TLDR
Custom-built parsing scripts extracted 2,894 FedEx shipments from DOJ-released invoices, spanning July 2000 through October 2005 (PAPER TRAIL Project, 2026a). Two accounts -- one personal, one corporate -- connected four primary locations: Manhattan, Palm Beach, Zorro Ranch, and St. Thomas. The records end abruptly four months before Epstein's first arrest.
The Invoices Nobody Read
Among the hundreds of thousands of pages in DOJ Data Set 10, there are stacks of FedEx invoices (U.S. Department of Justice [DOJ], 2025). They are not dramatic documents. They list tracking numbers, sender addresses, recipient addresses, weights, charges, and reference codes. Individually, each one is trivial. Collectively, they map a logistics network that moved packages between Epstein's properties for five years.
Our automated parser extracted 2,894 individual shipments from these invoices (PAPER TRAIL Project, 2026a). The source PDFs span document identifiers EFTA01312563 through EFTA01337164. Each invoice was processed through optical character recognition, parsed into structured fields, and loaded into a PostgreSQL database for analysis.
Two Accounts, Four Locations
The shipments billed to two FedEx accounts. The first, account 1144-2081-6, was registered to "Jeffrey E Epstein" at 457 Madison Avenue Floor 4, New York, NY 10022. The second, account 2292-07504, was registered to "NYSG LLC" at the same address (PAPER TRAIL Project, 2026a). NYSG LLC -- New York Strategy Group -- originated 199 shipments from the Madison Avenue office, functioning as what corporate filings describe as a financial advisory firm for ultra-high-net-worth individuals.
The primary shipping routes connected four locations that readers of this series will recognize: the 457 Madison Avenue office in Manhattan, the Palm Beach area (358 El Brillo Way), Zorro Ranch in Stanley, New Mexico, and 6100 Red Hook Quarter in St. Thomas, USVI. These four nodes form the geographic skeleton of Epstein's domestic operations, and the FedEx data confirms they were connected by a regular flow of physical packages.
The October 2005 Cutoff
The records end abruptly in October 2005. Epstein was arrested in Palm Beach in March 2006. The four-month gap means the FedEx corpus captures the period before law enforcement attention but not the period during it. This is not necessarily evidence of destruction -- it may simply reflect the scope of what was seized. But it creates a hard boundary in the data (PAPER TRAIL Project, 2026c).
That boundary has analytical consequences. When our cross-domain synthesis engine compared FedEx shipping patterns against financial transaction timelines, it flagged ten contradictions -- all of which turned out to be timing mismatches caused by the FedEx corpus ending before other data domains begin (PAPER TRAIL Project, 2026d). In other words, the October 2005 cutoff is the single largest source of apparent inconsistency in the entire cross-domain analysis.
OCR Challenges
FedEx invoices are not easy documents for automated text extraction. The fields are small, the print is thermal, and the formatting varies across invoice generations. Average text-extraction confidence for FedEx records sits around 0.4-0.5 -- well below the 0.8+ confidence typical of typed correspondence. This meant extensive post-processing was required.
A cleanup script corrected 801 rows using more than 280 exact name mappings, pattern-based corrections, and garbage-to-NULL conversions (PAPER TRAIL Project, 2026a). Names that text extraction rendered as character soup were either mapped to known individuals or explicitly marked as unparseable rather than left as misleading data. Of the 148 third-party shipments, 31 (21%) have no parseable date at all.
What the Analysis Produced
The FedEx analysis script generated 27 CSV exports covering temporal patterns, sender and recipient networks, cost distributions, weight anomalies, and geographic routing (PAPER TRAIL Project, 2026b). These exports revealed patterns that would be invisible in any individual invoice: shipping volume peaks, seasonal routing changes, and the identification of who had access to Epstein's billing accounts.
Ghislaine Maxwell appears in the records as both sender and recipient, including international shipments to family members at Moulsford Manor in Oxfordshire and to Katie Orchard care of Alexander Mann in London (PAPER TRAIL Project, 2026e). JEGE Inc. aircraft maintenance logistics appear in five reference fields, including entries for "2 ENGINE JEGE" and tail number N908JE, confirming that third-party shipments supported aviation operations.
Zorro Ranch Provisioning
Among the 148 third-party billed shipments, Zorro Ranch in Stanley, New Mexico was the number one destination with approximately 29 shipments (PAPER TRAIL Project, 2026e). This is consistent with the ranch's role as a remote property requiring regular supply shipments from the Manhattan office. The shipping patterns to New Mexico provide an independent confirmation of the property's operational status during this period -- a data point that complements the real estate records and corporate filings documented elsewhere.
The Bigger Picture
2,894 packages across five years, billed to two accounts, connecting four properties. The FedEx data is not the most dramatic evidence domain in this corpus. But it is among the most concrete. Every shipment has a tracking number, a date, a sender, a recipient, a weight, and a charge. It is the kind of evidence that does not require interpretation -- it requires only counting.
References
PAPER TRAIL Project. (2026a). FedEx shipment database [Data set]. fedex_shipments table, PostgreSQL.
PAPER TRAIL Project. (2026b). FedEx analytical exports [Data set]. _exports/fedex/
PAPER TRAIL Project. (2026c). Observations [Data set]. OBSERVATIONS.md, OBS-3.
PAPER TRAIL Project. (2026d). Cross-domain contradiction analysis [Data set]. _exports/synthesis/contradictions_all.csv
PAPER TRAIL Project. (2026e). FedEx third-party shipment analysis [Data set]. research/fedex_third_party_analysis.md
U.S. Department of Justice. (2025). Epstein document release, Data Set 10 [Government records]. Document IDs EFTA01312563-EFTA01337164.