That is new information! I wasnt even able to get that ‘no images produced’ page, good to know thank you. I just hit a file corruption error when I tried to dl from the DOJ. Thank you for the information. I guess this means the content is still missing in a way but at least accounted for.
Oh no…I didn’t know this, on one hand now i need to run another scan, but on the other it could reveal something, the torrent has 500k+ files so there is still a gap. I will run the scraper again and do a new analysis in the next day or two.
Follow-up to my Dataset 9 indexing post. I pulled the adjacent files from my local copy of the torrent. What I found is… notable.
The 3 missing files aren’t random corruption. They all cluster around one event: Epstein’s girlfriend Karyna Shuliak leaving St. Thomas (the island) in April 2016. And one of the gaps sits directly next to an email where Epstein recommends her a novel about a sympathetic pedophile—two days before the book was publicly released.
Two of the missing files (326497 and 534391) are the same document processed twice—once with redactions, once without—208,000 files apart in the index.
| Redacted Batch | Unredacted Batch | Content |
|---|---|---|
| 326494-326496 | 534388-534390 | AmEx travel booking, staff emails |
| 326497 - MISSING | 534391 - MISSING | ??? |
| 326498-326500 | — | Email chain continues |
| 326501 - MISSING | — | ??? |
| 326502-326506 | — | Reply + Invoice |
| — | 534392 | Epstein personal email |
Random file corruption hitting the same logical document in two separate processing runs, 208,000 positions apart? That’s not how corruption works. That’s how removal works.
I pulled everything around the gaps. It’s all one email chain from April 10, 2016:
The event: Karyna Shuliak (Epstein’s girlfriend) booked on Delta flight from Charlotte Amalie, St. Thomas → JFK on April 13, 2016.
St. Thomas is where you fly in/out to reach Little St. James. She was leaving the island.
The chain:
The unredacted batch (534xxx) reveals the email addresses that are blacked out in the redacted batch (326xxx):
The document immediately after missing file 534391:
From: "jeffrey E." <jeevacation@gmail.com>
To: Karyna Shuliak
Date: Sun, 10 Apr 2016 19:52:13 +0000
order http://softskull.com/dd-product/undone/
He’s telling her to buy a book. The same day she’s being booked to leave his island.
“Undone” by John Colapinto (Soft Skull Press)
On-sale date: April 12, 2016
Epstein’s email: April 10, 2016
He recommended it two days before public release.
Publisher’s description:
“Dez is a former lawyer and teacher—an ephebophile with a proclivity for teenage girls, hiding out in a trailer park with his latest conquest, Chloe. Having been in and out of courtrooms (and therapists’ offices) for a number of years, Dez is at odds with a society that persecutes him over his desires.”
The protagonist is a pedophile who resents society for judging him.
The author (John Colapinto) is a New Yorker staff writer, former Vanity Fair and Rolling Stone contributor. Exactly the media circles Epstein cultivated.
So now we know the context:
EFTA00326497 — Between AmEx confirmation and Groff’s forward. Probably the PDF ticket attachment referenced in the emails.
EFTA00326501 — Between the forward chain and Shuliak’s reply. Unknown.
EFTA00534391 — Immediately before Epstein’s personal email about the pedo book. Unknown, but its position is notable.
How did Epstein have this book before release? Advance copy? Knows the author?
What is 534391? It sits between staff logistics emails and Epstein’s direct correspondence. Another Epstein email? An attachment?
Are there other Shuliak travel records with similar gaps? Is April 2016 unique or part of a pattern?
What else is in the corpus from jeevacation@gmail.com?
Try the DOJ links (all return errors):
Check the torrent: Pull the EFTA numbers I listed. Confirm the gaps. Confirm the adjacencies.
Grep the corpus: Search for “QWURMO” (booking reference), “Shuliak”, “jeevacation”, “Colapinto”
Three files missing from 531,256. All three cluster around one girlfriend’s April 2016 departure from St. Thomas. Same gaps appear in two processing batches 208,000 files apart. One gap sits adjacent to Epstein personally recommending a novel about a sympathetic pedophile, sent before the book was even publicly available.
This isn’t random corruption.
Full analysis + all code: https://github.com/degenai/Dataset9
If anyone has the torrent and wants to grep for Colapinto connections or other Shuliak trips, please do. This is open source for a reason.
My question is, why is the total download size so large and the range of displayed documents so little? Only 15% of the known documents are individually served on the site, and some arent seen until page 10,000
Ok everyone, I have done a complete indexing of the first 13,000 pages of the DOJ Data Set 9.
KEY FINDING: 3 files are listed but INACCESSIBLE
These appear in DOJ pagination but return error pages - potential evidence of removal:
EFTA00326497
EFTA00326501
EFTA00534391
You can try them yourself (they all fail):
https://www.justice.gov/epstein/files/DataSet 9/EFTA00326497.pdf
The 86GB torrent is 7x more complete than DOJ website
DOJ website exposes: 77,766 files
Torrent contains: 531,256 files
Page Range Min EFTA Max EFTA New Files
0-499 EFTA00039025 EFTA00267311 21,842
500-999 EFTA00267314 EFTA00337032 18,983
1000-1499 EFTA00067524 EFTA00380774 14,396
1500-1999 EFTA00092963 EFTA00413050 2,709
2000-2499 EFTA00083599 EFTA00426736 4,432
2500-2999 EFTA00218527 EFTA00423620 4,515
3000-3499 EFTA00203975 EFTA00539216 2,692
3500-3999 EFTA00137295 EFTA00313715 329
4000-4499 EFTA00078217 EFTA00338754 706
4500-4999 EFTA00338134 EFTA00384534 2,825
5000-5499 EFTA00377742 EFTA00415182 1,353
5500-5999 EFTA00416356 EFTA00432673 1,214
6000-6499 EFTA00213187 EFTA00270156 501
6500-6999 EFTA00068280 EFTA00281003 554
7000-7499 EFTA00154989 EFTA00425720 106
7500-7999 (no new files - all wraps/redundant)
8000-8499 (no new files - all wraps/redundant)
8500-8999 EFTA00168409 EFTA00169291 10
9000-9499 EFTA00154873 EFTA00154974 35
9500-9999 EFTA00139661 EFTA00377759 324
10000-10499 EFTA00140897 EFTA01262781 240
10500-12999 (no new files - all wraps/redundant)
TOTAL UNIQUE FILES: 77,766
Pagination limit discovered: page 184,467,440,737,095,516 (2^64/100)
I searched random pages between 13k and this limit - NO new documents found. The pagination is an infinite loop. All work at: https://github.com/degenai/Dataset9
We definitely need a crowdsourced method for going through all the files. I am currently building a solo cytoscape tool to try out making an affiliation graph, but expanding this to be a tool for a community, with authorization to just allow whitelisted individuals work on it, that’s beyond my scope and I can’t volunteer to make such an important tool, but I am happy to offer my help building it. I can convert my existing tool to a prototype if anyone wants to collaborate with me on it. I am an amateur, but I will spend all the Cursor Credits on this.