TLDR
The corpus contains 224 parsed wire transfers totaling $24,059,535.45, spanning April 2004 to August 2019 across five banking institutions. Vision-language model processing recovered 58.9% of dates, 92.9% of originators, and 94.6% of beneficiaries from degraded bank documents that text-only OCR could not read.
What a Wire Transfer Corpus Looks Like
A wire transfer is a paper trail in its purest form: an originator, a beneficiary, an amount, a date, and sometimes a memo field explaining the stated purpose. When you have 224 of them spanning 15 years across five banks, you have the skeleton of a financial network (PAPER TRAIL Project, 2026a).
The wire transfers database table holds these 224 records, parsed from Deutsche Bank, TD Bank, Charles Schwab, FirstBank Puerto Rico, and Morgan Stanley documents scattered across Data Set 10. The total value of all transfers with non-null amounts (219 of 224) is $24,059,535.45 (PAPER TRAIL Project, 2026a). This is not the total amount of money that flowed through Epstein's financial network — Senator Wyden's Treasury investigation documented $1.08 billion in 4,725 wires (PAPER TRAIL Project, 2026b). These 224 represent what was extractable from the released bank documents.
Recovery From Degraded Documents
When Script 16 first parsed the bank records, many wire transfers had missing fields. Dates were absent. Originator names were blank. Beneficiary fields were unreadable. The bank documents — scanned, faxed, photocopied, and rescanned — had degraded to the point where text-only OCR (the standard process of converting images of text to searchable characters) produced garbled output (PAPER TRAIL Project, 2026c).
This is where the vision-language model became essential. A vision-language model, or VLM, is an AI system that processes images and text together — it can look at a scanned page the way a human would, reading handwritten annotations, interpreting faded stamps, and deciphering partially obscured text. Script 16g processed 66 documents covering 215 pages using Qwen2.5-VL-7B (a 7-billion-parameter open-source VLM). It recovered 47 dates, 44 originator names, and 2 beneficiary names, updating 53 wire records (PAPER TRAIL Project, 2026d).
Earlier sub-scripts had already made progress. Script 16b filled 139 of 200 null originator fields ($5.59 million in wires). Script 16c reprocessed 70 Regulation E forms (standardized consumer banking disclosures), recovering 36 recipients worth $1.47 million. Script 16e recovered 90 of 104 null beneficiaries, achieving 93.8% coverage (PAPER TRAIL Project, 2026c).
The result: 132 dates recovered (58.9%), 208 originators identified (92.9%), 212 beneficiaries identified (94.6%) (PAPER TRAIL Project, 2026a).
The Major Flows
Three wire flows dominate the corpus. The largest is the $14.66 million that flowed into Darren Indyke's IOLA (Interest on Lawyer Account — a special trust account governed by bar association rules, meant to hold client funds temporarily) through just three wires: $8 million and $4.66 million from Southern Trust Company via Charles Schwab, and $2 million from Epstein's personal Deutsche Bank account (PAPER TRAIL Project, 2026e).
Second is Harlequin Dane LLC, which received $7.75 million in seven wires from Southern Financial LLC. Indyke was sole signer on the Harlequin Dane account. The money flowed out to Seaford Avenue Capital, Signature Title Group, JR Watersports, and Jetsmarter (PAPER TRAIL Project, 2026e).
Third is a cluster of wires from F T Real Estate Inc. to Indyke's personal account: $3 million from FirstBank Puerto Rico with a memo reading "real estate transaction" (PAPER TRAIL Project, 2026a).
The Migration Pattern
Multiple wire memos reference "to close DB account" as Deutsche Bank accounts were systematically shuttered in early 2019 (PAPER TRAIL Project, 2026a). After Deutsche Bank terminated the Epstein relationship in December 2018, funds migrated to TD Bank (New York State Department of Financial Services [NYDFS], 2020). The TD Bank SAR (Suspicious Activity Report — a mandatory filing banks make when they observe transactions they cannot explain through normal business purposes), filed October 1, 2019, captured $47.3 million in suspicious activity across 25 subjects during this transition period (PAPER TRAIL Project, 2026e).
This is not speculation. The wire memos say it explicitly. The account numbers match. The timeline aligns: Deutsche Bank out, TD Bank in, Charles Schwab as intermediary, FirstBank Puerto Rico for real estate transactions. Five banks, one financial network (PAPER TRAIL Project, 2026a).
The Butterfly Trust Correction
One of the most important outcomes of the VLM recovery process was a correction. The Butterfly Trust had been previously reported as involving $15 million in wire activity. When Script 16e recovered the actual beneficiary fields and amounts from source documents, the verified figure was $150,000 in parsed wire amounts — two orders of magnitude lower (PAPER TRAIL Project, 2026c). The $2.65 million figure cited by NYDFS represents total Trust disbursements documented in the consent order, not the subset captured in the wire transfer table (NYDFS, 2020).
This correction matters. It demonstrates that the pipeline self-corrects when better data becomes available, and that inflated figures — however dramatic — are replaced by verified ones.
What 224 Wires Cannot Show
These 224 wires are a fraction of the total financial activity. They do not include the Butterfly Trust's 120-plus disbursements to women with Eastern European surnames. They do not include the $606.9 million that flowed through Southern Financial LLC. They are the wires that survived the document release process, the scanning process, the OCR process, and the parsing process — each step losing information (PAPER TRAIL Project, 2026a).
The 224 wires are not the financial story. They are the financial evidence that made it through.
References
New York State Department of Financial Services. (2020, July 6). In the matter of Deutsche Bank AG (Consent Order). https://www.dfs.ny.gov/industry_guidance/enforcement_discipline/ea20200706_deutsche_bank
PAPER TRAIL Project. (2026a). Wire transfer records [Database table]. PostgreSQL wire_transfers, 224 rows, $24,059,535.45, db=epstein_files.
PAPER TRAIL Project. (2026b). External government sources inventory [Research document]. research/external_government_sources.md
PAPER TRAIL Project. (2026c). Bank records classification and wire parser [Computer software]. app/scripts/16_parse_bank_records.py
PAPER TRAIL Project. (2026d). VLM wire NULL field recovery (Script 16g) [Data set]. 66 documents, 215 pages, 53 wires updated.
PAPER TRAIL Project. (2026e). TD Bank SAR extraction [Research document]. research/td_bank_sar_extraction.md
PAPER TRAIL Project. (2026f). Bank document classification [Database table]. PostgreSQL bank_documents, 229K rows, db=epstein_files.
PAPER TRAIL Project. (2026g). Wire transfer exports [Data set]. _exports/bank/