The DOJ Quietly Deleted 892 Epstein Documents. We Built a System to Prove It.

On an unknown date between 2024 and 2025, the Department of Justice removed 892 documents from its Epstein Files Transparency Act release. There was no press release. No Federal Register notice. No update to the EFTA portal. The files simply stopped resolving. HTTP 404. Gone.

We know this because independent researchers got there first.

Before the deletions, rodrigopolo computed SHA-256 cryptographic hashes for every file across all 12 EFTA datasets -- 1.38 million documents in total. Separately, beak2825 was monitoring the DOJ servers and logged every HTTP response change. Together, they preserved the receipts.

We ingested all of it. Every hash. Every deletion record. Every modification timestamp. And what we found raises questions that the DOJ has never publicly addressed.

The Numbers

What Was Deleted

The overwhelming majority of deletions -- 866 of 892, or 97% -- came from Dataset 9. The remaining 26 deletions were scattered across Datasets 1, 2, 3, 8, and 10.

The deleted documents span a wide range of EFTA identification numbers, from EFTA00045963 to EFTA01260086. This is not a contiguous block. The deletions were selective.

The DOJ's stated reason

The DOJ has indicated that Dataset 9 documents were removed because they contained Child Sexual Abuse Material (CSAM). If true, that removal is not only justified but legally required under federal law.

We are not disputing that justification. CSAM has no place in any public release, and we fully support its removal.

The questions that remain

But the stated reason raises its own questions:

Why were 32 documents in other datasets silently modified? If DS9 was pulled for CSAM, what explains content changes in Datasets 2, 3, 4, 5, 6, 8, and 12? These modifications were never announced or explained.

Why was the deletion so thorough? When we attempted to reconstruct what was in the deleted documents, we discovered something unusual. The deletions didn't just remove PDFs from the server. They removed:

• Every document title and summary
• Every person-to-document link
• Every OCR text extraction
• Every entity extraction record
• Every co-occurrence relationship

Out of 892 documents that once existed in a fully indexed, searchable database, only 24 fragmentary entity records survive. One mentions Ghislaine Maxwell. Others contain truncated names: "Jeff Pagl," "Laura Menningei," "Fogelman Lawre." The rest -- addresses, dates, an organization code.

That level of cleanup goes beyond pulling files from a web server. Someone systematically scrubbed every database table that referenced these documents.

Why was there no public announcement? The Epstein Files Transparency Act was passed specifically to ensure public access to these records. When the government removes nearly 900 documents from that release, the public has a right to know -- even if the reason is legitimate.

The Modifications Nobody Mentioned

Beyond the 892 deletions, 32 documents across 8 datasets had their content modified after initial publication:

Dataset 8 accounts for two-thirds of all modifications. These are not CSAM removals -- they are content changes to documents in datasets the DOJ has never flagged.

When a government agency modifies evidence documents after publication, without disclosure, it undermines the integrity of the entire release. Every researcher, journalist, and attorney working with these files needs to know whether the version they have matches the version that was originally published.

Now they can check.

How We Verify: The Integrity System

Today we're launching a document integrity verification system built on three independent data sources:

1. Independent Hash Verification

We ingested 1,380,911 SHA-256 hashes from rodrigopolo's repository, which independently computed cryptographic fingerprints for every file across all 12 EFTA datasets. A SHA-256 hash is a one-way mathematical function: if even a single byte of a document changes, the hash changes completely. This is the same technology used to verify software downloads, blockchain transactions, and legal evidence chains of custody.

Every document on this site now displays its hash verification status:

• Verified (green shield) -- the document's current content matches its independently computed hash
• Pending -- the document hasn't been verified yet
• Mismatch (red alert) -- the document's content has changed since the hash was computed
• Removed from DOJ (red banner) -- the document was deleted from government servers

2. Deletion and Modification Tracking

We ingested every deletion and modification event from beak2825's monitoring project, which tracked HTTP responses from the DOJ EFTA servers over time. Each event is logged with a timestamp, the affected document ID, the dataset, and the type of change.

Documents that were deleted by the DOJ display a prominent red banner explaining that the original was removed from government servers.

3. Continuous Monitoring

An automated integrity agent checks the DOJ servers twice daily, verifying a random sample of 50 documents. Once a week, it performs deep verification on 20 high-value documents -- downloading the full PDF and computing a fresh SHA-256 hash against our stored baseline. If any document disappears or changes, the system generates an alert.

The full methodology, dataset breakdowns, and real-time statistics are available on our Integrity Dashboard.

What You Can Do

Verify it yourself

Every hash in our system comes from a public GitHub repository. You can clone it, compute the hashes yourself, and compare them against the DOJ's current servers. The tools are open source. The math is deterministic. You don't have to trust us -- you can verify.

# Clone the hash repository
git clone https://github.com/rodrigopolo/epstein-doj-library-sha256
 
# Verify any document
sha256sum downloaded-document.pdf
# Compare against the hash in the corresponding VOL file

Download the torrent

The DS9 documents that were removed from DOJ servers still exist in a torrent maintained by the same researchers who computed the hashes. We are not hosting this torrent, but its existence means the deleted documents are not truly gone -- they are just no longer available from the government that was ordered to release them.

Watch the dashboard

The Integrity Dashboard updates in real time. If the DOJ deletes or modifies another document tomorrow, you'll see it there.

The Bigger Picture

The EFTA was passed because Congress recognized that the public had a right to these records. The DOJ published 3.5 million pages. And then, quietly, 892 of those pages disappeared.

Maybe every one of those 892 documents contained CSAM and the deletion was entirely justified. If so, the DOJ should say that publicly, with specificity, and explain the 32 modifications in other datasets while they're at it.

But the default posture of a government agency should never be to silently alter a mandated public release. When the same agency that gave Epstein a sweetheart plea deal in 2007 removes evidence from a transparency release without public notice, the burden falls on them to explain -- not on the public to discover it through independent cryptographic audits.

We built this system because accountability shouldn't depend on whether someone happens to be watching. Now someone always is.