AI-generated report (Claude, Anthropic) — iteratively fact-checked against source documents but may contain errors. Verify claims against linked EFTA sources before citing. No affiliation with Anthropic.

Epstein Files — Investigation Reports

What This Is

165+ forensic analysis reports examining the DOJ's 2.91 million page Epstein case file release (January 30, 2026). Every claim in these reports cites specific EFTA document numbers that can be verified against the original PDFs on justice.gov.

These reports are the analysis layer. The underlying data — searchable databases of all 2.91 million pages — lives in the companion repo: rhowardstone/Epstein-research-data.


First Session Bootstrap

If this is your first time working with these files:

  1. Read METHODOLOGY.md and WRITING_GUIDE.md. These contain the investigation pipeline, fact-checking rules, and writing standards. Memorize the key points.
  2. Note the absolute paths to both this repo and the companion data repo in your memories.
  3. Verify database setup by running the verification query in the data repo's CLAUDE.md.
  4. Store a directory structure index in your memories so you can navigate reports without re-scanning.

These auxiliary files are kept separate from CLAUDE.md to save context window tokens. Read them once, internalize the rules, and refer back when needed.


Directory Map

overview/ — Start here (10 reports)

Executive summaries and master findings.
- FINAL_INVESTIGATION_REPORT.md — Comprehensive findings across all investigation lines
- PHASE4_BRIEFING_KIT.md — Condensed briefing with key evidence chains
- SESSION9_MASTER_FINDINGS.md — Latest session consolidated findings

congressional/ — Congressional briefings (8 reports)

Materials prepared for congressional oversight.
- CONGRESSIONAL_SUBPOENA_GUIDE.md — 64 witnesses, 490 verified EFTA URLs, recommended subpoena targets
- WITNESS_BRIEF_INDYKE.md / WITNESS_BRIEF_KAHN.md — Deep-dive witness preparation briefs

financial/ — Follow the money (20 reports)

Shell entities, wire transfers, Deutsche Bank compliance failures, art market laundering.
- SHELL_ENTITY_MAP.md — Complete map of Epstein's corporate structure
- FORENSIC_ACCT_*.md (6 parts) — Forensic accounting series: trust drawdowns, money sources, inter-entity flows
- CRYPTO_NETWORK_INVESTIGATION.md — Cryptocurrency and financial network analysis

individuals/ — Named person investigations (28 reports)

Deep dives on specific people appearing in the files.
- BILL_CLINTON_INVESTIGATION.md / DONALD_TRUMP_INVESTIGATION.md — Political figures
- INVESTIGATION_6_LEON_BLACK.md / LEON_BLACK_PROSECUTION_FAILURE.md — Apollo Global
- WILLIAM_BARR_INVESTIGATION.md — AG recusal and NTOC tip handling
- PSEUDONYM_CODENAME_REGISTRY.md — 273 pseudonyms/codenames found in the corpus

institutional/ — System failures (19 reports)

How institutions enabled or failed to stop the operation.
- DOJ_DOCUMENT_ALTERATION_FORENSICS.md — 212,730 changes detected between original and current DOJ-hosted documents
- DOJ_DOCUMENT_REMOVAL_AUDIT.md — ~64,259 documents removed from justice.gov after initial release
- FBI_302_MISSING_SERIALS_DOSSIER.md — FBI interview records that should exist but are missing from production
- SECONDARY_BATES_STAMP_ANALYSIS.md — Hidden numbering systems revealing production manipulation
- DS12_EXPANSION_ANALYSIS.md — 23 new documents quietly added March 2026

evidence/ — Device and physical forensics (13 reports)

Computer forensics, phone records, PLIST analysis, blackout periods.
- DEVICE_FORENSICS_COMPLETE.md — All seized electronic devices
- PHONE_RECORDS_INVESTIGATION.md — Call records analysis
- BLACKOUT_PERIOD_INVESTIGATION.md — Gaps in the evidentiary timeline

pqg_lines_of_investigation/ — Prosecutorial gaps (11 reports)

What the grand jury subpoenas demanded vs. what was actually produced.
- 00_INDEX.md — Start here: overview of all 10 investigation lines
- Topics: temporal blackouts, redacted targets, tech company gaps, travel records, Deutsche Bank, financial no-returns, crypto dead ends, MCC death investigation

intelligence/ — Intelligence connections (4 reports)

government-officials/ — Branch-by-branch analysis (7 reports)

Organized by party and branch: executive, judicial, legislative (Democrat/Republican/Independent).

victims/ — Victim analysis (4 reports)

Trafficking routes, victim census, interview analysis. All victim-identifying information is redacted.

methodology/ — Data quality and methods (12 reports)

Redaction analysis, data quality audits, evidence reliability assessments, corpus inventory.

Other directories

Directory Reports Focus
art/ 4 Art market investigation
scientists/ 3 Science network, David Shaw, biotech connections
social-networks/ 6 Geffen, Kotick, Peggy Siegal, Reuben Brothers
raw-dataset-analysis/ 11 Per-dataset deep dives (DS8, DS10)
internet-theories/ 3 Corpus searches testing internet claims (pizzagate, occult, misc)
recovered_corrupted_pdfs/ 1 Recovered data from corrupted PDFs in the DOJ release

Find reports mentioning a person or topic

grep -rl "Leon Black" --include="*.md" .

Search with context

grep -rn "EFTA00074206" --include="*.md" -C 2 .

Find reports about a specific EFTA document

grep -rl "EFTA01688067" --include="*.md" .

Accessing Original PDFs

Every EFTA number cited in these reports can be opened as a PDF. The DOJ is the canonical source (requires age gate); two independent mirrors serve byte-identical raw PDFs with no gate.

Source URL Template Notes
DOJ justice.gov/epstein/files/DataSet%20{N}/EFTA{NUMBER}.pdf Canonical. Requires dataset number + age gate.
RollCall media-cdn.rollcall.com/epstein-files/EFTA{NUMBER}.pdf Raw PDF, no gate, no dataset needed. Full coverage.
Kino/JDrive assets.getkino.com/documents/EFTA{NUMBER}.pdf Raw PDF, no gate. DS1-7 + DS9-11. Missing DS12.
Kino/JDrive (DS8) assets.getkino.com/documents/vol00008-official-doj-latest-efta{number}.pdf DS8 uses different naming. Note lowercase efta.
JMail Viewer jmail.world/drive/EFTA{NUMBER}.pdf (or vol00008-... for DS8) Formatted viewer. Uses Kino CDN.

To find the correct dataset number for DOJ URLs, use the companion data repo's database:

sqlite3 full_text_corpus.db "SELECT dataset FROM documents WHERE efta_number = 'EFTA00074206';"

Or use this boundary table:

Dataset EFTA Range
1 00000001 – 00003158
2 00003159 – 00003857
3 00003858 – 00005586
4 00005705 – 00008320
5 00008409 – 00008528
6 00008529 – 00008998
7 00009016 – 00009664
8 00009676 – 00039023
9 00039025 – 01262781
10 01262782 – 02205654
11 02205655 – 02730264
12 02730265 – 02858497

Note: Dataset boundaries have gaps — always verify via the database when possible.


Critical Rules

  1. Never include real victim names or identifying details. Use pseudonyms (Jane Doe, JD#1) or EFTA references only. The goal is exposing the system and the perpetrators, never retraumatizing or identifying victims.

  2. Corpus absence ≠ non-existence. A document not appearing in the EFTA production may be under seal, filed in a different case, or outside the scope of the DOJ release. Do not assume non-compliance from a gap.

  3. Present data, not conclusions. These reports document what the files contain. Readers draw their own inferences.

  4. Verify before citing. Cross-check EFTA numbers against the actual documents before repeating claims.


Companion Repository

The searchable databases behind these reports: rhowardstone/Epstein-research-data — 6.3 GB full-text corpus with FTS5 search, concordance metadata, redaction analysis, image descriptions, audio transcripts, and more. See that repo's CLAUDE.md for setup instructions and a SQL query cookbook.


Updating

# Pull new reports and data files
git pull origin main

To check what's new on the remote before pulling:

git fetch origin main
git log HEAD..origin/main --oneline

Always fetch from the remote before reporting what's changed — your local checkout may be behind.


Read-Only Distribution

These repositories are distributed read-only. If you cloned them to investigate locally:

Flag an error or leave a note
Ask about this report

Ask a question — the AI has the full report loaded and can also search the full corpus.