Epstein Files: Structured Data Exports

Structured data exports from the forensic analysis of the DOJ Jeffrey Epstein file release under the Epstein Files Transparency Act (Public Law 119-38). All 12 DOJ datasets plus House Oversight Estate and FBI Vault materials: 1.4M+ documents, 2.9M+ pages. Structured data exports from the forensic analysis of the DOJ Jeffrey Epstein file release under the Epstein Files Transparency Act (Public Law 119-38). All 12 DOJ datasets plus House Oversight Estate and FBI Vault materials: 1.38M+ documents, 2.77M+ pages.

epstein-data.com — searchable interface with full-text search, AI research assistant, document viewer, redaction analysis, deposition transcripts, geographic heatmap, and more.

Investigation Reports — 180+ forensic reports with DOJ source citations, organized by topic.

Desktop Install — run the full Claude-powered investigator locally with all databases. No tech skills required. (CLI version)

---

Investigation Reports

180+ reports across 18 categories. Each report cites specific EFTA documents you can verify at epstein-data.com. Browse all reports in the grid view.

Executive Summaries — Final Investigation Report | Master Report | Phase 4 Briefing Kit

Individual Investigations — Trump | Rothschild | Pseudonym & Codename Registry

Institutional Failures — DOJ Document Removal Audit | DS12 Expansion Analysis | Alteration Forensics

Congressional Briefings — Subpoena Guide | Witness Brief: Indyke | Deposition Analysis: Indyke

Financial Forensics — DiIorio / Apollo Whistleblower | Shell Entity Map | Leon Black Prosecution Failure

Evidence & Device Forensics — Evidence Compilation | Device Forensics | MCC Inmate Witness Interviews

Intelligence Networks — FBI Intelligence Investigations | Israel Deep Dive

Social Network Analysis — Geffen | Kotick / Activision | Peggy Siegal Pipeline

Victim Census & Routes | Art Market | Science Network | Government Officials | Dataset Analysis | Methodology | Internet Theories | Prosecutorial Query Graph

---

Quick Start

All databases are in a single release: v5.2 — Complete Database Collection (PII-redacted).

# Create the data directory (required for search tools)
mkdir -p data

# Download the split corpus parts (~2.3 GB total)
wget https://github.com/rhowardstone/Epstein-research-data/releases/download/v5.2/ftc_clean_part_aa
wget https://github.com/rhowardstone/Epstein-research-data/releases/download/v5.2/ftc_clean_part_ab
wget https://github.com/rhowardstone/Epstein-research-data/releases/download/v5.2/ftc_clean_part_ac

# Reassemble, decompress, and move to data/
cat ftc_clean_part_* > full_text_corpus.db.gz
gunzip full_text_corpus.db.gz
mv full_text_corpus.db data/

# Search the full text corpus
sqlite3 data/full_text_corpus.db "SELECT efta_number, page_number, substr(text_content, 1, 200) FROM pages WHERE text_content LIKE '%Leon Black%' LIMIT 10;"

# FTS5 full-text search (faster)
sqlite3 data/full_text_corpus.db "SELECT p.efta_number, p.page_number, substr(p.text_content, 1, 200) FROM pages_fts fts JOIN pages p ON p.rowid = fts.rowid WHERE pages_fts MATCH 'shell company' LIMIT 10;"

---

Databases

All databases available in the v5.2 release.

Database	Compressed	Description
full_text_corpus.db (after reassembly)	2.3 GB (split)	Master text database (split into ftc_clean_part_aa/ab/ac). 1.38M docs, 2.77M pages with full text and FTS5 search. DS1-12 + House Oversight (DS99) + FBI Vault (DS98).
concordance_complete.db	137 MB	Cross-reference with email threads, folder inventory, production metadata.
alteration_results.db	183 MB	212K change units with diff text, pixel-diff, LLM classification of post-release DOJ document modifications.
redaction_analysis_v2.db	166 MB	2.6M redaction boxes, 850K doc summaries, 39K reconstructed pages, 107K extracted entities.
redaction_analysis_ds10.db	87 MB	Dataset 10 deep redaction analysis.
image_analysis.db	64 MB	38K+ extracted images with AI-generated descriptions.
secondary_stamps.db	—	835K documents mapped to parallel Bates numbering systems (R1, JPM-SDNY, DB-SDNY, SDNY-GM).
deposition_transcripts.db	—	Whisper-transcribed depositions with speaker diarization (Wexner, Bill Clinton, Hillary Clinton, and others).
document_status.db	—	Document availability tracking across DOJ website.
handwriting_transcriptions.db	—	Handwritten document transcriptions.
native_files.db	—	Metadata for 3,800+ non-PDF native files (XLSX, media, etc.).
transcripts.db	1.7 MB	1,600+ media files, 435 with speech, 190K words (faster-whisper large-v3).
ocr_database.db	25 MB	Tesseract OCR results for degraded documents.
spreadsheet_corpus.db	—	Native spreadsheet data extracted from Dataset 8.
prosecutorial_query_graph.db	2.5 MB	Subpoena analysis: riders, returns, clause fulfillment, investigative gaps.
knowledge_graph.db	764 KB	524 curated entities, 2,096 relationships.
communications.db	—	Email thread analysis.
persons_registry.json	—	1,600+ persons merged from 9 sources with aliases and categories.

---

Data Exports

Knowledge Graph

File	Records	Description
knowledge_graph_entities.json	606	Curated entities: people, shell companies, organizations, properties, aircraft, locations.
knowledge_graph_relationships.json	2,302	Typed relationships with weights, date ranges, and source/target entity names.

Person Registry

File	Records	Description
persons_registry.json	1,614	Unified person registry from 9 sources. Includes name, aliases, category, search terms.

Entity Extraction

File	Records	Description
extracted_entities_filtered.json	8,085	Filtered entities: 3,881 names (2+ documents), 2,238 phone numbers, 1,489 amounts, 357 emails, 116 organizations.
extracted_names_multi_doc.csv	3,881	Names appearing in multiple documents with counts and sample EFTA references.

Image Catalog

File	Records	Description
image_catalog.csv.gz	38,955	Complete image catalog with AI descriptions, people identified, objects, settings.

Document Summaries

File	Records	Description
document_summary.csv.gz	519,438	Per-document redaction summary for every EFTA document.

Reconstructed Pages

File	Records	Description
reconstructed_pages_high_interest.json.gz	39,588	Pages where hidden text was recovered from under redactions.

---

DOJ Document Audit

The doj_audit/ directory documents a comprehensive audit of the DOJ Epstein Library conducted in February-March 2026.

The audit identified ~64,000 documents that were temporarily removed from the DOJ website after the initial public release, as well as ~24,000 documents with post-release size changes suggesting modification. As of March 2026, previously removed documents have been restored to justice.gov. The audit data is preserved for the historical record.

File	Records	Description
doj_audit/CONFIRMED_REMOVED.csv	67,784	Documents confirmed removed at time of audit (since restored).
doj_audit/FLAGGED_documents.csv	96,112	All flagged documents with DOJ URLs and dataset info.
doj_audit/FLAGGED_documents_details.csv	102,223	Detailed metadata including status, category, document type, priority score.
doj_audit/SIZE_MISMATCHES.csv	23,989	Documents with file size changes between original and current DOJ versions.

Full reports: Document Removal Audit | Alteration Forensics

Alteration Analysis

The alteration_analysis/ directory contains analysis of documents where content was modified between versions.

File	Records	Description
alteration_analysis/classified_alterations.csv	21,803	Classified alteration types with sensitivity ratings and LLM reasoning.
alteration_analysis/removed_entities_export.csv	146,209	Entities (names, accounts, phone numbers) removed between document versions.

---

EFTA Numbering and DOJ URLs

EFTA numbers are assigned per page, not per document. A 10-page document consumes 10 consecutive EFTA numbers.

URL pattern: https://www.justice.gov/epstein/files/DataSet%20{N}/EFTA{XXXXXXXX}.pdf

Dataset	EFTA Start	EFTA End	Notes
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	Largest (103 GB)
10	01262782	02205654
11	02205655	02730264
12	02730265	02731783	Post-release expansion

No gaps between datasets 1-11 — every apparent gap is a multi-page document at a boundary. Dataset 12 has internal gaps (~100K unassigned numbers).

Also available as efta_dataset_mapping.csv and efta_dataset_mapping.json.

---

Processing Tools

The tools/ directory contains 34 Python scripts used to build all databases from raw PDFs. Use these to replicate or extend the analysis.

All tools auto-detect the data directory via EPSTEIN_DATA_DIR environment variable, or look relative to the script location.

Core Pipeline

Tool	Description
ingest_house_estate.py	Ingest House Oversight Estate documents (Concordance format)
ingest_spreadsheets.py	Ingest native XLS/XLSX/CSV into full_text_corpus.db
transcribe_media.py	GPU transcription with faster-whisper large-v3
redaction_detector_v2.py	Spatial redaction analysis: find black bars, extract hidden text
build_person_registry.py	Build unified person registry from 9 sources
build_knowledge_graph.py	Construct entity relationship graph
document_classifier.py	Classify document types
populate_evidence_db.py	Build evidence/receipts database

Search & Analysis

Tool	Description
person_search.py	FTS5 cross-reference search with co-occurrence and CSV export
congressional_scorer.py	Score documents by redacted-name density for reading room prioritization
name_search.py	Name-specific corpus search
search_judicial.py	Search corpus for federal judges
search_gov_officials.py	Search corpus for government officials

Prosecutorial Query Graph (PQG)

Tool	Description
pqg_00_extract_concordance.py	Extract subpoena concordance data
pqg_01_decompose_riders.py	Decompose subpoena rider clauses
pqg_02_match_returns.py	Match returns to subpoenas
pqg_03_score_fulfillment.py	Score clause fulfillment
pqg_04_build_graph.py	Build prosecutorial query graph
pqg_05_report.py	Generate PQG investigation reports

Data Integrity

Tool	Description
find_missing_efta.py	Gap detection across EFTA numbering
recover_missing_efta.py	Recover missing EFTAs from DOJ server or forensic carving
run_post_ingestion_pipeline.sh	Chain all post-ingestion steps

Recovered Corrupted PDFs

The recovered_corrupted_pdfs/ directory contains text forensically recovered from 5 corrupted PDFs through byte-level carving: EFTA00593870, EFTA00597207, EFTA00645624, EFTA01175426, EFTA01220934. See: Corrupted PDF Forensics

---

Raw Data Sources

Source	Contents
DOJ Epstein Library	Datasets 1-12 (individual PDFs)
Archive.org DS9	103.6 GB, largest single dataset
Archive.org DS11	25.6 GB, 267K PDFs
Archive.org DS1-5	First 5 datasets combined
House Oversight	Estate documents, DOJ-provided records

See COMMUNITY_PLATFORMS.md for a directory of 79+ community tools, mirrors, and analysis platforms.

---

Integration Notes

• EFTA numbers are the universal key. Every document in the DOJ release has one.
• Join on efta_number, not document_id, when linking across databases.
• FTS5 search: SELECT ... FROM pages_fts WHERE pages_fts MATCH 'term' — join on rowid.
• Entity efta_numbers arrays give cross-references: "this entity appears in these documents."
• Image filename format: EFTA{number}_p{page}_i{index}_{hash}.png.

License

Analysis of public government records released under the Epstein Files Transparency Act (Public Law 119-38). Underlying documents are U.S. government works. Structured data released under CC BY-NC-SA 4.0.

Contact

Questions or corrections? Use the feedback form at epstein-data.com.

Defective Redactions Analysis

NEW: Defective Redactions in DOJ Court Filings — Recovery of hidden text from 12,000+ court filing PDFs with faulty redactions (Dec 2025 - Feb 2026).

• Quick Start Guide — Copy/paste technique, download tools, case inventory
• Technical Analysis — Full vulnerability assessment and methodology
• Public Guide — Non-technical explanation for journalists and researchers

Key Findings: .3M+ property tax details, 6M+ in undisclosed loans, previously unknown financial entities, evidence destruction instructions — all recoverable from original Wayback Machine archives.