An independent analysis of the largest document release in DOJ history
In late 2025, the Epstein Files Transparency Act (EFTA) compelled the Department of Justice to release approximately 3.5 million pages of documents related to Jeffrey Epstein — spanning FBI interviews, financial records, flight logs, legal correspondence, and internal communications across twelve separate datasets.
These documents reveal systematic failures: prosecutorial misconduct, institutional cover-ups, and an accountability gap that spans decades. This investigation exists to ensure those documents are not buried by their own volume — to read what was released, connect what was separated, and surface what was meant to stay hidden.
We are not a news organization. We are independent researchers building a structured, searchable, evidence-linked database from raw government disclosures — making it possible to trace connections, build timelines, and verify claims against primary sources at a scale that manual review cannot achieve.
Investigative Stories
Long-form reporting organized into four sections: The Cover-Up, The Network, Follow the Money, and The Operation. Each story is built on primary source documents with numbered citations you can verify.
Case Files
Structured investigation dossiers that track specific threads — from the 2007 non-prosecution agreement to financial networks. Each case file catalogs evidence, entities involved, and open questions still under investigation.
Entity Profiles
Detailed dossiers on individuals connected to the Epstein network, classified by evidence strength — not guilt. Each profile links to source documents, known connections, and timeline events.
Evidence Room
A public, searchable interface to the full document corpus. Full-text search across 1.37 million documents, an interactive network graph of entity connections, and a chronological timeline of events.
~3.5M
Pages Released
1.37M+
Documents Indexed
53
Entities Profiled
8
Stories Published
125+
Source Citations
44
Open Questions
AI helps us read 3.5 million pages. Humans make every judgment call.
This investigation uses Claude AI (Anthropic) as a research partner — not an autonomous decision-maker. AI assists with document analysis, research synthesis, and initial article drafting. The collaboration works across three layers:
Document Processing
AI-assisted classification, entity extraction, and redaction detection across the corpus. Documents are typed (FBI 302, email, financial record), severity-scored, and cross-referenced against the entity database. Every AI output is flagged for human review before it enters the database.
Archer: Document Review Copilot
When a human reviewer opens a document, Archer surfaces entity mentions, key quotes, and connections to other documents in the corpus. It highlights patterns the reviewer might miss across hundreds of pages — but the reviewer decides what matters and what gets recorded.
Detective: Research Assistant
A read-only AI assistant with access to the full database. It answers research questions, finds missed connections between entities, and surfaces patterns across thousands of documents. It can query but never modify the data.
Every article is reviewed, fact-checked, and edited by the named editor before publication. Every factual claim is traced to a specific EFTA document identified by Bates number. AI suggestions require explicit approval before any database change. Articles are marked with an “AI-Assisted” indicator in their metadata.
Entities are classified into six tiers based on evidence strength, not guilt. Presence in the database does not establish criminal conduct.
Direct Evidence
Convicted, charged, or named as abuser in forensically authenticated victim journals
Immunized
Named co-conspirators in 2007 NPA who received blanket immunity
Circumstantial
Documentary evidence of suspicious conduct without direct evidence
Associated
Documented contact without evidence of criminal awareness
Victim / Witness
Identified victims or witnesses who provided testimony
Peripheral
Household staff, pilots, prosecutors, defense attorneys, investigating officers
Redactions in DOJ documents are categorized into four types to identify potential misuse and prioritize review.
Category A: Victim Protection
Protects victim identity, privacy, and safety. These redactions are legitimate under EFTA and should be preserved.
Category B: Legal Privilege
Covers attorney-client privilege, grand jury material, and ongoing investigations. Legitimacy must be verified case-by-case.
Category C: Institutional Protection
Protects institutional reputation or conceals government misconduct. Often violates EFTA Section 2(b) prohibition on embarrassment-based redactions.
Category D: Perpetrator Protection
Conceals perpetrator identity or criminal conduct. No EFTA exemption permits this. Highest scrutiny required.
Derek Emsbach, Editor — responsible for editorial direction, fact-checking, and final review of all published content.
This platform was designed and built by Cyclops Digital LLC as a research tool for investigative journalists, researchers, and advocates seeking accountability.
Special thanks to the survivors who made their stories public, the investigative journalists who refused to look away, and the legislators who passed the EFTA to force transparency.