Layer 1: Graphics state save (q)

Layer 2: INVISIBLE OCR TEXT LAYER (Text Rendering Mode 3)

Layer 3: SCANNED IMAGE (/Im0 Do) - rendered ON TOP of text layer

Layer 4: WHITE RECTANGLE (clip mask) + BLACK EFTA LABEL at bottom

Layer 5: End

Evidence from content streams:

• Stream 24 (171 bytes): White rectangle fill + hex-encoded "EFTA00000476" label

• Stream 25: End

• Stream 4 (34 bytes): Image rendering

• Stream 30 (142 bytes): White rectangle fill + hex-encoded "EFTA00001932" label

• Stream 31: End

Critical Detail: Text Rendering Mode 3

From the raw content streams:

3 Tr

PDF Text Rendering Mode 3 means the text is neither filled nor stroked - it is completely invisible. This is the standard method used by OCR software (such as ABBYY FineReader, Adobe Acrobat's OCR, or OmniPage) to create a "searchable" text layer behind a scanned image. The text exists only for search/copy functionality, not for visual display.

FINDING 2: OCR SIGNATURE PROOF

The text layers exhibit unmistakable OCR signatures:

Wildly Varying Font Sizes

Real PDF text typically uses 2-10 font sizes. Having 197-266 unique sizes means OCR software is assigning different sizes to each word to match the spatial dimensions detected in the scan.

Wildly Varying Horizontal Scaling (Tz)

Standard OCR Font Names

All four PDFs use identical non-embedded standard fonts:

FINDING 3: IMAGE ANALYSIS

Both PDFs contain a single embedded image per page

96 DPI is extremely low for OCR purposes (typical OCR requires 300+ DPI for good accuracy). This explains the garbled text output.

Images are DIFFERENT between versions

Pixel-level comparison shows the images were re-scanned or re-processed:

Despite the massive pixel-level difference, the images look visually similar to the human eye - this indicates re-scanning from the same physical document (or re-processing with different settings).

EFTA00001932 has an ADDITIONAL redaction in the current version">EFTA00001932 has an ADDITIONAL redaction in the current version

Visual comparison reveals the current (re-released) version of EFTA00001932 has a new black rectangle that does not exist in the December 19 original:

• Original: 1 black rectangle at image coordinates (73,105)-(165,222) - top-left area (likely covering the greeting/name)

• Current: 2 black rectangles - the original one PLUS a new one at approximately (370,524)-(403,676) - middle area of the letter

This new redaction is baked into the scanned image itself, not a PDF annotation overlay. It was applied by re-scanning or re-processing the physical document with an additional physical or digital redaction.

FINDING 4: WHAT IS THE "EXPOSED TEXT"?

EFTA00000476 (Financial Ledger)">EFTA00000476 (Financial Ledger)

The document is a photograph of a financial ledger lying on a manila envelope. The image shows:

• A spreadsheet/table with columns for dates, descriptions, and dollar amounts

• Black marker redactions covering certain cells in the physical document

• The document was photographed (not flatbed scanned) at low resolution

The "213 lines of exposed text" are OCR's attempt to read this low-resolution photograph:

04044 so 4,10y yentaYI ory a 4

Afaoutt W a Paso pew teoi 016.4

L290 /39 51 92100'0I

This is not "exposed hidden text" - it is garbled OCR of the visible printed content in the photograph, mangled by:

EFTA00001932 (Handwritten Victim Letter)">EFTA00001932 (Handwritten Victim Letter)

The document is a handwritten letter in blue cursive ink on decorative stationery paper with a cartoon owl design. The "47 lines of exposed text" are OCR's attempt to read cursive handwriting:

ear e.i-freA;

1- i.O,,-_)c ot( hac.i. a wonderi-iti 110ii-

dali SeaSO11.

Manual reconstruction suggests this reads approximately:

Dear [name],

I hope [you] had a wonderful holiday

season.

The letter is a victim thank-you letter to Epstein, expressing gratitude for:

• Holiday/Christmas celebrations

• Trips to Palm Beach, Las Vegas, Mexico, and an island

• Flying her sister out to visit

• Use of a Manhattan apartment

• Help seeing her mother

• "Pushing me to be at my best"

This is consistent with the well-documented grooming pattern where victims were conditioned to express gratitude for material benefits.

FINDING 5: NO BLACK RECTANGLE OVERLAY HIDING TEXT

Annotation Check

Drawing Object Check

The only drawing objects are white-filled page borders. No black rectangles exist as PDF drawing objects.

Where are the black rectangles?

The black rectangles visible in these documents are baked into the scanned images themselves. They are part of the pixel data of the embedded raster image. This means:

For EFTA00000476 (financial ledger):

• 143,650 near-black pixels in the image (17.1% of image area)

• Largest black region: 998x561 pixels (the table area with multiple column redactions)

• These are physical marker/tape redactions on the printed document, captured in the photograph

For EFTA00001932 (victim letter):

• Original: 1 black region at (73,105)-(165,222) = 92x117 pixels

• Current: Same region PLUS new region at (370,524)-(403,676) = 33x152 pixels

• The additional redaction in the re-release was applied to the image (re-scanned or digitally added to the raster)

FINDING 6: ORIGINAL vs. CURRENT COMPARISON

Text Layer Differences

The text layers differ because different OCR runs produce different results from different scans of the same physical document. The current versions were re-scanned and re-OCR'd, producing slightly different (but equally garbled) text.

Key differences for EFTA00001932:

• Original OCR produces 257 words; Current OCR produces 229 words

• The original has "Mexico", "Vegas", "Friends"; the current has "Arizona", "Christmas", "circus"

• Both are equally garbled attempts at reading the same handwriting

• The current version lost some text near the new redaction area

File Size Differences

The slight size differences are consistent with different compression of the re-scanned images and different OCR text content.

CONCLUSION

Answer to the Key Question

Assessment of the Viral "Poorly Redacted" Claim for These Specific Files

For EFTA00000476 and EFTA00001932 specifically, the claim that redacted text can be "exposed" by copying/pasting from behind black rectangles is overstated. The "exposed text" is the OCR engine's garbled attempt at reading visible (non-redacted) content from a low-resolution scan. It does not reveal information hidden by the redactions.

However, it is worth noting:

• EFTA00001932 did receive an additional physical redaction in the re-release, covering what appears to be a signature or name area. This confirms that DOJ recognized at least some content needed additional redaction.

• The OCR text, while garbled, does provide fragmentary readable content from the non-redacted portions (e.g., recognizable words like "Christmas," "Manhattan," "Mexico," "friends," "wonderful")

• Other documents in the collection may have different redaction methods - this analysis applies only to these two specific files

Root Cause of Garbled "Exposed Text"

The text appears garbled because:

APPENDIX: Technical Evidence

EFTA00000476 Original - OCR Layer (first 500 bytes)">Content Stream: EFTA00000476 Original - OCR Layer (first 500 bytes)

%WB0AiUxr

q

1 0.06 -0.06 1 17.43 -26.32 cm

BT

0 0 0 rg

0 0 0 RG

1 0 0 1 252.8 489 Tm

77.33 Tz

3 Tr/OPBaseFont0 8.33 Tf(\)1 )Tj

1 0 0 1 246.23 475.12 Tm

65.18 Tz/OPBaseFont0 15.62 Tf(4 )Tj

Key operators:

Content Stream: Image Rendering Layer

q

864 0 0 576.75 0 0 cm

/Im0 Do

Q

This renders the image at full page size, ON TOP of the invisible text layer.

EFTA00000476 Original)">Font Inventory (EFTA00000476 Original)

(9, 'n/a', 'Type1', 'Courier', 'OPBaseFont0', 'WinAnsiEncoding')

(10, 'n/a', 'Type1', 'Helvetica', 'OPBaseFont1', 'WinAnsiEncoding')

(11, 'n/a', 'Type1', 'Helvetica-Bold', 'OPBaseFont2', 'WinAnsiEncoding')

(12, 'n/a', 'Type1', 'Times-Roman', 'OPBaseFont3', 'WinAnsiEncoding')

(13, 'n/a', 'Type1', 'ArialMT', 'OPExtFont0', 'WinAnsiEncoding')