Data Narrative 20 — Epstein Forensic Finance Project

The Verification Wall

You saw the numbers. You had questions. Good. So did I. Here's what $144 million in claimed transactions looks like when you check the bank documents.
Randall Scott Taylor
February 2026
These narratives are living documents — markdown-rendered views into a relational database that updates as new records are verified, reclassified, or corrected. They are not PDFs. Archiving or saving will capture an HTML snapshot that may contain outdated figures if the underlying data has since been updated. For current numbers, always reference the live version.
$144.4M
Claimed (Noise)
$0
Verified (Noise)
$310.5M
Black Verified
15
Bank Documents
42
Verified Wires

In Narrative 19, I published the full publication ledger — 10,964 transactions, $2.146 billion, every name the money touched. The internet saw names and dollar amounts. Donald Trump: $65 million. Bill Gates: $37 million. Bill Clinton: $18 million. And they reacted.

Half the responses said I was covering for someone. The other half said I'd been compromised. A few just wanted to know: how did those numbers get there, and are they real?

That's the right question. So I ran every name through the verification wall — the same forensic standard I apply to every entity in this project. Every document has a Bates number. The question is whether it points to a bank record. No bank document, no transaction.

What Is a Bates Stamp?

A Bates stamp is a page number. Every page in a legal document production gets one. When Deutsche Bank turned over records to SDNY prosecutors, each page got a stamp — DB-SDNY-0002962, DB-SDNY-0003484, etc. When the DOJ released the Epstein corpus under the Transparency Act, those pages got EFTA numbers — EFTA01075607, EFTA01527348. They don't change. Any researcher can pull the original document by the number.

Here's the thing: every document in this corpus has a Bates number. The forwarded Bloomberg newsletter has one. The news article about Putin has one. Ghislaine Maxwell's Christmas card list has one. Having a Bates stamp doesn't make something a financial record. The question is what's on the other side of it — a wire transfer instruction with account numbers and routing codes, or a CNN article that happened to mention two names near a dollar sign. When I say "verified," I mean the Bates stamp points to a bank document. A wire instruction. An account statement. Not a news clipping.

Season 1 Finale
N19 — I Audited $2.1B in Epstein Financial Records
10,964 transactions across 14 banks, 8 shell entities, and 123 connected nodes.
I

The Wall

Claimed Volume vs. Verified Bank Records
Every document in this corpus has a Bates number. The question is what the Bates stamp points to. Gray bars: dollar volume from documents where the stamp points to news articles, court filings, and email forwards. Gold bars: where the stamp points to a bank document — wire instructions, account statements, transaction records. Scale = $310M.
Bates → news / court filings
Bates → bank documents
Person of Interest
Bank Docs
Donald Trump
$64.7M → $0
$00 bank docs
Bill Gates
$36.6M → $0
$00 bank docs
Bill Clinton
$17.6M → $0
$00 bank docs
Prince Andrew
$13.9M → $0
$00 bank docs
Ehud Barak
$6.0M → $0
$00 bank docs
Mort Zuckerman
$4.2M → $0
$00 bank docs
Jes Staley
$1.4M → $0
$00 bank docs
Reid Hoffman
$611 → $0
$00 bank docs
Leon Black
$310.5M — 42 VERIFIED WIRES
$310.5M15 bank docs
Joi Ito
$2.97M — 8 records
$2.97MConfirmed
Les Wexner
$1.44M — 9 records
$1.44MConfirmed

Eight names. $144.4 million in claimed transactions. Every source document has a Bates number — they're all in the DOJ corpus. But not one of those Bates stamps points to a bank document. Not one verified wire. Not one bank statement entry. Not one account number. The Bates stamps are real. The documents behind them are news articles, court filings, and political newsletters.

Then there's Leon Black. $310.5 million across 42 verified wires, each one traceable to a specific Deutsche Bank document with a Bates stamp. Same corpus. Same pipeline. One name produces bank documents. The rest produce news clippings.

II

Anatomy of a Phantom

Trump's $64.7 million came from 75 records in the fund_flows_audited table — a text-extraction pipeline that reads documents, finds two entity names and a dollar amount in the same paragraph, and logs them as a "transaction." Every single record was low confidence. The pipeline's own classifier tagged every one as NOTABLE_NOISE.

NLP ExtractionAmountActual Source
VLADIMIR PUTIN → DONALD TRUMP$5,600,000DS9 Email — Russian diplomatic supplies article ⓘ
DONALD TRUMP'S → DONALD TRUMP$5,800,000DS9 Email — Epstein sweetheart deal coverage ⓘ
MITT ROMNEY → DONALD TRUMP$5,000,000DS9 Email — Powell endorsement coverage ⓘ
GEOFFREY BERMAN → DONALD TRUMP$4,600,000DS9 Email — Ukraine consulting / Skadden Arps ⓘ
DONALD TRUMP → HUNTER BIDEN$650,000DS9 Email — A&F CEO / FBI article ⓘ
SAID TRUMP → CARES ACT$1,200DS9 Email — Coronavirus relief article ⓘ
The Possessive Grammar Problem

"Donald Trump's $3.4 million settlement." The NLP creates two entities — DONALD TRUMP'S as sender and DONALD TRUMP as receiver — and logs a $3.4 million self-transfer. Twenty-four records. $23.9 million. From apostrophe-S.

The pattern holds for every noise-tier POI. Gates' $36.6 million came from Boris Nikolic and Melinda Gates co-occurring near dollar figures in text. His single bank memo hit was a Morgan Stanley market commentary mentioning "Lions Gates 'Hunger Games' opens with M at the domestic weekend box-office." A film studio. Not Bill Gates.

Clinton's $17.6 million includes THOMAS JEFFERSON → BILL CLINTON: $425,000 — a founding father wiring money to a president. In 2015. Prince Andrew's $13.9 million is court filing text with names near dollar amounts in legal narratives. All of it from the same extraction pipe.

Verification Result

We searched master_wire_ledger (481 records), publication_ledger (10,964 records), and verified_wires (185 records). Three tables backed by bank documents — wire instructions, account statements, CHIPS/SWIFT logs. 11,630 financial records total. Trump, Clinton, Gates, Prince Andrew, Barak, Zuckerman, Staley, Hoffman — none of them appear in a single one. Their names exist in the corpus. Their Bates stamps point to news articles and court filings. Not one points to a bank document.

III

How the Noise Machine Works

The fund_flows_audited table is a text-extraction pipeline. Not a bank ledger. It processes documents from the DOJ corpus, identifies entity names using named-entity recognition, finds dollar amounts in proximity, and logs them as potential financial flows.

Epstein's
Email Inbox
NLP Entity
Extraction
Name + Dollar
Co-occurrence
fund_flows
_audited

The problem: Epstein's staff forwarded him political news roundups, market commentaries, and legal briefings. Dataset 9 — emails — generated 77% of all fund_flows_audited records. The NLP processed 2,533 unique email files and extracted 5,673 "transactions." Almost all of it is news coverage that happened to mention famous names near dollar signs.

The Noise Floor

fund_flows_audited: 7,355 total records. 94.8% low confidence. $3.34 billion in phantom volume. The pipeline's own NOTABLE_NOISE category — $277 million — is literally where Trump, Clinton, Putin, Pompeo, and Hunter Biden live. The pipeline knew they were artifacts.

IV

What Real Money Looks Like

Leon Black's financial footprint doesn't come from NLP guesswork. It comes from bank documents. Each wire traces to a Deutsche Bank record bearing a Bates production stamp from the SDNY litigation. These aren't inferences. They're photocopied wire instructions with account numbers, routing codes, dates, and authorized signatures.

DateFromToBatesAmount
2013-10-15Leon & Debra BlackSouthern Trust Co.DB-SDNY-0002962$8,500,000
2013-12-18Black Family Partners LPSouthern Trust Co.DB-SDNY-0003484$10,000,000
2014-04-29Leon & Debra BlackSouthern Trust Co.DB-SDNY-0003819$15,000,000
2014-07-15Narrow Holdings LLCSouthern Trust Co.DB-SDNY-0004009$20,000,000
2014-10-16Black Family Partners LPSouthern Trust Co.DB-SDNY-0004172$13,000,000
2015-10-14BV70 LLCGratitude America Ltd.DB-SDNY-0004999$10,000,000
2017-03-31BV70 LLCPlan D LLCDB-SDNY-0006113$22,500,000
10/31/2008Jeffrey EpsteinBlack Family Partners LPEFTA01075607$50,000,000

Southern Trust Company is Epstein's entity. Every "MONEY IN" wire represents Black family money flowing into Epstein-controlled accounts. The reverse — Epstein sending $76.9 million back to Black Family Partners in 2008 — came from a Bank of America statement bearing FinCEN SAR-corroborated Bates stamp EFTA01075607.

The Black entity universe spans 28 unique names in the publication ledger. Nine identified staff members. Two banking institutions. Senator Wyden's S.2746 names the same entities. Congress is mapping the same architecture.

The Authorization Chain

Wechsler: $84.4M · Alexanderson: $61.5M · Debra Black: $58.9M · Ada Clapp: $15M · Halperin: $13.9M · Spinella: $8.2M · Hannan: $3.5M · Joslin: $925K · Cohen: $884K

Banks: JPMorgan ($83M, 27 records) · Deutsche Bank ($62M, 12 records)

Shells: Plan D LLC ($44.5M) · BV70 LLC ($40.5M) · Southern Trust ($33M)

Forensic Workbook
Full Dataset — Verified Wire Ledger, Entity P&L, Shell Layering
Google Sheets — 8 analytical tabs. Every wire traceable to source documents.
V

Why This Is Hard

The DOJ released 1.48 million documents. No schema. No index. No table of contents. Bank wire instructions sit in the same file dumps as forwarded Bloomberg newsletters, legal memos quoting settlement figures, court filings narrating decades of financial history, and Ghislaine Maxwell's Christmas card list. Every one of these documents contains names. Most of them contain dollar signs.

That's what makes this corpus dangerous to analyze at scale. A news article about a $5.8 million police settlement that mentions Trump looks identical to an NLP pipeline as a $5.8 million wire instruction naming Trump — unless you pull the Bates-stamped source document and see it's a CNN forward. A court filing recounting Wexner's $26 million trust contribution reads like a bank memo until you check the document behind the Bates stamp and find narrative text, not a transaction record.

The verification wall exists because the data demands it. Gross extraction volumes capture everything the pipeline touches — real wires, news mentions, OCR garbage, possessive grammar artifacts, and a founding father apparently sending $425,000 to a living president. The only thing that separates signal from noise in a corpus this size is the bank document. Bates stamp or it didn't happen.

This narrative applies that standard to every name the internet asked about after N19. The pipeline did its job. It extracted what it was designed to extract. But extraction is not evidence. The wall is.

VI

What Comes Next

N19 mapped $2.146 billion across 123 connected nodes. This narrative separated the noise from the signal. Now we go deeper.

Season 2 takes every verified person in the money — Black, Wexner, Ito, the shell operators, the banks — and runs a full forensic profile on each one. Not surface-level dollar totals. Wire-by-wire reconstruction. Source entity tracing. Authorization chain mapping. Where the money came from before it entered the network and where it went after it left. Every entity that touched real money in N19 gets its own deep dive, backed by the same standard that built this wall: Bates stamp or it didn't happen.