Missing FBI 302s: A Gap Analysis of the EFTA Production

Executive Summary

The DOJ's Epstein Files Transparency Act production contains internal FBI indices that list more interview records than were actually published. By cross-referencing three key documents — the FBI's own serial log, the prosecution's disclosure index, and the interview master list — we can identify specific FD-302 interview reports that the government's records say exist but that do not appear in the published files.

This is not about documents that were uploaded and taken down. These appear to be records that were never collated into the production — despite being listed in the FBI's own internal indices within the same production.

For the broader context on how the DOJ production was assembled from multiple document streams — and quantitative evidence that 57-75% of reviewed pages were withheld across the two trackable numbering systems — see The Hidden Numbering System: Secondary Bates Stamps. The 3501.xxx serial numbers analyzed here correspond to System 6 in that report.

The Three Key Index Documents

1. Interview Master List — EFTA00016612

EFTA00016612 (Dataset 8, 4 pages)

A document titled "Interviews 136" listing 136 FBI interview serial numbers in the 3501.XXX format. These are FBI case file serials from the Epstein/Maxwell investigation. The serials run from 3501.002 through 3501.516, with a separate section labeled "Interview/Contact with third party" listing three additional entries (3501.24, 3501.44, 3501.45).

Full serial list (136 entries):

2. Prosecution Disclosure Index — EFTA00095751

EFTA00095751 (Dataset 9, 63 pages)

The prosecution's master disclosure index for United States v. Ghislaine Maxwell (20 Cr. 330), dated November 6, 2021. This document lists 423 unique FBI serials (3501.000 through 3501.546) with their individual sub-records — every photograph, interview note, FD-302, audio recording, and supporting document that was produced to the Maxwell defense team.

Each entry includes:
- Serial and sub-serial number (e.g., 3501.045-001)
- Date of the underlying record
- Description (e.g., "Interview 302," "Interview Notes," "Photograph")
- Date produced to defense

This is the Rosetta Stone — it tells us exactly what should exist for each serial.

3. FBI Master Serial Log — EFTA01684300

EFTA01684300 (Dataset 10, 166 pages)

The master FBI serial log for case SOD-NY-3027571 — the 2019 SDNY investigation. Lists approximately 701 serials using a different numbering system (integer serials 1–701 rather than the 3501.XXX format). Heavily redacted. Contains FD-302 entries, search warrants, pen registers, arrest records, CHS (confidential human source) intelligence reports, and NTOC tips.

Important: This log uses a different serial numbering system from EFTA00016612. The 3501.XXX system appears to be the FBI's case file management numbering, while EFTA01684300 uses a sequential serial numbering within the SDNY case.

Case Study: Serial 3501.045 — The Missing 302s

Serial 3501.045 is identified in the interview index as one of the 136 interview serials. The user identified this serial as relating to Victim 1 in the Trump accusation slides — a reference to the well-known photograph of Epstein and Donald Trump that was used in FBI victim identification.

What the Disclosure Index Says Should Exist

According to EFTA00095751 (page 7), serial 3501.045 has 15 sub-records:

That's 4 FD-302 interview reports, 3 sets of interview notes, 3 photographs, 2 FBI intake reports, 1 FBI report, 1 law enforcement report, and 1 license.

What Actually Exists in the EFTA Production

Searching the full text corpus for all documents referencing "3501.045":

7 of 15 sub-records found.

What Is Missing

8 of 15 sub-records are missing, including 3 of the 4 FD-302 interview reports and all 3 sets of interview notes.

The only published 302 for this victim (Interview One, EFTA01245620) contains the notable passage on page 6:

"[Administrative Note: [The victim] agreed to allow the Agents to take a photograph of the cropped image (attached electronically as a 1A attachment). Of note, the particular image sent to her by [friend] was recognized by Agents as a widely distributed photograph of JEFFREY EPSTEIN and current United States President DONALD TRUMP. The cropped image [the victim] provided to the Agents only included JEFFREY EPSTEIN.]"

The victim identified Epstein from this photograph as "JEFF," the man who sexually abused her beginning at approximately age 13 at Sea Pines Plantation, South Carolina.

Three additional FBI interviews of this same victim — dated June 7, August 20, and October 16, 2019 — are listed in the government's own index but do not appear in the published files.

Corpus-Wide Statistics

FD-302 Documents in the Production

Distribution by Investigation

Distribution by Dataset

The 136 vs. 419 Discrepancy

The interview index (EFTA00016612) lists 136 interview serials. The corpus contains 419 standalone FD-302 documents. These numbers are not directly comparable because:

1. Each serial can have multiple sub-records (e.g., 3501.045 has 4 FD-302s)
2. The 419 count includes 302s from multiple investigations (MCC death, Palm Beach, SDNY)
3. The 136-serial list appears to cover only the SDNY sex trafficking prosecution interviews

The prosecution disclosure index (EFTA00095751) lists 423 unique serials with potentially thousands of sub-records. The full gap analysis requires parsing all 63 pages of that index against the corpus.

The Disclosure Index: What It Reveals

EFTA00095751 is governed by Protective Order paragraphs 7, 8, 9, 10, 15, and 17 — it was filed under seal in the Maxwell case. Key observations:

Named Persons with Dedicated Serials

The following individuals have dedicated FBI serial files (names corrected from OCR):

Largest Victim Files (Names Redacted)

Production Waves

Document Types in the Index

The index catalogs these types of FBI records:
- FD-302 interview reports
- Interview notes (handwritten and typed)
- Interview recordings (audio, full tape, clip, side A/B)
- Video clips
- Grand jury transcripts
- Depositions
- Photographs and photo arrays
- DAVID (Florida DMV) records
- FBI intake reports and crisis intake forms
- FBI electronic communications (ECs)
- FBI memoranda
- Palm Beach PD incident reports, evidence tracking, property receipts
- MySpace pages (social media evidence)
- Birth certificates
- Letters (from attorneys including Gerald Lefcourt, Jay Lefkowitz, Ken Starr, Joe Whitley, Alexander Acosta, Stuart Mermelstein, Jack Goldberger, Jeffrey Herman)
- OIG (Office of Inspector General) reports
- Cherokee County Sheriff's Office reports
- Money transfer records
- Pro Tech Monitoring reports (sex offender monitoring)
- Text messages between witnesses and FBI agents
- DAG (Deputy Attorney General) submissions

The Master Serial Log: Intelligence Entries

EFTA01684300, while using a different numbering system, reveals several categories of FBI work product that warrant separate attention:

Confidential Human Source (CHS) Reports

NTOC (National Threat Operations Center) Tips

Foreign Assistance

Programmatic Corpus Search: All 136 Interview Serials

A Python script searched all 2,914,901 pages of the corpus for documents stamped with each of the 136 interview serial numbers. Runtime: ~10 minutes. Full results in fbi_302_gap_analysis.json.

High-Level Results

Serials with ZERO Documents in the Production

Top 30 Serials by Document Count

The "Expected vs. Found" Problem

The prosecution disclosure index (EFTA00095751) should tell us exactly how many sub-records exist per serial. However, the OCR quality of this 63-page scanned document is extremely poor — Tesseract produced garbled text where serial numbers and sub-serial numbers are frequently misread.

Example: For 3501.045, we manually verified 15 sub-records on page 7 of the disclosure index. The automated parser found only 3 sub-serials due to OCR errors. This means the programmatic "expected" counts are severe undercounts and cannot be relied upon for gap calculation.

What we CAN say: The corpus search found 6 actual documents for 3501.045, and our manual verification confirmed that 7 of 15 sub-records exist as EFTA documents (the 7th — EFTA01245632, the crisis intake — didn't have "3501.045" in its OCR text, so the script missed it). That means 8 of 15 are missing.

[UPDATE — March 5, 2026]: The DS12 expansion published the 3 missing Interview 302s for serial 3501.045 (EFTA02858481, EFTA02858491, EFTA02858495). The interview gap for this serial is now closed (10 of 15 sub-records published; 5 remaining items are agent notes and grand jury transcripts). This is the only victim serial for which the interview gap was addressed. See DS12 Expansion Analysis.

A reliable full gap analysis would require either:
1. Better OCR of EFTA00095751 (the original PDF is available at justice.gov)
2. Manual review of all 63 pages
3. A dedicated OCR pass with a more capable engine than Tesseract

Methodology

Data Sources

• Full text corpus database (full_text_corpus.db): 1,416,711 documents, 2,914,901 pages — FTS5 indexed
• All three index documents read in full from the database, page by page

Search Approach

• FD-302 identification: Documents with "FD-302" AND "FEDERAL BUREAU OF INVESTIGATION" on page 0
• Serial matching: LIKE queries for 3501.XXX patterns across all pages
• Sub-record verification: Cross-referencing disclosure index entries against EFTA documents stamped with matching serial numbers

Limitations

• The OCR quality of EFTA00095751 (prosecution disclosure index) is poor — it is a Tesseract extraction of a scanned PDF. Some serial numbers and names may be misread.
• EFTA01684300 (master serial log) is heavily redacted — most dates appear as ######## and many names are truncated.
• The 3501.XXX serial numbering system and the integer serial system in EFTA01684300 have not been fully mapped to each other.
• A complete gap analysis of all 423 serials and their sub-records against the EFTA production has not yet been performed. The 3501.045 case study demonstrates the approach.

What This Means

The Proven Gap (3501.045)

For the one serial we audited manually, more than half the records were originally missing. The FBI's own disclosure index says 15 sub-records were produced to the Maxwell defense team. Only 7 appeared in the initial EFTA production.

[UPDATE — March 5, 2026]: The DS12 expansion published the 3 missing Interview 302s, bringing the count to 10 of 15 sub-records published. The interview gap for this serial is now fully closed. The 5 remaining unpublished items are agent notes (3501.045-002/-004/-006) and grand jury transcripts. This is the only victim serial for which the interview gap was addressed — all other serials retain their original gaps.

The Scale of the Problem

If the 3501.045 gap rate (~53% missing) is representative, then across 136 interview serials with potentially thousands of sub-records, hundreds of FBI documents may be absent from the published files. The disclosure index alone runs 63 pages.

Three Categories of Missing Records

1. FD-302 Interview Reports: The core product of FBI interviews — the detailed written summaries of what witnesses and victims told the FBI. These are the most consequential missing documents.

2. Interview Notes: The raw handwritten or typed notes agents took during interviews. These sometimes contain details not in the formal 302.

3. Supporting Records: Law enforcement reports, licenses, photographs, audio recordings, and other materials attached to each interview serial.

Possible Explanations

• Redaction scope: Some records may have been withheld entirely (rather than redacted) due to victim privacy, ongoing investigations, or national security
• Production error: Records that were disclosed to Maxwell's defense but not properly ingested into the EFTA production pipeline
• Classified material: The master serial log contains S//NF (Sources/Methods — Not Releasable to Foreign Nationals) entries that would not be publishable
• Sealed records: Some materials may remain under seal by court order

The disclosure index explicitly notes it is "SUBJECT TO PROTECTIVE ORDER PARAGRAPHS 7, 8, 9, 10, 15, and 17" — which governs handling of sensitive materials in the Maxwell case.

Recommended Next Steps

1. Better OCR of EFTA00095751: The justice.gov PDF is available. A dedicated OCR pass (or manual transcription of the 63 pages) would allow a complete accounting of every expected sub-record
2. Deep-dive additional serials: Replicate the 3501.045 manual audit for the largest serials (3501.226, 3501.103, 3501.162, 3501.055) to establish the gap rate across multiple interview files
3. Cross-reference CHS intelligence reports: The master serial log references confidential human source reports on topics including Israeli foreign influence, Chabad, Masha Drokova, and classified Epstein material — are any of these in the published files?
4. FOIA request: The disclosure index provides specific serial numbers and dates that can be cited in a targeted FOIA request for the missing 302s
5. Congressional inquiry: The interview index (EFTA00016612) and disclosure index (EFTA00095751) together constitute a roadmap for oversight — the government's own records documenting what should exist

See Also

• The Hidden Numbering System: Secondary Bates Stamps — How the production was assembled from multiple document streams, and quantitative evidence that 57-75% of reviewed pages were withheld. The 3501.xxx serials analyzed here are System 6 in that report.
• Corpus Inventory — Complete evidence chain for the EFTA production.

All EFTA references link to documents in the DOJ's Epstein Files Transparency Act release at justice.gov/epstein. Database queries run against full_text_corpus.db (2,914,901 pages). This analysis relies on Claude Code running Opus 4.6, which can make mistakes.