for new uses effective April 16, 2026. It says that new mirroring, redistribution, derivative works, republication, commercial use, and machine learning training require written permission from R. S. Taylor.

Limited quotation for criticism, commentary, news reporting, or academic research is permitted with attribution under the repository’s stated terms.

The underlying United States government documents remain separate from the repository’s copyright claim. Researchers can cite and analyze the original EFTA evidence independently.

How This Tool Supports EpsteinWiki Research

The repository can help EpsteinWiki researchers locate financial evidence, test entity relationships, identify transfers, map shell companies, and prioritize records for review.

It works best with the EpsteinWiki Search Hub, the EpsteinWiki report on Tommy Carstensen’s money network, the Epstein Data archive, and the Epstein Data financial reports.

The repository supplies a financial reconstruction model. Epstein Data supplies direct access to the evidence. EpsteinWiki should cite the original record while explaining how the analytical model connects it to the wider financial system.

Key Takeaways

Epstein Financial Forensics uses automated extraction and forensic accounting methods to reconstruct financial activity across the EFTA corpus.
The project reported 6,397 ledger records totaling $2.308 billion as of April 16, 2026.
That total includes activity that may be legitimate and must not be described as proven criminal proceeds.
The repository separates evidence into four layers and four confidence tiers.
Tier One through Tier Three records are described as supported by bank documents, court exhibits, or several sources.
Tier Four contains extracted transactions with unresolved entity matching.
The project removed approximately $70 million in errors and duplicates during an April 2026 audit.
Automated extraction can mistake statement totals, news articles, and nearby names for transactions.
Every important financial claim should be checked against the Bates stamped EFTA evidence.

Why Epstein Financial Forensics Matters

The Epstein files contain financial evidence spread across statements, transfers, emails, exhibits, corporate records, and bank compliance material. The repository attempts to turn those fragments into a structured financial system that researchers can examine.

Its strongest contribution is the separation of verified transactions from unresolved automated findings. Its correction history also shows why large financial datasets require constant auditing. The tool can reveal important leads, but the original bank record remains the authority.

Sources

Tags:

Financial Network