Epstein Files CSAM Label Audit: 39 Withholding Cards, 660 Visual Entries and the Unresolved Count
Snapshot
| Field | Details |
|---|---|
| Record | Administrative withholding-label audit of released Epstein records. |
| Scope | 39 located withholding cards in 13 source PDFs; research cutoff November 9, 2026. |
| Visual-entry tally | 660 annotated entries: 589 thumbnails, 40 images, 30 photos and one picture. |
| Other units | Three cards specify one document, one photo frame and four pages; one additional card has no quantity. |
| Primary sources | Complete 39-card ledger and 13 parent-document subtotals, with original identifiers and page locators. |
| Evidence limit | These are administrative annotations, not a verified count of unique files, victims, incidents or offenses. Withheld content was unavailable. |
| Content note | Discusses records concerning child sexual abuse material. No abuse imagery is reproduced and no child is identified. |
Overview
Evidence inventory and source guide
Content note: This article discusses administrative records concerning child sexual abuse material (CSAM). It does not reproduce abuse imagery, identify children, or invite readers to locate withheld material. Source links point to published administrative records; broader source PDFs can contain distressing investigative material.
The public Epstein files contain a countable record of material withheld under the label “CSAM NOT SCANNED.” EpsteinWiki’s November 9, 2026 label audit located 39 distinct cards in 13 PDFs. Their handwritten annotations enumerate 660 visual entries: 589 thumbnails, 40 images, 30 photos and one picture. Three additional cards specify one document, one photo frame and four pages. A fourth supplies no quantity.
The tally establishes what the public paperwork says. It does not establish 660 unique original files, 660 victims, 660 incidents or 660 judicially determined offenses. The withheld contents were not available for this review, and an image may recur as a thumbnail, print, copy or backup. Those limitations are essential to understanding both the audit and the much smaller investigative estimates appearing elsewhere in the released record.
This article converts the audit into a complete source guide: every located card, every parent-PDF subtotal, the separate CD review flags, the investigative chronology, the search method, file-integrity information and the questions the available records cannot answer. All 13 card-bearing public PDF copies were reacquired during article preparation; their page counts and SHA-256 hashes match the audit. The 39 card annotations were checked against those copies, and the arithmetic was recalculated.
Key Takeaways
- 39 cards, 13 PDFs, 660 annotated visual entries. These are separate units, not interchangeable totals.
- Thirty-five cards contain visual counts. Three contain other quantified units, and one has a blank quantity line.
- Twenty-eight cards are in Data Set 3, one in Data Set 4 and ten in Data Set 10. A data-set number identifies a release grouping, not a separate investigation or victim group.
- Eight review sheets contain 23 flagged CD-label occurrences. One uses CSAM without a qualifier; 22 use possible or suspected language. They cannot be added to the card total.
- The official figures measure different collections. Reviewed files, responsive files, downloaded material, published images and withholding-card entries require different denominators.
- The missing step is a file-level reconciliation. Public counts alone do not resolve duplication, final classifications, connections between collections or completed victim-identification referrals.
What is being counted
A withholding card is a published placeholder page. Its Bates identifier locates that page within the released record. The PDF containing it may have a different starting identifier. For example, the card EFTA00004156 appears on PDF page 52 of EFTA00004105. Counting the card and its parent PDF as two independent withholding events would inflate the result.
The audit uses visual entry as an accounting term for the quantities expressly annotated as thumbnails, images, photos or a picture. It does not convert every withheld unit into an image. A document can contain text; a page can contain several images or none; a photo frame is a container description whose image count is not supplied. The blank card is recorded as unknown, never zero.
| Unit | Quantity | Treatment |
|---|---|---|
| Thumbnails | 589 | Included in the visual-entry subtotal |
| Images | 40 | 33 plus 7 |
| Photos | 30 | 29 in EFTA01682933 plus one in EFTA01722953 |
| Picture | 1 | EFTA00004878 |
| Visual entries | 660 | 589 + 40 + 30 + 1 |
| Document | 1 | EFTA00005564; separate unit |
| Photo frame | 1 | EFTA01697410; image count unspecified |
| Pages | 4 | EFTA01713390; image count unspecified |
| Unquantified card | 1 | EFTA01729433; no quantity entered |
These distinctions also explain why “at least 660 unique images” would be an unsupported description. Without deduplication, an entry count cannot supply a lower bound on the number of distinct original images. Conversely, the 660 does not cap the amount of withheld material: some cards use other units, one is blank, and additional labels may be absent from searchable text.
Complete 39-card ledger
PDF positions are one-based. The card identifier is the Bates stamp printed on the placeholder; the source link opens its parent document at the relevant page. Handwriting is transcribed with its original unit. The audit’s report reproduces these administrative cards on pages 9–47.
| Card Bates identifier | Annotation | Source and PDF page | Data set |
|---|---|---|---|
| EFTA00004156 | 33 IMAGES | EFTA00004105, p. 52 | 3 |
| EFTA00004283 | 2 Thumbnails | EFTA00004250, p. 34 | 3 |
| EFTA00004300 | 33 Thumbnails | EFTA00004250, p. 51 | 3 |
| EFTA00004303 | 51 Thumbnails | EFTA00004250, p. 54 | 3 |
| EFTA00004308 | 13 Thumbnails | EFTA00004250, p. 59 | 3 |
| EFTA00004309 | 24 Thumbnails | EFTA00004250, p. 60 | 3 |
| EFTA00004330 | 100 Thumbnails | EFTA00004250, p. 81 | 3 |
| EFTA00004335 | 81 Thumbnails | EFTA00004250, p. 86 | 3 |
| EFTA00004338 | 15 Thumbnails | EFTA00004250, p. 89 | 3 |
| EFTA00004530 | 2 Thumbnails | EFTA00004477, p. 54 | 3 |
| EFTA00004819 | 5 Thumbnails | EFTA00004800, p. 20 | 3 |
| EFTA00004878 | 1 Picture | EFTA00004800, p. 79 | 3 |
| EFTA00004919 | 25 Thumbnails | EFTA00004898, p. 22 | 3 |
| EFTA00004970 | 27 Thumbnails | EFTA00004898, p. 73 | 3 |
| EFTA00005285 | 32 Thumbnails | EFTA00005284, p. 2 | 3 |
| EFTA00005294 | 4 Thumbnails | EFTA00005284, p. 11 | 3 |
| EFTA00005297 | 4 Thumbnails | EFTA00005284, p. 14 | 3 |
| EFTA00005300 | 18 Thumbnails | EFTA00005284, p. 17 | 3 |
| EFTA00005345 | 54 Thumbnails | EFTA00005284, p. 62 | 3 |
| EFTA00005346 | 12 Thumbnails | EFTA00005284, p. 63 | 3 |
| EFTA00005357 | 2 Thumbnails | EFTA00005284, p. 74 | 3 |
| EFTA00005360 | 54 Thumbnails | EFTA00005284, p. 77 | 3 |
| EFTA00005361 | 16 Thumbnails | EFTA00005284, p. 78 | 3 |
| EFTA00005368 | 1 Thumbnail | EFTA00005284, p. 85 | 3 |
| EFTA00005377 | 7 Thumbnails | EFTA00005284, p. 94 | 3 |
| EFTA00005380 | 3 Thumbnails | EFTA00005284, p. 97 | 3 |
| EFTA00005564 | 1 Document | EFTA00005563, p. 2 | 3 |
| EFTA00005565 | 4 Thumbnails | EFTA00005563, p. 3 | 3 |
| EFTA00006039 | 7 images | EFTA00006039, p. 1 | 4 |
| EFTA01682991 | 2 Photos | EFTA01682933, p. 59 | 10 |
| EFTA01683001 | 6 Photos | EFTA01682933, p. 69 | 10 |
| EFTA01683002 | 6 Photos | EFTA01682933, p. 70 | 10 |
| EFTA01683003 | 12 Photos | EFTA01682933, p. 71 | 10 |
| EFTA01683004 | 2 Photos | EFTA01682933, p. 72 | 10 |
| EFTA01683006 | 1 Photo | EFTA01682933, p. 74 | 10 |
| EFTA01697410 | 1 Photo Frame | EFTA01697410, p. 1 | 10 |
| EFTA01713390 | 4 pages | EFTA01713378, p. 13 | 10 |
| EFTA01722953 | 1 photo | EFTA01722953, p. 1 | 10 |
| EFTA01729433 | No quantity entered | EFTA01729394, p. 40 | 10 |
The previously unresolved cards EFTA00004878 and EFTA00005564 read “1 Picture” and “1 Document,” respectively. Only the former belongs in the visual subtotal. EFTA01729433 visibly leaves the quantity line blank.
The 13 source PDFs
The card-bearing PDFs contain 893 physical pages in total. That is a source-container page count, not 893 cards or images. The following subtotals provide an independent check on the ledger.
| Source PDF | Data set | Cards | Visual entries | Other units |
|---|---|---|---|---|
| EFTA00004105 | 3 | 1 | 33 | None |
| EFTA00004250 | 3 | 8 | 319 | None |
| EFTA00004477 | 3 | 1 | 2 | None |
| EFTA00004800 | 3 | 2 | 6 | None |
| EFTA00004898 | 3 | 2 | 52 | None |
| EFTA00005284 | 3 | 12 | 207 | None |
| EFTA00005563 | 3 | 2 | 4 | 1 Document |
| EFTA00006039 | 4 | 1 | 7 | None |
| EFTA01682933 | 10 | 6 | 29 | None |
| EFTA01697410 | 10 | 1 | 0 | 1 Photo Frame |
| EFTA01713378 | 10 | 1 | 0 | 4 pages |
| EFTA01722953 | 10 | 1 | 1 | None |
| EFTA01729394 | 10 | 1 | 0 | No quantity entered |
The Data Set 3 cards account for 623 visual entries, Data Set 4 for seven, and Data Set 10 for 30. The ten Data Set 10 cards include six photo-count cards within EFTA01682933, one photo-count card at EFTA01722953, the photo frame, the four pages and the blank card.
CD review flags: 23 occurrences, not 23 proven unique discs
The eight media review sheets describe classification and handling decisions for CDs. Their terminology differs from the explicit withholding-card label. “Possible” and “suspected” preserve uncertainty; neither should be silently converted into a final forensic classification. Several sheets also request that mixed content be split, showing why a disc-level flag cannot establish the status of every file on that disc.
| Review sheet | Local CD labels | Occurrences | Meaning |
|---|---|---|---|
| EFTA00244892, p. 1 | CD1, CD2, CD3 | 3 | A few suspected photo items; instructions to split the media. |
| EFTA00244893, p. 1 | CD1, CD2, CD3 | 3 | Possible CSAM. Repeats the local labels on the preceding sheet; do not assume three additional physical discs. |
| EFTA00246249, p. 1 | CD14, CD17, CD20, CD21 | 4 | Suspected mixed content. CD14 requires further discussion; other entries call for splitting. |
| EFTA00247011, p. 1 | CD3, CD4, CD5 | 3 | Suspected content requiring splitting. Source image reads CD5; extracted text can misread it as CDS. |
| EFTA00247084, p. 1 | CD2 | 1 | Only two pictures described as suspected CSAM; splitting requested. Their overlap with card entries is unknown. |
| EFTA00276492, p. 1 | CD02, CD03, CD08, CD10 | 4 | Possible CSAM. |
| EFTA00278453, p. 1 | CD14, CD17 | 2 | CD14: CSAM. CD17: possible CSAM. Other damaged or partially extracted discs on the sheet are not automatically CSAM entries. |
| EFTA00279288, p. 1 | CD18, CD22, CD23 | 3 | Possible CSAM. Other entries described as image files or a blank CD are excluded. |
The arithmetic is 3 + 3 + 4 + 3 + 1 + 4 + 2 + 3 = 23 occurrences. Local numbering, repeated sheets and unknown file overlap prevent translating that into 23 unique discs. The two suspected pictures on EFTA00247084 are already part of this separate review-sheet inventory; adding them to 660 would assume an independence the records do not establish.
Drives, restored backups and evidence item 1B136
EFTA00161465 and EFTA00172473 describe two Samsung drives containing possible CSAM among media delivered to the Office of the Deputy Attorney General. They describe the same transfer, not two separate pairs of drives. Two drives is a container count, not an image count.
The March 26, 2025 dashboard in EFTA00164853 identifies evidence item 1B136 as an 85 GB collection with 33,985 files and says it includes CSAM. The total describes the review population. It does not classify all 33,985 files as CSAM. Separate entries concerning restoration of 1B144 and 1B145 describe backup activity; storage capacity does not reveal how many images, duplicates or classified items those backups contain.
EFTA01649116 addresses restored archived evidence and handling of 1B136. The March 17, 2025 discussion in EFTA01649100 distinguishes CSAM from the separate category described in the record as child erotica and seeks consistent media-redaction instructions. These records help explain the administrative setting. They do not supply another compatible image subtotal.
Read the supporting administrative PDF
EFTA00172473–EFTA00172480: Epstein Overview: FBI Response, eight pages. This is the processing and delivery record cited above, not the 39-card ledger or the withheld material.
If the PDF does not appear, open the PDF in a new tab or download the PDF.
Investigative counts and review history
The chronology below preserves the date, collection and unit attached to each figure. A small investigative count and a larger placeholder-entry count can coexist when they refer to different populations or processing stages. The available records do not provide the crosswalk needed to decide exactly how these populations overlap.
| Date and record | Statement | Counting limitation |
|---|---|---|
| August 2021 EFTA00159835, p. 4 | A trial-preparation update for August 16–20 reports approximately 38,934 images reviewed, several categorized as CSAM, further review and an intended NCMEC referral. | 38,934 is the review population. “Several” is not an exact subtotal, and a planned referral is not proof of completion. |
| March 9, 2023 EFTA00162440, p. 1 | About a dozen CSAM images from a known series. | Approximate investigative count; not a count of all later withholding entries. |
| July 2024 EFTA00038617, p. 2 | CID summary approved July 17 describes a small number of CSAM images on one device. | No exact number supplied. |
| March 2025 EFTA00164742, p. 2 | A March 20 email forwarded March 21 quotes a March 11 prosecutor response describing approximately a dozen photographs on one device as internet-obtained, rather than produced by Epstein. | This is the response’s characterization, not an independent finding about every withheld item. |
| July 10, 2025 EFTA00164897 | Approximately 15–20 commercial CSAM images; more than one million images and videos extracted; approximately 34,000 marked responsive; no CVIP submitted during the investigation. | Only 15–20 is the CSAM-specific estimate. Extracted and responsive populations are separate denominators. |
| July 2025 DOJ/FBI memorandum | Over ten thousand downloaded videos and images in a combined category of CSAM and other pornography. | The memorandum does not divide the categories and cannot establish a CSAM-only count above 10,000. |
| January 30, 2026 DOJ production letter, p. 2 | 180,000 images and more than 2,000 videos included in the publication. | Publication totals, not CSAM totals. |
| May 29, 2026 Bondi interview, pp. 66–67 | Discussion of withheld CSAM and downloaded material. | No numerical inventory in this passage. |
Victim identification and the referral question
CVIP is the Child Victim Identification Program. NCMEC describes victim identification as a process that helps investigators identify children depicted in abuse material. The 2021 intention to make a referral and the July 2025 statement that no CVIP was submitted warrant comparison, but their coexistence does not establish that a promised submission occurred, that every possible referral route was covered, or that particular survivors were or were not notified.
A defensible reconciliation would identify the collection, submitting agency, date, referral identifier and outcome without exposing survivor identities. The published passages do not supply that complete chain.
The 2023 estate video review
EFTA00039001 records April 19, 2023 correspondence about approximately 1,100 videos being reviewed to meet civil document-production obligations. It discusses a potentially problematic video and a request for FBI guidance. May 18 correspondence in EFTA00038997 reports additional videos flagged as potential CSAM. Neither record supplies a final confirmed CSAM count for the full collection.
Related correspondence appears under EFTA00039004, EFTA00039008, EFTA00039011, EFTA00039014, EFTA00039017 and EFTA00039019. The audit identifies these as overlapping correspondence records. Multiple copies of an email chain cannot be counted as separate discoveries or separate videos.
The audit also cites the review order in Government of the United States Virgin Islands v. JPMorgan Chase Bank, ECF 112 as procedural context. A process for reviewing potentially relevant media is not a final classification of that media. The approximately 1,100-video population therefore remains outside the 660-entry card total.
Earlier do-not-scan correspondence
The audit’s broader search identifies eight correspondence records discussing DocLab kicksheets marked “possible child pornography” with instructions not to scan: EFTA00013245, EFTA00013261, EFTA00015211, EFTA00073201, EFTA00101087, EFTA00147405, EFTA00147522 and EFTA00147819. The historical terminology is retained here only to identify the records and search method.
EFTA00147819 and EFTA00101087 preserve the 2019–2020 scanning correspondence and questions about locating excluded material. They help explain why absence from a scanned production does not prove absence from the evidence originally collected. They do not enumerate another eight images or establish a one-to-one match with the 39 later cards.
An earlier OPR account reproduced in Maxwell filing 204-3, PDF p. 72 is the audit’s locator for prosecutor Marie Villafaña’s assessment that investigators had not found photographs they characterized as child pornography in the earlier home-photo review. The underlying OPR report’s indexed text corroborates the attributed assessment; the Maxwell attachment’s exact PDF-page offset was not independently confirmed for this article. That historical assessment concerns a different review. It cannot substitute for classification of the later device collections, estate videos or withholding cards.
What the administrative label establishes
The cards prove the presence of a published administrative withholding label and its written annotation. They do not permit independent inspection of the withheld material, age determination, identification of a depicted person, reconstruction of an offense or confirmation of the legal rationale applied to every item.
The January 4, 2026 DOJ Attorney Review Protocol illustrates why classification and counting must remain separate. Page 4 includes descriptions of child sexual abuse in statements and reports within its CSAM category. Page 7 includes directions to use the “Depict/contain CSAM” tag with full withholding when segregation is not possible or a partial redaction would still expose protected private information. A tagged record therefore is not automatically one image.
The protocol postdates the March 2025 revision printed on the cards. It cannot establish that each earlier placeholder was generated under identical rules. It supplies later processing context, not a retroactive explanation for every card.
The February 14, 2026 DOJ report to Congress identifies CSAM among withholding categories but does not reconcile the cards with device findings, review populations, estate media or duplicates. The audit’s count should be read at the level the evidence supports: a documented inventory of located labels and annotations.
Reconciling the reported 49-document search result
Patrick Duggan’s article reports 49 documents tagged with the withholding phrase. The supplied audit describes its own inspection of the underlying search index and the publisher’s download archive. These are the audit’s recorded search observations, not a claim that a fresh search today will return identical results.
According to the audit, its exact-phrase query returned 44 records: 29 single-page JP2 OCR records and 15 parent-document records representing eight distinct PDFs. The 29 page records correspond to the initial 29 Bates identifiers in Data Sets 3 and 4. The parent records included duplicate representations. Its unquoted query returned 95 results, including investigative emails, review instructions, unrelated references and an IT-system acronym.
The archive’s csam-catalogs folder contained EFTA00004250.pdf, EFTA00004477.pdf and EFTA00005284.pdf, already included in the card ledger. That check was at filename-manifest level; it should not be expanded into a claim that every archived file was independently authenticated.
The historical 49-result snapshot is unavailable in the evidence used here. The current-state reconciliation in the audit does not reconstruct that historical snapshot. Nor does the number 49 establish 49 distinct cards, 49 distinct PDFs or 49 additional images. The ten Data Set 10 cards were located by the audit’s broader corpus search, not by subtracting or adding search-engine result counts.
Research method, coverage and reproducibility
What the supplied audit searched
The November 9 report records a scan of 1,380,932 rows with 1,380,932 distinct EFTA document identifiers in a versioned derivative text corpus. It searched CSAM, the complete withholding phrase, possible and suspected variants, older child-pornography terminology and scanning instructions. It normalized whitespace, Markdown markers and HTML formatting and used a compact-character check for punctuation or merged-spacing variants.
The audit reports 116 records containing CSAM as a standalone term, 39 normalized full-phrase occurrences in 13 records, and 380 broader candidates combining CSAM and scanning-related phrases. Relevant scanning contexts resolved to the 13 card-bearing PDFs and eight overlapping older correspondence records. Training examples, other criminal cases, news bulletins and DOJ computer-menu references were excluded from Epstein media counts.
What was checked for this article
The complete 50-page supplied report was read, including its 39 reproduced administrative cards and appendices. All 13 card-bearing public PDF copies were reacquired, their page counts and hashes compared with the report, and the 39 handwritten annotations visually checked. The eight CD sheets were acquired and their flags checked, including the CD5/CDS OCR error. Key investigative estimates, estate-review correspondence and official review/publication documents were checked separately against primary-document copies.
The 1.38-million-record corpus scan itself was not rerun for this article conversion. Its row counts, candidate counts and historical index-query results remain attributed to the supplied audit. That distinction preserves what was independently rechecked without implying a new full-corpus census.
Community transcription check
A targeted check of Epstein Forums found no indexed results for CSAM, the withholding phrase or the sampled source identifiers EFTA00164897 and EFTA00004250. The publicly accessible forum homepage exposed no forum listings, and its search endpoint was not retrievable through the research tool. No forum transcription was used, so no individual transcriber is credited. This is an access and search limitation, not evidence that the forum contains no relevant work.
What the method cannot cover
OCR can miss labels, especially faint handwriting, scans with no text layer, rotated pages and punctuation variants. A derivative corpus may omit documents, lag later releases or contain duplicate representations. A public search cannot retrieve wholly withheld files. The audit also conducted targeted checks of House, FBI Vault and court sources, not a separate bulk OCR census of every non-EFTA release.
The findings are therefore complete for the located 39-card ledger, not a guarantee that no other placeholder exists anywhere. A later revision should preserve the old ledger and record new card identifiers, sources, units and reasons for changing totals.
Corpus version and source-PDF integrity
The audit identifies the corpus as ishumilin/epstein-files-ocr-complete, revision 98225cc44979a3c84de2e0f1a0d9ea6f75b380c2. The recorded file is data/dataset.parquet, 1,414,456,548 bytes. Its reported SHA-256 is 49022b8dfa77f6ee4d2df318e3190d2434562af444c69187c6bd71cb9f53ad02.
The following hashes identify the inspected Roll Call public copies. All matched when reacquired for this article. A hash establishes byte identity with the audited copy; it does not authenticate the withheld contents, prove completeness of a government production or show that every possible mirror is identical.
| PDF and official locator | Pages | SHA-256 |
|---|---|---|
| EFTA00004105 DOJ locator | 52 | a4060afdde63a4a25516ef6e7fcfc48f24e3e9cd3fb17eb72f3767785a494f61 |
| EFTA00004250 DOJ locator | 98 | c2513e335c5c23c7bfecf26eb52b0fa5dc663b0805fd107281f69a4ef00a4881 |
| EFTA00004477 DOJ locator | 100 | 0f90a29d3801d2bbcb4c90d0de9abf021c1d48527325ffae11fc6bedb9c1322a |
| EFTA00004800 DOJ locator | 98 | ddc9f128ad266fe71536fa82e0657b3c9812dcf3c7d69826b074552a6048a948 |
| EFTA00004898 DOJ locator | 77 | cf6db5a36f5662bc09574ebf241a3f570d001310ed58461ddd194543cf4f9c19 |
| EFTA00005284 DOJ locator | 97 | bc5007bf24890c3e1de12a52f617fee71540eb53ed1e7b8bf27af6831f39a688 |
| EFTA00005563 DOJ locator | 4 | 62e1ec94357ac6cceb80a5b40b81c0dcd2740418ae896554ccf8e5906c9a86f0 |
| EFTA00006039 DOJ locator | 4 | 2c50f75a0278de62284ac89f6b67a5d9393f309ac74219884575bf8363167696 |
| EFTA01682933 DOJ locator | 74 | 410c25fbcae7f70c1c4460c4058cfdef9c3883a3d5496ca5680c17d23f5f713a |
| EFTA01697410 DOJ locator | 1 | c5263ac984109e12b39b38aabfceecd645beb4617f3892728da39d12c8ccab66 |
| EFTA01713378 DOJ locator | 100 | e7280c87832666ac8d594543f3b33c3eb96ddf170d794fadafb39a659e25f578 |
| EFTA01722953 DOJ locator | 2 | b94e0b7f0abfb3a3d362e970f38328eb3d562e449091653b784bb44d52c7eb64 |
| EFTA01729394 DOJ locator | 86 | 7761516ec1e658131823d1710f508488fe9f4b07d5511955a64bba01daa2fded |
Version history: The supplied full-corpus edition replaces provisional 622 and 629 tallies. It resolves the two formerly unclear initial annotations and adds ten Data Set 10 cards. Earlier totals should be identified as superseded rather than presented as separate findings.
Questions That Remain
- What does the blank card represent? EFTA01729433 requires an administrative quantity and content-type explanation. No inference from adjacent pages can replace that missing field.
- How many images, if any, are represented by the document, photo frame and four pages? An index or content-type inventory could answer without releasing protected images.
- Which entries are duplicates? File hashes, parent-child relationships, evidence-item identifiers and duplicate-group metadata could connect thumbnails, prints, discs, restored backups and original files.
- What became of the possible and suspected flags? Final review classifications, dates and reasons are needed for the CDs and estate-video review.
- How do the 660 annotated visual entries relate to the approximately 15–20 commercial images? A collection-by-collection crosswalk is needed before identifying a contradiction or aggregating the counts.
- Which victim-identification referrals were completed? Non-identifying submission and response records could resolve the difference between plans and completed steps.
- How complete is the searchable public record? A release manifest, withheld-record inventory and subsequent production history would clarify records outside the audited corpus.
- Which instructions governed each card? Dated processing guidance and batch metadata would show whether criteria changed across investigations and releases.
These questions concern administrative accountability. They do not require public access to abuse imagery, identifying information about children, private addresses or survivor names. Missing metadata should remain a gap, not a basis for assigning guilt or inferring the contents of a withheld file.
Conclusion
The located paperwork supports a reproducible tally of 39 withholding cards in 13 PDFs and 660 annotated visual entries. Separate CD flags and investigative estimates cannot be added to that tally without a documented relationship between collections. A unique-file total and the withheld material’s contents remain unestablished.
Questions and Answers
Does the audit establish 660 separate CSAM files?
No. It establishes 660 visual entries written on withholding cards. Distinct-file identity and overlap have not been resolved.
Can the CD flags or 1,100 videos be added to 660?
No. Disc labels and reviewed videos are different units, their classification status differs, and overlap with the cards is unknown.
Does the July 2025 memorandum establish more than 10,000 CSAM images?
No. Its figure combines CSAM with other pornography and combines images with videos. It supplies no CSAM-only subtotal.
Does a blank quantity line mean no material was withheld?
No. It means the public card provides no quantity.
Does the audit independently determine the age of anyone depicted?
No. It reviews labels and administrative records, not the withheld contents.
Why retain both official links and public copies?
Official locators show the release location. Public copies and hashes make the specific inspected version traceable. PDF page positions and Bates identifiers together reduce confusion when a parent filename differs from the page stamp.
What can responsibly be concluded?
The located public paperwork supports a precise administrative tally: 39 cards, 13 PDFs and 660 annotated visual entries, with other units and uncertainties explicitly separated. A unique-file total remains unestablished.
Sources and Related Reading
- FBI Redaction Training and Instructions: the broader processing and training record.
- DOJ Epstein Library: official release gateway.
- FBI Vault: Jeffrey Epstein: related FOIA releases, which are not identical to an EFTA corpus census.
- July 2025 DOJ/FBI memorandum.
- January 4, 2026 Attorney Review Protocol.
- January 30, 2026 production letter.
- February 14, 2026 report to Congress.
Source basis: EpsteinWiki, Epstein Files CSAM Audit: 39 Withholding Cards and 660 Visual Entries, updated November 9, 2026, 50 pages. This article preserves the report’s full located card ledger and unit distinctions while distinguishing newly rechecked sources from the report’s historical corpus and index-search observations.