Skip to main content
Skip to content
Data Analysis

Dataset 10 - Interim Redaction Analysis Findings

28 EFTA citations1,249 words10 persons referenced

Note: This is a historical interim snapshot generated at ~15% scan completion. It has been fully superseded by DS10_COMPLETE_FINDINGS.md, which documents the complete scan results. All findings below (trust beneficiary names, victim interview transcripts, plea deal negotiations, financial records) are documented more completely in the final report. Additionally, per REDACTION_TEXT_LAYER_ANALYSIS (Report #93), the "hidden text" referenced throughout this report is predominantly garbled OCR from a

Dataset 10 - Interim Redaction Analysis Findings

Note: This is a historical interim snapshot generated at ~15% scan completion. It has been fully superseded by DS10_COMPLETE_FINDINGS.md, which documents the complete scan results. All findings below (trust beneficiary names, victim interview transcripts, plea deal negotiations, financial records) are documented more completely in the final report. Additionally, per REDACTION_TEXT_LAYER_ANALYSIS (Report #93), the "hidden text" referenced throughout this report is predominantly garbled OCR from an invisible Tr=3 text rendering layer placed over scanned images — not content deliberately concealed beneath PDF redaction annotations. While the identified content is real and has been verified, the data provenance is OCR extraction, not redaction-defeat.
Generated: 2026-02-05 (scan in progress) Status: ~15% scanned (77K of 503,154 PDFs), scan ETA ~1 hour Database: Dataset 10 document text database

Scale Comparison

MetricDS1-9,11-12 (all combined)DS10 (15% scanned)
--------------------------------------------------------
PDFs scanned16,28477,000
Total redactions179,139268,110
Bad overlays (recoverable text)70,94091,646
Proper redactions108,199176,464
Documents with hidden text9,13322,398
DS10 is the largest and most document-rich dataset by a factor of 30x. At only 15% scanned, it already contains more recoverable hidden text than all other datasets combined.

Key Findings

1. Epstein Trust Documents - Beneficiary Names Under Redaction

Multiple documents reveal the hidden names of Epstein's trust beneficiaries under redactions:

  • Jean-Luc Brunel - "GHISLAINE N. MAXW JEAN LUC BRUNEL an[d] Third of the Trust Agreement" (EFTA01297516) - Brunel was a French fashion agent who died in custody awaiting trial for trafficking
  • [Eva Anders]son Dubin - "SSON DUBIN, if she survi[ves]" (EFTA01266434) - Wife of billionaire Glenn Dubin
  • Lawrence Paul Visoski Jr - Epstein's pilot, beneficiary (EFTA01266380)

EFTA01333133, EFTA01333327)">2. Victim Interview Transcripts (EFTA01333133, EFTA01333327)

Detailed Palm Beach Police Department interview transcripts with extensive content under redactions:

  • "Epstein then instructed her to give him a massage pointing to a specific lotion for her to use"
  • "asked if Epstein ever asked her age and she stated he had"
  • "she stated she had just turned seventeen"
  • "she is still a minor"
  • "Jeff would try to get away with more and more on each massage"
  • "Epstein ejaculated which meant the massage was over. She was paid $200.00"
  • "she was paid $200.00 each time she went. Each time she went she was reminded not to speak of what [happened]"
  • "Epstein apologized for his actions and subsequently paid her a thousand dollars for that visit"
  • Descriptions of a "large white vibrator with a huge head on the tip"
  • "photographs of naked [subjects]"
  • "Epstein's penis was deformed"
  • References to Juan Alessi (Epstein's butler), Ghislaine, school recruitment
  • "from her family in Yugoslavia into the United States to be [brought]" - trafficking reference

EFTA01302111)">3. Lawyer Negotiations About Plea Deal (EFTA01302111)

Hidden email exchange about Epstein's plea negotiations:

  • "conversation re yesterday's call... regarding the New York Times"
  • "proof positive of what transpired. Proof positive of a breach of the rules of conduct"
  • "David Weinstein's actions" - breach of local rules of conduct
  • "talks about appeals, waivers... I think that is a non[starter]"
  • "our current proposal, while not what we would prefer, is currently the best overall solution"
  • References to "Ken" [Starr? Acosta?] as "a man of great faith"
  • "he believes it has failed miserably. His fear is that the Bob [?] to be a rubber stamp for the office"
  • "multiple charging statutes, even including the 1591" (18 USC 1591 = sex trafficking statute)
  • "patently unfair to Mr. Epstein"

EFTA01334281)">4. "Epstein's Buddy Clinton" (EFTA01334281)

Document mentions Maxwell identifying herself, a weapon being kept, and reference to "Epstein's Buddy Clinton" on page 28.

EFTA01368199)">5. Prince Andrew / Vanity Fair (EFTA01368199)

"Vanity Fair. The controversy over Prince Andrew's continuin[g]..." - article or communication about the Prince Andrew scandal.

6. Financial Records Under Redaction

  • Deutsche Bank - 66+ entries with "DEUTSCHE BANK AG, 60 WALL STREET" (Epstein's primary bank)
  • JPMorgan Chase - 9+ entries including "JPM-SDNY-00000331" (SDNY case discovery production)
  • TD Ameritrade - EFTA01267639 contains 1,204 bad redactions - entire portfolio statements, account positions, interest credits
  • MoneyGram wire transfer records - "WIRE TRANSFER RECORDS THAT INCLUDES NAMES" (EFTA01265800)
  • MCC financial records - Account transaction details (EFTA01265978) - appears to be Metropolitan Correctional Center commissary/account records

EFTA01339374)">7. Aircraft Sales (Post-Death Estate Disposal) (EFTA01339374)

Extensive email chain about sale of Sikorsky S76C+ helicopter (N162AE, SN 760472):

  • Aircraft N722JE and N162AE
  • Involves Larry Visoski (pilot), Darren Indyke (estate attorney), Richard Kahn
  • Wire transfer instructions
  • "LOI is in the name of Hype[rion]"
  • JPM Clearing Corp
  • Insured Aircraft / Guardian Jet brokers
  • "Helicopter 1029 LLC sn 760750"
  • Sale negotiations from 2019 into early 2020

8. Zorro Ranch (New Mexico) Property Records

40+ entries showing:

  • "ZORRO DEVELOPMENT CORP" - Epstein's shell company for the ranch
  • "49 ZORRO RANCH ROAD, STANLEY NM 87056"
  • New Mexico State Land Office address change requests
  • Oil/gas/mineral lease documents

EFTA01307744)">9. Island Structures Analysis (EFTA01307744)

Hidden descriptions of structures on Epstein's island:

  • "D1: Single Story Structure - 3 Dark double doors"
  • "D2: Potential 2-Story Structure"
  • "D3: Dome not seen on video footage / Potential Single Story Structure"
  • "Yellow Square: Potential Hidden Structure" - implies structures not visible in standard surveillance

10. Law Enforcement References

  • "Information provided by Victim-1 has proven reliable and has been corroborated by independent [evidence]" (EFTA01262900)
  • Alexander Acosta reference in case escalation (EFTA01353255)

11. Communication Records

  • Phone call detail records with locations (Palm Beach, New York, Queens, Boston, Connecticut, UK, France) (EFTA01334722)
  • ROAMING call records
  • Cell phone subscriber data

12. Seized Digital Evidence

  • Evidence of forensic disk imaging

Projected Final Numbers

At the current density rate (~1.19 bad overlays per PDF on average), the full DS10 scan will likely yield:

  • ~600,000+ total redaction regions
  • ~200,000+ bad overlays with recoverable text
  • ~140,000+ documents with hidden text

This would mean DS10 alone contains approximately 3x more recoverable hidden text than all other datasets combined.

Categories of Hidden Information (from 77K docs scanned)

CategoryDocument Count
-------------------------
Financial/Banking~640
Trust/Estate~146
Law Enforcement~38
Legal/Court~142
Property/Real Estate~75
Communications~385
Victim/Abuse~27
Named Associates~233

Next Steps

  • Complete full DS10 scan (ETA ~1 hour from report generation)
  • Extract all hidden text from DS10 for systematic review
  • Cross-reference DS10 names with knowledge graph
  • Search for additional powerful names that may emerge in remaining 85% of documents
  • Update HIDDEN_TEXT_COMPLETE_REVIEW.md with DS10 findings
    • View source on GitHub →