Nicole Junkermann

How she appears in these documents	Writing to Jeffrey Epstein in her own voice, and being written to by him, across the whole span of the record. Her sender name changes over the years from “Nicole Junkermann” to “Nicole Junkermann | Junkermanngroup” — a pipe on the sheets, which the extraction captures as a capital I; she signs “Nicole”123
Described in the released files as	”Nicole Junkermann; dynamic, smart and ambitious; is an impressive and successful entrepreneur and investor, fluent in six languages” — a World Economic Forum Young Global Leaders nomination form filed in Epstein’s papers for the 2007 nomination round, whose nominator block is an empty printed form4
Described by Epstein as	8 April 2009, introducing her to a third party: “my very good friend/old girlfriend. nicole junkerman,, german,, doing many things in china.. perfect combo for you”5
Span of the record	A courier billing entry of 18 February 2002 is the earliest document; the latest is an e-mail of 7 April 2019. The e-mail traffic itself begins on 17 March 2009. Measured over a population of 4,303 documents67
Earliest act recorded	18 February 2002: a FedEx shipment on Epstein’s office account from “J EPSTEIN”, 457 Madison Avenue, to “NICOLE JUNKERMANN” at a private address abroad, delivered 20 February 2002 at 11:49 and signed for8
What Epstein offered her	23 February 2013: “also why dont you consider working for/with me , organizing the worlds most intersintng poepl, you can invest alongsie, you can re structure house staff, be very helpful to me, and you might find it challenging.” Her whole reply is “Let s dicsuss it”9
The joint investment the documents record	Carbyne, formerly Reporty Homeland Security, in which she writes on 18 February 2019: “Carbyne invested 1.5 m (1m you and 0.5m me) net of fees is3.9m your share, Ehud carry1.5 m and my share 2m”10
What Epstein did for her	From December 2018 to February 2019 he read her China arbitration papers and answered with questions and drafting — “where is the co incorpprated. what is njf capital?” — and introduced her to a China specialist as “Nicole friend for 20 years. Has china interests”1112
Contact after the 2017 marriage	31 January 2019: her London office writes to Epstein — “My name is Luciano and I work for Nicole Brachetti Peretti in London” … “We have a parcel for you, but I wanted to check the correct address to send it to” — and she answers her own member of staff the same day, copying Epstein1314
Not to be confused with	Ugo Brachetti Peretti, a different Epstein correspondent who shares the married surname and is the subject of almost the whole "Brachetti Peretti" population. A sheet in the release, read at the image for this page, has him calling a man Epstein named “fernando” his brother; that the man is her husband is on the outside record and is on no sheet, and neither is any relation between her and Ugo Brachetti Peretti himself. His own correspondence is a subject in its own right and is not taken up here15

Nicole Junkermann is a German-born investor. Emails, courier bills and business papers released by the Justice Department under the Epstein Files Transparency Act record contact with Jeffrey Epstein running from a 2002 courier shipment to an e-mail of 7 April 2019, a span of seventeen years, and containing, in the later period, her own account of a disputed Chinese joint venture, a shared holding in Carbyne, the company previously called Reporty Homeland Security, and an invitation from Epstein to come and work for him that she answered with “Let s dicsuss it”.⁶⁷⁹

She is a living public figure and is named here in full. This page is about what the documents record her doing, and about what the release does and does not establish. Where the outside record is used, for her marriage, her public appointments and her later response, it is named as the outside record and is not attributed to any sheet.

A word about what this page does not do. The release contains a very large volume of personal correspondence between these two people. This page quotes from it only where the quotation carries a fact about an act: a decision, a sum, an introduction, an offer, a date. It does not characterise the relationship beyond what a quoted sheet says in its own words.

Three things to get right before reading anything else

First, Brachetti Peretti in this release is overwhelmingly a different person. A search for the phrase returns 153 pages in 122 documents. In those documents the forename printed immediately before the surname is Ugo 259 times across 103 documents, Ferdinando 8 times across 4, and Nicole twice, in two documents, both of them copies of the single January 2019 parcel thread quoted below. Ugo Brachetti Peretti corresponds with Epstein in his own right and from 2014, three years before this subject’s marriage. One sheet of that population is used here and no other: the February 2017 exchange in which he calls a man Epstein had named his brother, which is what the release, as against the newspapers, establishes about the two surnames.¹⁵

Second, Infront is mostly not the company. Her former company is Infront Sports & Media AG. The bare string Infront returns 124 pages in 115 documents, and 65 of those 115 are the extraction running the two words “in front” together: “plant infront of engineering”, “infront of 2 witnesses”, “infront of her house”. All 65 were opened. The remaining 50 are the company: 48 of them sit inside the name population used here, and two do not.¹⁶

Third, Carbyne and Magnum Photos each pull in someone else. Magnum Photos returns 66 pages in 50 documents and not one is in her name population; they are a Magnum photographer’s 2015 correspondence with Epstein’s assistant. Carbyne returns 85 pages in 50 documents, of which 24 are in her name population and 26 are not; the 26 belong to Ehud Barak’s side of the same company and are already covered at that page. The company is a real link between these people; most of the documents carrying its name are not hers.¹⁷

The population, and how it was built

The obvious search, "Nicole Junkermann", returns 3,676 pages in 3,419 documents. The surname alone, "Junkermann", returns 4,472 pages in 4,187 documents, which is 768 documents more. Neither is the whole record.⁶

Three further classes lie outside both. The figures that follow are increments over the union built so far, in the order given, and not counts of what each form returns outside the "Junkermann" population. The forms overlap, both with one another and with the one-n form, so a figure measured only against the base population counts some documents twice.

Starting from the 4,187 documents of the "Junkermann" population: the extraction spells the surname with one n in 171 documents, 66 of which contain no other spelling; all 66 were opened and read, and all 66 concern this subject. Epstein himself writes it that way, and so do FedEx billing sheets and Google Calendar alerts. Those 66 documents are the first increment.¹⁸ Six damaged spellings reachable by substitution then add 32 documents that neither of the first two forms had reached.¹⁹ The office domain token Junkermanngroup, which is a single word and therefore does not match a search for the surname, adds 11 more on top of those.²⁰ Three entity terms, NJF Capital, Winamax and Infront Sports, add 7 documents that carry no spelling of her name at all, including a four-sheet curriculum vitae.²¹

4,187 + 66 + 32 + 11 + 7 = 4,303, and the union is 4,303 documents. That the parts reconcile shows only that they partition the union without overlap; it is not a claim that all 4,303 are about her. The union is the set that had to be built to find out, the "Junkermann" population within it is the set that was classified, and named subsets were then read.⁶

Every document of the "Junkermann" population (4,187 of the 4,303, and not the whole union) was classified from the header block of its own first sheet, with whitespace flattened, with the sender label matched in eighteen forms, and under the header layouts this production uses. The result, over those 4,187 documents, and a measurement of the record and a parser together in the sense set out at ⁷ rather than a count of the record. Every choice the specification at ¹ leaves open is closed there in words except one, and the two cells that one choice moves are given below under both of its readings, the narrower test first:

class	documents	share
she is the sender	1,780	42.5%
she is addressed	2,011	48.0%
a header naming neither her as sender nor addressee	370 or 379	8.8% or 9.1%
no e-mail headers at all	26 or 17	0.6% or 0.4%

So 3,791 of 4,187 documents, 90.5%, are direct correspondence in one direction or the other, and that figure does not move with the one choice left open. The desk’s own 40-document sample had given 52% sender and 45% addressed; the whole-population figure is lower on the sender arm and higher on the addressee arm, and the combined direct-correspondence rate is 90.5% rather than 97%. The third-party documents are mostly Epstein’s own office traffic, his assistant’s scheduling sheets and calendar alerts, and the header-less documents are his calendars and his assistant’s schedule sheets, forensic device-extraction reports, a courier invoice, two profile documents of her and a translated press interview, with a World Economic Forum nomination form and eight further sheets joining them under the narrower of the two header tests. All of them are enumerated at ¹, and the authored ones among them are acts by somebody and are treated as such below.¹

2002 to 2004: the courier record

The oldest documents in the population are not e-mails. They are FedEx billing sheets for Epstein’s office account at 457 Madison Avenue, and they record parcels sent from that office to Nicole Junkermann seven years before the first surviving message. On every sheet read at the image for this page the direction is the same one: his office is in the sender column and is the payer, and she is in the recipient column.

The earliest is 18 February 2002. The sender line reads “J EPSTEIN / JEFFREY E EPSTEIN / 457 MADISON AVE / NEW YORK NY 10022-6843 US”; the recipient is “NICOLE JUNKERMANN” at a private address abroad. It was an international priority shipment, delivered on 20 February 2002 at 11:49 and signed for, with customs entered on 21 February 2002 and a declared value of USD 150.00.⁸

Seven months later, on 18 September 2002, Epstein’s office sent a ten-pound package by FedEx Standard Overnight to “NICOLE JUNKERMANN / HARVARD BUSINESS SCHOOL”. It was delivered at 10:00 the following morning and signed for; the charge was USD 22.83.²² Her own curriculum vitae, elsewhere in the release, lists “2002 Program for Management Development, Harvard, USA”.²¹

Three further shipments from that office to her are billed across December 2002 and January 2003, one of them picked up on Christmas Eve, recorded on the sheet as an undeliverable package and returned. On 26 July 2004 Epstein’s office sent her another envelope under the shipper reference “LESLEY”.²³

These sheets do not say what was in any of the parcels and this page does not suggest anything about their contents. What they establish is narrow and dated: that in 2002 Epstein’s office already held an address for her, used it, and was billed for it.

2009: “do yu mean nicole junkerman”

The earliest e-mail in the population dates from 17 March 2009.⁷ The earliest one that turns on her name is from September of that year, and it is an exchange about establishing who she was.

On 28 September 2009 a correspondent writing under the account name “Gmax”, a handle that a court record in this same release ties to Ghislaine Maxwell, wrote to Epstein at 1:21 p.m.:

Am in CA – that friend of yours whom I think lives in Paris was at the conference..she spoke to you last week..claims to be in charge of a hedge fund or something

Epstein answered at 19:16:28: “do yu mean nicole junkerman”. The reply came at 7:17 p.m.: “That’s her..I couldn’t remember her name”.²⁴ The identification of the handle rests on a court record, not on this sheet, and is footnoted rather than asserted here.²⁵

Nearly six months earlier, on 8 April 2009, Epstein had introduced her to a correspondent named in the thread as David Stern, in a message that sets out how he presented her:

i spoke to my very good friend/old girlfriend. nicole junkerman,, german,, doing many things in china.. perfect combo for you. call her [barred] she has a sports fund. - 300mm 37 years old she is moving to berlin. she is partner/girl of robert louis-dreyfuss

The telephone number in that sentence is blacked out on the sheet and is not supplied here; the bracketed marker is editorial and is not on the sheet. The reply came the next day from a correspondent whose sender name is barred on this copy, beneath whose signature the sheet prints a company block for Asia Gateway Ltd.: “Will meet Nicole for lunch in Munich on 22 April.”⁵

Epstein’s description checks against her own document. Her curriculum vitae in the release lists “2002 – 2005 Private equity manager for Robert Louis Dreyfus” and “2005 – 2012 Founder and Executive Partner & Investor, United in Sports, Luxembourg”, the sports fund that “has raised €242.2m”.²¹ That is a documented correspondence between what he wrote about her and what she wrote about herself. It is not evidence for his description of her as an “old girlfriend”, which is his characterisation and is quoted here as that.

2011 to 2015: Infront, the sale, and two-way traffic

By September 2011 the correspondence is routine and two-way. On 3 September 2011, under the subject “Re: Fw: Presse Release Infront”, Epstein wrote: “great news,,, come visit congatulations”. She replied the next morning: “I will let me close first, cash in the bank and then I come flying…”²⁶

On 14 September 2011 a four-message exchange in a single day runs: Epstein, “i understand that barclays in monaco, has some property, owned by laude ott in st bun / might be of Merest to me„ / when do I see you?”; her answer, “Done he will send me the contact details of the person in charge- you would like to get in contact with him directly or should it go through me?”; his “when is the dosing, approx?.. go through you”; and her reply, which begins “depends on antitrust have to see the schmocks trnr for my last bod meeting, they want me to stay on the board”.²

On 11 February 2015 Epstein wrote “any news?” She answered at 11:38 a.m.:

All ok

Infront was sold for over 1 billion

But tell me how are you?

He wrote back twenty-three minutes later: “infront? is that your former co?”³

That is the release’s plainest statement of what her business had been, and of what Epstein did not know about it. The string of his question returns 24 documents; two were opened and the other 22 were counted and not read.²⁷

2013: the offer

On 23 February 2013 at 09:02 p.m. Epstein wrote to her:

I have gates on wed, if you would like to join for part.. also why dont you consider working for/with me , organizing the worlds most intersintng poepl, you can invest alongsie, you can re structure house staff, be very helpful to me, and you might find it challenging.

Her entire reply, sent at 10:18:17 p.m. the same evening, is: “Let s dicsuss it”.⁹

No document read for this page records any employment of her by Epstein, any engagement letter, any payment to her from him, or any further discussion of that offer. The searches run to test that are named in the footnote, and the sets they returned were opened.²⁸

In the same period Epstein’s own office was chasing her. A Google Calendar alert of 5 February 2013 reads “Nicole Junkerman to join quant meetings?”; one of 14 February 2013, “do I need to reach out to Nicole Junkerman re quant meetings on the 21st?”; and one of 20 June 2013, “need to hear from Nicole Junkerman’s assistant!”.²⁹

That June the relationship ran the other way. On 20 June 2013 Epstein’s assistant Lesley Groff wrote to her: “Hello Nicole. How kind of you to help out here. Please let me know once you have coordinated the car and how I can get payment to you or the company.” The errand was a car between Rome and Berlin for two travellers whose names are barred on the sheets; Junkermann answered the same morning, “Hope is all well- beatrix will help to coordinate”, and on 21 June the arrangement was confirmed and paid: “Nicole paid - I’ll check with her and will get back to you.” Groff then offered to “have our accounting dept send her a reimbursement check”.³⁰

In February 2016 a message signed “Ehud Barak”, copied to Reporty’s chief executive, told her: “I hereby confirm to you, following a conversation with Amir Elichai, Reporty=CEO(cc’d), that we invite you to join the Advisory Board of Reporty once it=will be established”. That company later became Carbyne.³¹

2017 to 2019: the married name, Carbyne, and China

The outside record, and not the release, dates her marriage to Ferdinando Brachetti Peretti to 2017: an Italian newspaper report of March 2019 on the baptism of the couple’s daughter states that they had married two years earlier, in a ceremony attended only by the witnesses.³² The release records contact continuing after it, and the document that shows it is invisible to every search for her maiden name.

On 31 January 2019 a member of her London staff wrote to Epstein:

Good afternoon Mr Epstein

My name is Luciano and I work for Nicole Brachetti Peretti in London.

We have a parcel for you, but I wanted to check the correct address to send it to. We have the following on file

9 East 71st Street

between 5th and Madison

New York

USA

The address on file with her office is Epstein’s own Manhattan house. The message ends by asking about a telephone number for the courier; the number is blacked out on the sheet.¹³ Epstein answered “yes” at 11:59 UTC, a time fixed by the iPhone property-list date-received epoch printed on the second sheet of one copy of his reply, which decodes to 11:59:17 UTC and matches that copy’s own Sent: field to the minute. Her own sheet carries a bare Sent: Thursday, January 31, 2019 12:06 PM and no such epoch, so nothing printed on it fixes its zone; the two stamps are seven minutes apart only if both bare Sent: fields are on the clock the one epoch on this thread establishes, and that is an inference and not something the sheets settle. At 12:06 by her own sheet’s Sent: field she wrote to her member of staff, copying Epstein: “Luciano / Pls call his office ask for leslie”, then a line that is blacked out on the sheet, then “Thanks / Nicole / Sent from my iPhone”. The corpus text of that message shows nothing where the bar is.¹⁴

Three copies of that thread survive. Only two of them can be found by searching for the married name, because the third carries transmission damage through the middle of it.³³

Carbyne

Carbyne, formerly Reporty Homeland Security, is a shared holding. The release records her dealing with it directly with Ehud Barak and with Epstein. The company’s own story, Epstein’s investment in it and Barak’s role belong to ehud-barak and are not restated here.

On 11 June 2018 she wrote to Barak under the subject “News/ Questions”: “I have a couple of questions from my CFO, see attached. Can you let me know who should I address them to?” The attachment is named “Carbyne_questions_on_behalf of Nicole_Junkermann.docx”, and Barak forwarded her message to Epstein the same day with the covering word “FYI.”³⁴ Barak’s reply to her, later that day, sets out the state of the company and ends: “In regard to the Warrants, It has to be exercised before March 29 2019. No reason to do it now. We still have about nine months before action. I propose we will discuss the issue together with JE.”³⁵

On 15 February 2019 her chief financial officer sent her a warrant sensitivity table headed “Execution of warrants only calculation” with four valuation rows; she forwarded it to Epstein on 17 February, and he answered “Yes” and then, two and a half hours later, “and what is the profit separate from warrants on the same =aluation.?”³⁶ On 14 February he had already written: “yes. i am letting ehud do the warrants , I think you should do the same .”³⁷

On 18 February 2019 at 11:41 a.m. she set out her position in seven numbered points. The image of the sheet prints them as a numbered list:

1. You are my friend and I do not have a lot of them so let s try to find a fair solution for all of us
2. I am sorry you are in this situation as you are trying to help two friends
3. You are right I see it differently, you are a billionaire I am not and I am a mother you are not.. so I have to care about her future and I do not have a choice… and as you know China is difficult if not we would not have this conversation
4. So if I understand you right you want to help both friends?
5. Carbyne invested 1.5 m (1m you and 0.5m me) net of fees is3.9m your share, Ehud carry1.5 m and my share 2m
6. China is valued from FSB 1.2 to 1.5b if I can get at least 250m out of there
7. Can Ehud help?

Lets speak

Epstein replied at 15:56, in full: “If you recall months ago you said if you are doing it , of course I don’t like ir but will agree. I think as china has been a shock I understand if you want to change your mind.” Her answer, stamped 12:08 p.m. on the same sheet, goes to that first sentence: “Sorry but I recall that we agreed to speak about when the time comes and thats exactly what we are doing now / That has nothing to do with china”. His last word in the thread is “Ok”.¹⁰ The order of that exchange is taken from the nesting of the quoted blocks on the sheet, not from the face values of the stamps, which the sheet prints in three renderings: a top-level zoned Date: Mon, 18 Feb 2019 16:45:32 +0000, two Gmail-style in-thread attributions (“On Mon, Feb 18, 2019 at 11:41 AM” and “at 12:08 PM”) and one in the other client’s form (“On 18 Feb 2019, at 15:56”). Read as a single clock they would put her answer before the message it answers. No zone is assumed for any of them here.

China

The China matter runs through the same weeks. On 13 December 2018 Epstein introduced her to Robert Lawrence Kuhn with a one-line message: “Robert , Nicole friend for 20 years. Has china interests . I think you may be able to assist. See you on your return”.¹¹ She wrote to Kuhn on 17 December: “The pleasure is all mine, it’s good to be connected — and small world indeed! Do let me know when you are available in early January and I’ll be happy to meet you in NY.”³⁸

On 1 January 2019 Epstein wrote to her, in a message whose entire body is one line: “you didnt send me the chinese guys name”. She replied twice that afternoon, “We do not drink” and then “Tmr my lawyers office was closed”, and he answered “Then also send a detailed summary”.³⁹ No document read for this page records a reply to that request naming anyone, and the release does not establish which name he was asking for; the searches run to test that are named in the footnote.⁴⁰

The following day a name did reach him, in her papers rather than in an answer; the same name recurs four days later in her annotations to his 4 January draft, and never as a reply to the question he put. On 2 January 2019 she forwarded him a memorandum from her chief financial officer, and the forward carries beneath it a message of 31 December 2018 in which that officer had asked a Shanghai associate, “Can you please send me Mr Wu and Mrs Zhang passport copies”, and been told: “Mr. Wu and Ms. Zhang used their Chinese IDs in all documents where identity was required, so we could only get their ID numbers if needed. But the scanned copies of their ID are not accessible to the public.” Epstein’s reply the same evening is a list of questions:

where is the co incorpprated. what is njf capital? what happened in the past three years if you got no info? how did you find these people? how do you know the corp owns the asset.? is the jv in english or chinese ? are these party people? . what is the arbitration. ? who represents you. is this a party law firm ?

On 4 January he sent her a draft response: “my first draft response , is that arbitration , even if you win, , i m not sure you do. the fake document is a key if provable. but the chinese protect their own. my instict is that this needs finesse rather than a legal attack.”

Two days later she answered inside it, typing her own replies between his sentences. Through the interleaved passage on the copy at EFTA01009260 the two voices are distinguished by colour: his 4 January text prints in black and every one of her interpolated answers prints in red, so what follows is read off the image and not inferred. The colour separates the two voices inside that passage and nowhere else on the sheet: her own covering line above it, quoted at the end of this section, prints in black like his text.¹² Against his “did he claim you didnt produce what you promised, ?” she put: “he asked over the years from me personal guarantees which I did not provide, they claim we did not provide any help”. Against “he is not very powerful. but he is a local.”: “correct”. Against ”. unlikely a quick solution ,”: “I know but it is too large to leave it”. Against “was your former chinese partner involved in this mess?”: “I do not know”. Her fullest answer is responsive to his line about the disputed document but does not sit inside it: on the image the red run inside that sentence’s own parenthesis is a shorter one, and hers comes after his next sentence, “but the chinese protect their own.”, and before his “my instict is that this needs finesse rather than a legal attack.” It reads: “Yes we have a notary recording, and mobile recordings( mobile recordings in Chinese courts are valid as proof), Mr Wu is very emotional so He said too much, they went against the release of the recording..”.

Against “but heavy players needed. not likely kuhn.”, which was Epstein’s own view of what the China specialist he had put her in touch with three weeks earlier could do for her, she wrote:

I know but I am just a little remora, but when it comes to making powerful friends, nothing beats a shark..

Her own covering line, above the whole quoted block, is “Let me know when u want to speak”.¹² On 9 January she reported the outcome: “I have news / Arbitration results Wu claims were fully rejected.”⁴¹

Three weeks after his question of 1 January, Epstein sent her a description of his own. On 22 January 2019 at 00:59, under the subject line “do not pass on”, Epstein wrote to her in full:

Wu from Wenzhou (Zhejiang Province). super shady, no real high level party strength. but clever,sneaky. etc.

She replied at 9:45 a.m.: “Willi not / Thank you does not surprise me”.⁴² Wu is the counterparty in the arbitration she had reported on 9 January. Where Epstein got the description is not established by any document read for this page.

On 11 February 2019 he wrote: “should i stand down on really sport no valuation or news .? no problem for me, but it appears silly to chinese friend , that no info forthcoming”.⁴³ On 28 February 2019 he wrote: “china ? / i keep getting questions ? what woudl you like to say”. Her answer the next afternoon is a six-bullet intelligence summary on the counterparty and the asset: “FSB (Russians) valued the business between 1.5-2.5b (Euro)”, an “Official credit report form Chinese banking authority”, reported revenues set against “intelligence reports”, and “Mr Wu is owner/legal rep of more than 20 companies”. It closes: “I still believe that my target of 250m for my stake is in line with information I have at this moment. It would be great to hear from your contact what he believes is achievable and why as well as timeline.” Epstein’s reply, in one line: “the key is what entitisw generate what PROFITS and if you own any of them at all.”⁴⁴

The UK appointment: what the release holds, and what it does not

The outside record states that she was appointed in 2018 to the Healthtech Advisory Board of the Department of Health and Social Care, then led by Matt Hancock as Health Secretary of the United Kingdom,⁴⁵ and that after her correspondence with Epstein became public she lost two later positions: a Lancaster University visiting professorship, terminated, and a Royal Marsden Cancer Charity trusteeship, resigned.⁴⁶

The release holds one document about the appointment itself. Epstein’s files contain a five-sheet English translation of a Handelsblatt interview with her, datelined 27/11/2018 and bylined Peter Brors. Its second sheet puts the question:

Ms. Junkermann, the UK Department of Health, led by Matt Hancock, has just appointed you to the Scientific Advisory Board. Why?

and prints her answer:

I met the minister for the first time in the spring of this year, when he was still responsible for the technology department. After around two months, I had a letter from him saying that he wanted a direct link to young companies and science. The minister wants to find out more about how medical innovations can be introduced faster to the British healthcare system and thus into people’s everyday lives.

The same sheets carry her account of Winamax, of Infront and of buying into the Magnum photographic agency in 2017 as its first external investor.⁴⁷

Its evidentiary weight should be stated precisely. The document establishes that a press interview about the appointment was in Epstein’s papers by late 2018. It is not a record of the appointment, and no document read for this page records any communication between Epstein and any part of the UK government, or any role for him in the appointment. The searches run to test that are named in the footnote.⁴⁸

There is, separately, one exchange in which she asks Epstein about a health-technology board. On 3 November 2018 at 10:36 UTC, three weeks before the Handelsblatt interview, she sent him a link to a TED talk by the physician Ben Goldacre under the subject “What do you think about him?”, with one line of her own: “He will be my chairman for the advisory board for health& tech?” He answered at 11:10 UTC. Both times are taken from the Unix date-received epochs printed in the iPhone property-list trailer on the sheets themselves, which decode to 10:36:12 and 11:10:25 UTC and agree with each sheet’s own printed header to the minute; the zone is not assumed from any general rule. His reply:

i like his ideas he is not presidential enough to co=vey his ideas. as serious. he sounds more like a car sal=sman

(the = marks are transmission damage on the sheet and are shown as printed).⁴⁹

What the release establishes here, and what the outside record supplies. The two sheets establish the exchange itself: she sent him a talk by Ben Goldacre, asked whether Goldacre would be “my chairman for the advisory board for health& tech”, and he answered with an assessment of the man. No sheet read for this page states that the board in her 3 November message is the board she was appointed to. A sheet in the release does name such a board: the Handelsblatt interviewer’s question quoted above puts it to her that the Department “has just appointed you to the Scientific Advisory Board”. But nothing on that sheet or any other ties that board to the “advisory board for health& tech” of her own message, and the only sheets in the release that put Goldacre and a board in the same sentence are the two copies of her own 3 November message, where the board is unnamed and the chairmanship is a question she is putting to Epstein. The other half is on the public record and is cited to it and not to any sheet: the Department of Health and Social Care announced the Healthtech Advisory Board on 19 November 2018, sixteen days after this exchange, saying that “Dr Ben Goldacre will chair the board” and listing “Nicole Junkermann” among the members joining him.⁴⁵ Her Handelsblatt interview, dated eight days after that announcement, says the Department “has just appointed” her.

So the man whose TED talk she sent Epstein on 3 November 2018 is, on the government’s own press notice, the announced chair of the UK government board she was announced to sixteen days later, and she put him to Epstein, by name, before either appointment was announced. “Announced” is what the cited notice establishes and is all that is claimed here; whether either appointment was known to anyone before 19 November 2018 has not been searched for this page. The identification of Goldacre as that board’s chair rests on the gov.uk notice; the release itself neither makes the connection nor contradicts it.

On the departures the release establishes nothing at all. The last document in the whole population is dated 7 April 2019, four months before Epstein’s death and more than six years before the correspondence became public in the releases made under the Act. Nothing in the release can speak to what happened afterwards.⁷

Her response

A spokesperson for her issued a statement to The Tab, which printed it on 9 February 2026 and again on 11 February 2026 when it reported the end of her Lancaster University visiting professorship. That outlet is the party the statement was given to and not a re-reporter of it, and it is given here in full, in the three paragraphs that outlet prints, with its first sentence unrepaired:

The scale Jeffrey Epstein’s crimes is appalling. As a woman and a mother, Nicole was completely deceived and misled by him and deeply regrets their conversations regarding personal and professional matters.

These conversations took place at a time in her life when she was vulnerable, and she was manipulated into trusting him and consulting him with the most personal of matters. Nicole has been absolutely horrified to come to understand his true nature and the suffering he inflicted on women and girls through his abuse of trust, power and manipulation.

Her thoughts today are completely with his victims, and she looks forward to the day the real predators face justice.

Parts of the same statement are quoted in The Sunday Guardian of 16 February 2026, reporting her resignation that month as a trustee of the Royal Marsden Cancer Charity. Those two positions are the ones the outside record names as lost: the Lancaster visiting professorship, terminated, and the Royal Marsden trusteeship, resigned.⁴⁶

That is the outside record’s, not the release’s. No document read for this page carries any statement by her or on her behalf about Epstein’s offences, and the release, which stops in April 2019, could not contain one. It is set out here because a response belongs beside the allegation it answers.

What the extraction hides, on this record

This page turns up several faults in the production rather than in the subject, and most are set out where they arose: the courier sheets whose two address columns the extraction interleaves, at ²³; the colour that separates two voices on one sheet and is flattened into a single run of text, at ¹²; and the digit 4 captured as 9 through a run of 2014 headers, at ⁷. Two more are set out here in full, because each turns on a distinction the extraction erases: the first on what a search of her name can reach, the second on the difference between a blank and a bar. Both were settled at the image.

A four-sheet curriculum vitae is invisible to every search for her name, and not because the name is absent. The document sets out her career from Winamax in 1998 to the Ziggurat student-housing developments in 2012, and its letterhead prints “Nicole Junkermann” in light grey type at the head of every sheet. The extraction fails on that letterhead on all four sheets, dropping it altogether on the first and capturing it as unreadable noise on the others, so the corpus text carries the “NJ” monogram and nothing else of the head of the page. The document was reached through the entity terms Winamax and Infront Sports, not through her name.²¹

The Young Global Leaders nomination form in Epstein’s papers shows two different field states that the text layer renders identically. In the candidate block, the address, telephone, fax, e-mail and date-of-birth fields are barred, printed as black bars on the image. In the nominator block directly above, the first name, last name, position, organisation, address, telephone, fax and e-mail fields are empty printed form fields: labels with nothing after them and no bar. In the extracted text both blocks read the same way, which is to say they read like an unfilled form. Who nominated her is therefore not a redaction on this document; it is a blank.⁴

Related pages

• ehud-barak: Carbyne and Reporty, the SUM partnership and the warrants; the 26 Carbyne documents outside this subject’s name population belong there.
• lesley-groff: Epstein’s assistant, named in the January 2019 parcel thread and the reference line of a 2004 shipment.
• 9-east-71st-street: the address her office held on file for him.
• ugo-brachetti-peretti: the different Epstein correspondent who shares her married surname and to whom almost the whole "Brachetti Peretti" population belongs; that page reads the February 2017 sheet in which he calls a man Epstein named “fernando” his brother, which is the one document of his correspondence used here.
• ghislaine-maxwell: the court finding on which this page’s identification of the “Gmax” handle rests.
• jeffrey-epstein

Footnotes

1. All 4,187 documents of the "Junkermann" population — not the whole 4,303-document union — were classified from the header block of their own first sheet, never from the first sheet that matched. An earlier version of this footnote printed four point figures, named one hinge without computing what that hinge cost, and gave the “she is addressed” arm no printed definition at all, so that a reader implementing its words could not land on its table. The specification is therefore restated below with every choice closed in words but one, and every closure is priced: what it costs to close it the other way is given beside it. The specification. On each document’s own first sheet, with runs of spaces and tabs flattened to a single space and each line stripped: (1) A sender label is a line beginning with one of eighteen forms (From, Fran, Frorn, Frcm, Fram, Front, Fronl, Frnm, rom, Fnxn, Fromm, Fm, Frm, Fron, Erom, Frcrn, Ftom, Fiom), matched without regard to case and required to be followed by a colon, a space or the end of the line, so that Fromm is its own form and Tom is not a To. An addressee label is the same construction on To, Cc or Bcc. Both closures were run the other way: matching the sender label case-sensitively moves nothing at all, and confining the addressee set to To moves twelve documents out of the addressed class and into the third-party class. Bcc moves none. (2) The value of a label is whatever follows it on the same line after an optional colon; if nothing does, it is the next line as printed. Taking the next non-empty line instead moves nothing. Rule (2) is a limitation and not a choice, for the reason set out at ⁷; what the limitation costs is measured below. (3) A value names her when the string junkermann occurs inside it, without regard to case. This is why “Nicole Junkerman”, “Nicole Junkennann” and “junkerrnann” fall outside the classifier; the cost of that is measured at the end of this note. (4) A value that is itself a label line is not a value, and the label is read as carrying nothing. The label set for this rule is the one defined at (1), the eighteen sender forms and To, Cc, Bcc, plus Subject, matched by the same construction; and the rule is applied to the sender value only, not to addressee values. Both of those are choices and both are priced here rather than left to be inferred from the table. Subject is in the set because this production’s alert sheets print the subject line immediately below a withheld From, and dropping the set back to (1) alone moves one document: sender 1,780 to 1,781 and third-party 370 to 369 on the narrow reading and 379 to 378 on the wide. Adding Sent and Date to the set as well moves nothing at all, although it makes the rule apply far more often, to 187 first sheets rather than 48. Applying the rule symmetrically, to addressee values as well as to the sender value, costs two documents: addressed 2,011 to 2,009 and third-party 370 to 372 narrow, 379 to 381 wide, with the sender class unchanged. The rule is not a parser convenience but the handling a barred field requires. It applies to 48 of the 4,187 first sheets, and on ten of those the line it suppresses names her, which is where it moves the table. On each of the ten the extraction emits the sender label with nothing after it, and rule (2) reaches down to the next header line. On nine that line is a To line naming her: EFTA00357488, EFTA00357509, EFTA00357521, EFTA00357525, EFTA00454909, EFTA00499782, EFTA00542412, EFTA00542422 and EFTA00697681. On the tenth, EFTA02215101, a one-sheet alert, it is the subject line. All ten were rendered and read at the image, two at 400 dpi and eight at 200, and the empty field is a redaction on every one of them: each sheet prints the From value under a black bar, and on EFTA00697681, read again at 300 dpi for this note, the value is covered by three bars with white gaps between them rather than by a single continuous bar; on EFTA02215101 one bar covers the To and the From field together. The empty value in the text layer is the producing party’s withholding of the sender and not a blank field or an extraction fault. So nothing distinguishes the tenth from the nine, and rule (4) excludes all ten. Without rule (4) the sender class is 1,790; with it, 1,780. The 1,781 an earlier version of this footnote printed is what the set at (1) yields on its own, the nine removed and the tenth kept: a distinction that version did not state and that the image does not support. (5) The tests run in this order: sender, then addressee, then whether the sheet carries a header at all. A document is sender when the value of its first sender label names her; failing that, addressed when the value of any addressee label names her; failing that, third-party when the sheet carries a header; failing that, no headers. Running the header test before the addressee test moves nothing. Only the first sender label on a sheet is read, which is why any measurement of layout damage that counts later labels overstates the exposure to this table. (6) The one choice left open is what counts as “an e-mail header at all”, because both readings are defensible and no wording settles them. On the narrow reading a header is a sender or addressee label under (1) and nothing else; on the wide reading it is that set plus Sent, Subject and Date. Both were computed and the two cells the choice moves are printed as both. Each reading errs, in opposite directions: the wide one admits Date of Birth:, which is the only header-shaped line on the World Economic Forum nomination form at EFTA01113500, and so reads an authored form as e-mail traffic; the narrow one leaves in the header-less class three House Oversight e-mails that print their headers in layouts no line-initial label test can see, one as a pipe-delimited table and two as a single run-on line. Narrowing the set further, to sender labels alone, is closed by the words and not open, because an addressee label is an e-mail header as much as a sender label is, and it would cost thirteen documents. Twelve definitions of the header label set were implemented and the header-less count runs from 17 to 39 across them; the 33 the earlier version of this footnote printed is reached by none of them. The result: sender 1,780 · addressed 2,011 · third-party header 370 on the narrow reading and 379 on the wide · no headers at all 26 on the narrow reading and 17 on the wide. Both sum to 4,187, and that is worth nothing as a check: the four cells partition a fixed population, so every reading of every choice above sums to 4,187 too. A reconciling total proves partition, not membership. The parsed sender values date the change of style: of the 1,780, the commonest are “Nicole Junkermann” (813), “Nicole Junkermann I Junkermanngroup” (307, plus 283 more with a trailing bracket where the address follows). These are parsed strings from the text layer and are given as it emits them; the I is the extraction’s capture of the pipe the sheets print. Then a bare “junkermann” (107, plus 29 quoted) and “Nicole Junkermann <” (88). All six reproduce to the digit, and they were the one part of the earlier table that did. 1,120 of the 1,780 values were taken from the line below a bare label and 660 from the label’s own line. What rule (2) costs, measured where it bears. This production prints stacked headers in two shapes, and in the second, described at ⁷, the labels come as a block with their values as a block beneath, so that rule (2) takes the next label instead of the value. How many first sheets print the sender label that way has no single answer, and neither of the two figures the earlier version of this footnote printed, 152 sheets and 232 documents whose parsed sender value changes, is reached by any definition implemented here. Over five definitions the count runs 150 to 201 of 4,187, 3.6% to 4.8%: 187 where the line after the first sender label is another label, 150 where that next line is a label standing alone, 201 where the first sender label itself stands alone beside another such label, 103 where the next two lines are labels and 59 where the next three are. And a count of changed strings measures the wrong thing in any case, because rule (5) reads only the first sender label. What can move a cell is class change, and how many documents change class is not settled. Three implementations of the pairing, written independently of one another over the same 4,187 first sheets, do not agree. Four ways of pairing the label block with the value block gave 25 to 28 documents, 0.60% to 0.67%, putting sender at 1,767 to 1,770, addressed at 2,032 to 2,033 and third-party at 358 to 362. Twelve runs over six further definitions, varying the label set among the three used elsewhere in this note, varying the shortest run of bare labels that counts as a block between two and three, and running each with and without trimming the value block to the label block’s length (which moved nothing), gave five distinct answers, 3, 10, 13, 18 and 20 documents, 0.07% to 0.48%, putting sender at 1,766 to 1,782, addressed at 2,010 to 2,026 and third-party at 362 to 372. A third implementation, of eight pairings, returned figures spanning 9 to 39. None of the twelve runs here reproduces any of the four earlier figures, and none of them reaches 25. These are spans over the runs that were made and not bounds derived from named hinges, which is the distinction drawn at ⁷: a range over the runs made is a sample, not a bound. Nothing in the specification above closes the pairing rule, because rule (2) does not attempt the pairing at all, so there is no wording here for a reader to implement and no bound to derive from one. What every run of all three implementations agrees on is that the header-less count does not move, and that the change is under one per cent of the 4,187. So the block layout does not cost the five and a half percentage points the earlier disclosure invited the reader to discount by; what it does cost is unsettled, and the figures above are a sample and should be read as one. The sender figure is a lower bound, and by far more than the earlier version of this footnote allowed. Seventy-nine documents outside the sender class carry a sender value in which her name is damaged past rule (3): seventy-four in the addressed class and five in the third-party class. An earlier version of this note put that figure at thirty, found by pattern-matching the parsed sender strings for junk, iunk, lunk and nicole. Those thirty reproduce exactly, twenty-six addressed and four third-party, and they miss the commonest damaged form on the record. On every sheet of that class opened at the image for this note the sender is printed "junkermann", in quotation marks, and the extraction reads the opening "j as '5, so the parsed value is '5unkermann" and none of the four patterns can fire on it. That form and two fragmentary variants of it account for forty-eight of the seventy-nine, forty-five of them the bare string. The seventy-nine were arrived at twice, by two methods that agree on every document. The first extends the pattern list to the stems a damaged surname leaves: unker, nkerm, ermann and erman, beside the original four. The second uses no patterns at all: for each of the 2,319 documents outside the sender class whose first sheet yields a sender value, it takes the smallest number of single-character edits between junkermann and any substring of that value. At an edit distance of one that returns 57 documents, at two 77, at three 79, and at four and at five 79 still, so the set closes; and the seventy-nine it closes on are the same seventy-nine the stems reach. The same measure run against nicole adds nine documents, and all nine are Boris Nikolic, whose surname is two edits from her forename: they are homographs and are subtracted. Over the seventy-four in the addressed class the commonest values are '5unkermann" (45), “Nicole Junkerman” (5), “Nicole Junkennann” (5) and “Nicole Junkertnann” (4); the five in the third-party class are “Nicole Junkennann I Junkennanngroup”, “Nicole Junkerman”, “junkerrnann” twice and '5unkermann" cz. Each of those forms was sampled against the sheets it matches and is a real token in the extracted text and not an artefact of prefix matching. Five of the forty-eight were opened at the image at 300 dpi: EFTA00629581, EFTA00732833, EFTA00732835, EFTA00757091 and EFTA00766220. Every one of them prints From: "junkermann" on the paper, over a short message to Epstein’s own account. The surname is undamaged on the sheet; the damage is the extraction’s. They sit in the addressed class because the quoted original beneath each of them carries a To line naming her. The forty-ninth document the wider scan adds, EFTA02146275, was read at 300 dpi as well and prints From: Nicole Junkermann undamaged above a message of 21 February 2013. These are messages she sent, sitting outside the sender class. Seventy-nine is itself a floor, for the same reason thirty was. Both methods read the parsed sender value and nothing else, so neither can see a document whose sender value the extraction has destroyed past recognition, and neither reaches the 88 documents outside the sender class whose first sheet yields no sender value at all: 40 that carry no sender label and 48 whose label rule (4) empties. The header-less class is a further reason the sender figure is a floor, because the authored documents in it are acts by somebody. The header-less documents, enumerated. Under the wide reading, seventeen: Epstein’s monthly calendars for January and December 2014 (EFTA00285006, EFTA00284900); three schedule sheets of his assistant’s (HOUSE_OVERSIGHT_001529, HOUSE_OVERSIGHT_001532, HOUSE_OVERSIGHT_001537); three forensic device-extraction reports (EFTA00506709, EFTA01063888, EFTA01613781); an address-book card (EFTA00528250); an instruction sheet for people writing to him (EFTA00306074); a FedEx invoice (EFTA02825472); a monthly report (EFTA01074496); two message sheets that print no line-initial header label (EFTA02018707, EFTA02021060); the two profile documents of her named at ²¹ (EFTA00582749, EFTA02725548); and the translated press interview at ⁴⁷. Under the narrow reading, nine more: the nomination form at EFTA01113500; three to-do items carrying a subject line and nothing else (EFTA00413047, EFTA02168662, EFTA02169194); a fourth extraction report (EFTA00506839); and four more sheets of his assistant’s schedule traffic (HOUSE_OVERSIGHT_000616, HOUSE_OVERSIGHT_001542, HOUSE_OVERSIGHT_001543, HOUSE_OVERSIGHT_001553), the last three of them the e-mails named above. The curriculum vitae at ²¹ is not among them and cannot be. EFTA01919765 is not in the 4,187-document population at all, which is the point ²¹ itself makes: no sheet of it carries the string Junkermann in the extracted text, which is why the body reaches it through "Winamax" and "Infront Sports". An earlier version of this footnote, and of the body, named it as one of the header-less documents. Removing it changes no count, because it was never counted; the header-less figure changed for the unrelated reason given at (6). No correction for the damaged sender values or for the authored documents is applied to the table above. ↩ ↩² ↩³ ↩⁴

2. Nicole Junkermann to Jeffrey Epstein, “RE: Re:”, 14 September 2011, 1:33:07 p.m., whose quoted block carries the whole day’s exchange back to Epstein’s message of 12 September 2011, 6:36 p.m. Text layer only. https://epstein-data.com/EFTA01857199 pp.1–2. ↩ ↩²

3. Nicole Junkermann to “Jeffrey E.”, 11 February 2015, 11:38 a.m., and the thread above it. Read at the image at 400 dpi, on the copy whose second sheet prints “Infront was sold for over 1 billion / But tell me how are you?” in full. https://epstein-data.com/EFTA00864795 pp.1–2. A separate copy carrying her message alone, with its own header, is at https://epstein-data.com/EFTA02398275 p.1 (text layer). ↩ ↩²

4. “Young Global Leaders Nomination Form”, The Forum of Young Global Leaders, World Economic Forum, for the 2007 nomination round. Read at the image at 400 dpi. The candidate block carries her forename and surname, “Position: Entrepreneur”, “Organization: Various Enterprises”, a filled square beside “Female” against an empty box beside “Male”, and “Nationality: German”. Its address, telephone, fax, e-mail and date-of-birth fields are barred. Empty printed form fields, with no bar, in the nominator block above: first name, last name, position, organization, address, telephone, fax, e-mail — so the form as produced identifies nobody as the nominator. The corpus text renders both blocks identically, which is the point made in the body. https://epstein-data.com/EFTA01113500 p.1. The second sheet is the return-address page for the Forum’s Geneva office. ↩ ↩²

5. Jeffrey Epstein to David Stern, “Re: location idea,” 8 April 2009, 12:42:52 -0400, surviving inside the reply of 9 April 2009, 16:08:30 UTC. Read at the image at 400 dpi. The telephone number after “call her” is blacked out on the sheet; the bracketed marker in the quotation is editorial. The sender of the top message is barred on this copy; the confidentiality block printed beneath it names Asia Gateway Ltd., and that block is the whole basis for the attribution given in the text. https://epstein-data.com/EFTA00748642 p.1. The quotation follows the image, which differs from the corpus text of the same sheet in two ways: the sheet prints “old girlfriend” and the extraction emits “old irlfriend”; and the sheet prints two separate commas after “junkerman” and again after “german”, which the extraction closes into the double-comma ligature „ in both places. Nothing here is repaired; the image is the reading. ↩ ↩²

6. Populations measured against the FTS5 index of the full-text corpus on 11 September 2026, counting rowid only and resolving identifiers separately. "Nicole Junkermann" returns 3,676 pages in 3,419 documents; "Junkermann" 4,472 pages in 4,187 documents; "Junkermanngroup" 1,429 pages in 1,265 documents; "NJF Capital" 75 pages in 43 documents; "Winamax" 6 pages in 5 documents; "Infront Sports" 5 pages in 4 documents; "Nicole Brachetti Peretti" 2 pages in 2 documents; "NJF Holdings" 0. The union used on this page — the surname in every spelling below, plus the group token, plus the three entity terms — is 4,303 documents. Of the 4,187 in the "Junkermann" population, 4,175 carry an EFTA identifier, 11 a HOUSE_OVERSIGHT identifier and 1 a DOJ-OGR identifier; the scope was taken from the identifier prefix and not from the dataset column. ↩ ↩² ↩³ ↩⁴

7. Header dates were parsed from each document’s own first sheet across the 4,303-document union. Two earlier versions of this specification were each shown, by a reader implementing the printed words, to generate figures outside the range this footnote printed. It is therefore given below with every choice it leaves open either closed in words or named as a hinge, and with the count stated as the span over the hinges it names. The specification. On each document’s own first sheet, with runs of spaces and tabs flattened to a single space and each line stripped: (1) A label line is a line beginning Date or Sent, matched without regard to case, with or without a colon after it. The colon is optional, and that is not a convenience but a property of this production, measured over these 4,303 first sheets: it prints Sent with no colon on 3,220 header lines and Date with none on 9, and in 368 of the 4,246 documents that carry a Date or Sent label at all, not one such label carries a colon. Requiring the colon costs between 540 and 555 documents depending on the other choices below, and on the reading that produced the distribution further down it lands at 3,556, five hundred and three below the floor of the range printed here. An implementer who reads the colon as part of the label is not following this specification. The two label words are matched as printed and in no damaged form, unlike the sender label at ¹, which needs eighteen. That is a measured difference and not an oversight: of the 4,303 first sheets, 4,246 carry a clean Date or Sent label and 57 carry none, and not one of the documents this specification leaves undated carries a damaged form of either word in place of a clean one. Adding damaged label forms to the specification would move nothing. (2) The value is whatever follows the label on the same line; if nothing does, it is the next line as printed. (3) Read the label lines in the order they print and take the first value that yields a date. (4) A value yields a date when one of the four written forms 4 June 2014, June 4, 2014, 6/4/2014 and 2014-06-04, which are four renderings of the same day, appears anywhere inside it. It is searched for within the value and the value is not required to be the date, because these headers print as Monday, February 18, 2019 4:46 PM and Mon, 18 Feb 2019 16:45:32 +0000; requiring the whole value to be the date returns zero documents. Month names are matched in full and abbreviated, with or without a trailing point; the slashed form is read month first; the year is four digits. The two hinges. These are the choices the words above leave open deliberately, and they are the only ones left open. First, whether the reader stops at the first label line or scans on past one whose value yields no date. Second, whether the comma in June 4, 2014 is required or optional. Two binary choices are four readings and all four were run cold over the same 4,303 first sheets: stop and comma required, 4,059 dated and 244 undated · stop and comma optional, 4,061 / 242 · scan and comma required, 4,107 / 196 · scan and comma optional, 4,110 / 193. So a reader implementing the printed words lands on one of exactly four figures, and the count of documents with a header date lies between 4,059 and 4,110, the count without between 193 and 244. That span is derived from the named hinges and is not an envelope drawn round the runs that happened to be made, which is the distinction this footnote previously failed: a range over the runs made is a sample, not a bound. The other choices a reader might think open were each implemented both ways to check that they are closed by the words above: the colon, abbreviated month names, Sept as an abbreviation, a trailing point on an abbreviation, the order in which the four forms are tried, and whether a blank line between a label and its value is skipped. Thirty-two further runs, and not one of them moved any of the four figures by a single document. The stacked layout was mis-described until now, and correcting it is the reason rule (2) is stated as a limitation and not as a choice. This production prints stacked headers in two shapes, not one. In the first the label stands alone on its line and its value is on the line below, which is what rule (2) reads. In the second the labels come as a block (From: Sent: To: Subject: on consecutive lines) with their values following as a block beneath, and very often with a line of the message body interposed between the two blocks, so that the blocks cannot be paired simply by counting down them. EFTA02630742 p.1 prints its first sheet in that second shape. On such a sheet rule (2) takes the next line, which is the next label, and the document is left undated, so the figures above are a parser’s reach and not the record’s. Four ways of pairing the two blocks were implemented and none of them dates those sheets either, because the interposed body line destroys the alignment; that is a limitation this footnote discloses rather than one it has solved. What it costs is measured, and measured where it matters: of the 193 documents left undated on the scanning, comma-optional reading, 98 print some parseable date somewhere on their own first sheet, and exactly four print one inside the December-to-April window used at ⁴⁰, where all four are named and all four have been read. The runs that produced this footnote’s earlier figures are kept here as provenance, and the specification above does not generate all of them, which is the point of stating it as it now stands. Before the colon, the abbreviations and the within-value search were written down, an independent check of the same two branches returned 4,063 / 240 and 4,124 / 179, and two further parsers written independently of the one that produced the distribution below returned 4,107 / 196 and 4,113 / 190. Of those, 4,107 / 196 is reachable from the specification above and 4,063, 4,124 and 4,113 are not; each came from a wording that left at least one of the now-closed choices open. So this total is not a measurement of the record to the digit; it is a measurement of the record and a parser together, and it is given here as a range and not as a figure. The range printed above, 4,059 to 4,110, is narrower than the 4,059 to 4,124 this footnote used to print, and it is narrower because it is derived rather than observed: the older figure was the widest pair among the runs that had been made, and a reader implementing the older wording could and did land outside it in both directions. The distribution below was produced by the parser this page first used, which returned 4,110 / 193, and it is now reproduced exactly rather than approximately. An implementation of the specification above on its scanning, comma-optional reading returns the same 4,110 documents and the same count in every one of the eleven year buckets, a difference of zero. The three documents this footnote could not previously account for are accounted for: they are the comma-optional cases, the three whose month-day-year header prints without the comma the exemplar June 4, 2014 shows: EFTA02215661 p.1, of 21 June 2017, and HOUSE_OVERSIGHT_001543 p.1 and HOUSE_OVERSIGHT_001553 p.1, of 2 and 3 December 2014. They fall in exactly the two year buckets this footnote named, and the residue it called unexplained was the second of the two hinges above, unnamed. An earlier version of this footnote told the reader the re-parse differed in “exactly three documents” and asked that the totals be read “as accurate to within a handful of documents”. Both were false: the two readings of this footnote’s own wording differ by forty-eight documents, the widest pair by sixty-five, and by as much as fourteen in a single year bucket, in 2013. What does not move: all four readings of the specification above agree exactly on the earliest header date, 17 March 2009, and on the latest, 7 April 2019, and the only figure of this family in the body, the union of 4,303, reproduces exactly. But parser agreement is not what the span claims in the lede, the infobox and the body rest on, and an earlier version of this footnote offered it as though it were independent corroboration when it is not. Readings that are not faithful to the specification above move the terminus, and move it a long way: requiring the colon puts the latest header date at 24 July 2019, and refusing abbreviated month names puts it, on the scanning branch, at 13 August 2019. Both of those are damaged strings this page has already opened at the image, where the sheets print 2014, and they are set out in full below. The terminus therefore rests on the image reading and on nothing else; the agreement of the four readings is consistent with it and is not a second source for it. One parser’s distribution by year, given as that and not as a count of the record: 2009 114, 2010 471, 2011 464, 2012 319, 2013 465, 2014 673, 2015 301, 2016 194, 2017 503, 2018 356, 2019 250. Earliest header date 17 March 2009; latest 7 April 2019. The documents without a header date, however many the parser makes them, are not undated documents: they include the FedEx billing sheets at ⁸, ²² and ²³, whose own printed invoice dates run from 1 March 2002 to 2004, and which are the earliest documents in the population. This bounds the claim to the 4,303 documents and to nothing wider. The terminus was tested, and testing it at the image settled it more plainly than the corpus text could. Every sheet of all 4,303 documents was scanned for any 2019 date in four written forms; 9 documents carry a string later than 7 April 2019 in their extracted text, and all 9 were opened. Two are the same forwarded conference call-for-abstracts of 3 November 2018, whose listed deadlines genuinely do run to “25-28 June 2019” and which is not a later message. The other seven were rendered and read at the image at 400 dpi, and not one of them prints a 2019 date anywhere. In every case the sheet prints 2014 and the extraction emits 2019: the image reads “Sent: Wednesday, August 13, 2014 2:29 PM” where the corpus text emits “August 13, 2019” (EFTA00720859 p.1, EFTA00995364 p.1); “Sent: Thursday, July 24, 2014 11:52 PM” where it emits “July 24, 2019” (EFTA01914523 p.1, EFTA01914608 p.1); “Sent: Saturday, April 12, 2014 1:06 AM” where it emits “April 12, 2019” (EFTA01927504 p.1); “Date: Wed, 4 Jun 2014 09:22:54 -0600” where it emits “4 Jun 2019 09:22:59” (EFTA00695312 p.3); and “Sent: Wednesday, June 04, 2014 4:55 PM” where it emits “June 09, 2019” (EFTA00990274 p.1). Seven documents but five distinct strings, because the August and the July pairs are each one message surviving in two copies. The fault is uniform: the extraction captures the digit 4 as 9, in the year every time, in the seconds on one sheet, and in the day of the month on one. The last of the seven is the strongest of them, and it is worth setting out. EFTA00990274 is a single sheet whose own top header prints Date: Wed, 04 Jun 2014 14:57:41 +0000 and which carries two further undamaged quoted headers, “On Wed, Jun 4, 2014 at 8:58 AM” and “On Wed, Jun 4, 2014 at 8:52 AM”. The corpus text of that one sheet emits “Sent: Wednesday, June 09, 2019 4:55 PM”, which is two 4-to-9 substitutions inside a single string, with the correct value printed three more times on the same page. A reader working from the corpus text alone would find the weekday test failing on this sheet and on no other, because there the day of the month is damaged too: 9 June 2019 was a Sunday and 9 June 2014 a Monday, and neither is the Wednesday the string claims. On the other six the weekday survives the damage intact and agrees with 2014. Nothing read for this page postdates 7 April 2019. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹

8. FedEx invoice 7-134-39206, invoice date 1 March 2002, billed to “JEFFREY E EPSTEIN / 457 MADISON AVE / NEW YORK NY 10022-6843”. Read at the image at 400 dpi, printed page 4 of 7: shipped 18 February 2002, FedEx Intl Priority, tracking ID 830981617038, sender “J EPSTEIN”, recipient “NICOLE JUNKERMANN” at a private address abroad, delivered 20 February 2002 at 11:49 and signed for, declared value USD 150.00, customs entry date 21 February 2002, entry number CHC56268700000. https://epstein-data.com/EFTA01315280 p.4. A second copy of the same invoice is at https://epstein-data.com/EFTA00219889 p.4 (text layer only). ↩ ↩² ↩³

9. Jeffrey Epstein to Nicole Junkermann, 23 February 2013, 09:02 p.m., and her reply of 10:18:17 p.m. the same evening. Read at the image at 400 dpi; the spellings are the sheet’s own and are reproduced unaltered. https://epstein-data.com/EFTA02720599 p.1. "Let s dicsuss it" returns 24 pages in 24 documents and "house staff" AND "Junkermann" 28 pages in 28 documents; one of each was opened, and the remainder were counted and not read. ↩ ↩² ↩³

10. Nicole Junkermann to Jeffrey Epstein, 18 February 2019, 11:41 a.m., with his reply of 15:56, her reply of 12:08 p.m. and his “Ok” of 16:45:32, all on one sheet. Read at the image at 400 dpi; the seven points are a numbered list on the sheet, the ellipsis in point 3 is the sheet’s own, and “is3.9m” and “carry1.5 m” are run together as printed. The quotation follows the image and differs from the corpus text of the same sheet at four points: the extraction runs the list marker into the next word, emitting “2.1 am sorry” where the sheet prints “2. I am sorry”; it emits “(lm you” where the sheet prints “(1m you”; in his 15:56 message it emits “I don’t like it” where the sheet prints “I don’t like ir”; and in the fifth message quoted below it emits “I I have” where the sheet prints “1 I have”, a numeral one followed by a capital I. The fourth was settled at 600 dpi on p.2: the first character carries the diagonal upper flag and flat base of the digit and the second has the symmetrical top and bottom serifs of a capital I. That fourth difference is this production’s 4-to-9 fault at ⁷ running the other way, a digit captured as a letter, and an earlier draft of this footnote carried the extraction’s form of it. None of the four is repaired here; all four are read off the image. Her list closes on the sheet with “Lets speak”, two spaces between the words as printed, and his 15:56 reply is quoted in the body in full, both sentences: the first of them is the assertion her 12:08 answer disputes. https://epstein-data.com/EFTA01030134 p.1, which carries the four messages, the numbered list and the fifth message’s header, and p.2, which carries the fifth message’s body; the document is two sheets, pp.1–2. The sheet carries a fifth message the body does not narrate: beneath her seven points a quoted header prints “From: J … Sent: 18 February 2019 11:54”, whose body runs onto p.2 — “im free after 2 pm ny time . if convenient . lets continue. . 1 I have litle idea of carbyne results to date. , if your guy would like to tell me in addition to how much we are giving up. how much we would have made, that is a first step. . counting only future losses confuses me.” Its face value, 11:54, is later than the 11:41 attribution of the message nested above it, which is the plainest demonstration on this sheet that the stamps are not on one clock; the order used in the body is the nesting order and no zone is assumed for any stamp. "Carbyne invested" returns 13 pages in 13 documents; one was opened at the image, a second — EFTA02630742, a further copy of this same thread — was read in the text layer as one of the five documents at the parser margin listed at ⁴⁰, and the other 11 were counted and not read. “FSB” in point 6 is her own abbreviation and is explained by her message of 1 March 2019 at ⁴⁴, which opens the same list with “FSB (Russians) valued the business between 1.5-2.5b (Euro)”. ↩ ↩²

11. Jeffrey Epstein to Nicole Junkermann and Robert Lawrence Kuhn, 13 December 2018, 10:53 p.m., whose entire body is the sentence quoted. Text layer only. https://epstein-data.com/EFTA02611645 p.1. "Nicole friend for 20 years" returns 5 pages in 3 documents and all three were opened. ↩ ↩²

12. Nicole Junkermann to Jeffrey Epstein, “Fwd: Really Sports,” 2 January 2019, 20:30:35 UTC, forwarding her chief financial officer’s memorandum with attachments; and Epstein’s reply of 20:43:52 UTC, quoted here in full. Those two sheets read in the text layer only: https://epstein-data.com/EFTA01009653 p.1; https://epstein-data.com/EFTA01009648 pp.1–3, whose second and third sheets carry the 31 December 2018 exchange between her chief financial officer and a Shanghai associate quoted in the body. His message of 4 January 2019, header Date: Fri, 04 Jan 2019 01:55:53 +0000, is at https://epstein-data.com/EFTA01009490 p.1, read at the image at 600 dpi. The quotation follows that image, which differs from the corpus text of the same sheet at one point: the sheet prints “even if you win, , i m not sure you do”, with two separate commas and a space between them, and the extraction of this copy closes them to a single comma while the extraction of other copies emits the double-comma ligature „. Neither is repaired here and neither is invented; the image is the reading. The same shape appears at ⁵. Her reply of 6 January 2019, header Date: Sun, 06 Jan 2019 12:51:01 +0000, is at https://epstein-data.com/EFTA01009260 p.1, read at the image at 400 and 600 dpi, and in three further copies: https://epstein-data.com/EFTA02606254, and https://epstein-data.com/EFTA01009258 and https://epstein-data.com/EFTA02606204, which carry it inside his reply of 13:02 the same day. It is an interleaved reply, her lines set inside his 4 January draft, and on EFTA01009260 p.1 the two voices are distinguished by colour through that interleaved passage: his text prints in black and every one of her interpolations within it prints in red. Her covering line at the head of the sheet, “Let me know when u want to speak”, is hers and prints in black, so the colour rule holds inside the quoted passage and not across the whole sheet. On the placement of her longest answer: the red run inside his “the fake document is a key if provable(…)” is the shorter “Mr Wu is not very patient so he went for the quick solution as he never thought I would react”, which this page does not quote; her “Yes we have a notary recording…” run begins after his “but the chinese protect their own.” and ends before his “my instict is that this needs finesse rather than a legal attack.” The body states its position on the sheet rather than placing it inside the disputed-document sentence. Red runs on the sheet that this page does not quote are “Agree think it will be a la Chinese.. I have to file another claim against the company and not against the shareholders to enforce the legal representative He was convinced to win( reason tell u on the phone)”, “Agree this is just a sideshow but we had to react as they filed for the claim” and “at your convenience”; none of them bears on anything asserted here. The extraction flattens the colour and emits both voices as one continuous run of text, in which the two are indistinguishable; the attributions in the body are taken from the image alone and from no other copy. "where is the co incorpprated" returns 1 page in 1 document; "my first draft response" returns 6 pages in 6 documents and all six were opened; two are his 4 January message standing alone, two are her interleaved reply, and two are his 13:02 reply carrying hers. ↩ ↩² ↩³ ↩⁴

13. “Office | Junkermanngroup” to Jeffrey Epstein, “Parcel for you,” 31 January 2019 — a message that survives only inside the quoted blocks of the copies, where two of them attribute it to 6:57 a.m. and one to 17:57 (see ¹⁴) — with Epstein’s “yes” of 11:59, whose stamp is settled at ¹⁴, and the office’s “Thank you so much.” of 12:09, which is this sheet’s own top-level Sent: field. Read at the image at 400 dpi, both sheets. https://epstein-data.com/EFTA02628055 pp.1–2. Two findings from the image that the corpus text does not carry: the telephone and mobile lines of the signature block are barred, and the sentence “And the courier will also need a phone number for the delivery. Could I use this one” is followed by a black bar. In the corpus text both read as though nothing were there. This is the sheet that prints cleanly and is the copy quoted. ↩ ↩²

14. Nicole Junkermann to her own member of staff, copied to Epstein, “Re: Parcel for you,” 31 January 2019, 12:06 p.m., carrying her office’s earlier message in its quoted block. Both of its sheets read at the image at 400 dpi. The office’s opening message survives in this thread only as a quoted attribution, and the copies do not agree on its stamp: two of them print “On Thu, Jan 31, 2019 at 6:57 AM” and this one prints “On 31 Jan 2019, at 17:57”, an eleven-hour difference. Both are quoted in-thread attributions, each written by the client doing the quoting rather than by the sender, and nothing read for this page settles which clock either of them is on; no copy carries that message with a top-level header of its own. The times given in the body are taken only from top-level Sent: fields, and on this thread exactly one of those is confirmed directly: the copy at https://epstein-data.com/EFTA02628104 p.1 prints Sent: Thursday, January 31, 2019 11:59 AM, and its p.2 carries an iPhone property-list date-received value of 1548935957, which decodes to 11:59:17 UTC. This document carries no property-list trailer on either sheet: p.1 ends at the quoted message and p.2 carries only the office signature block, in which two contact lines are barred. Both sheets were read at the image for the trailer specifically, because an absence on a sheet is the easiest thing on this page to get wrong. Its Sent: Thursday, January 31, 2019 12:06 PM is therefore a bare field with no zone printed beside it and no epoch anywhere in the document to fix one, and the body says so rather than reading the 11:59 stamp’s zone across onto it. https://epstein-data.com/EFTA02628068 p.1 for the message and its quoted block and p.2 for the signature block; the document is two sheets, pp.1–2. The sender name on this sheet is her maiden name and the quoted message beneath it is the one that uses the married name — both on one sheet. ↩ ↩² ↩³ ↩⁴

15. "Brachetti Peretti" returns 153 pages in 122 documents. In those documents the forename token printed immediately before “Brachetti” is Ugo 259 times, in 103 documents; Ferdinando 8 times in 4 documents; Nicole twice, in the two documents at ³³; and Sylvianne once. "Ugo Brachetti" alone returns 139 pages in 109 documents and "Ugo" 303 pages in 265. Ugo Brachetti Peretti is a different correspondent of Epstein’s, and none of his correspondence is used on this page except the single sheet named next. What the release establishes is more than a shared surname. An earlier version of this footnote said, in bold, that it established “a surname and no more”, and that was wrong. On EFTA00627738 p.1 of 1 — a sheet inside the "Brachetti Peretti" population and outside this page’s own union of 4,303, rendered and read at the image at 400 dpi for this footnote and not taken from the text layer or from another desk page — Epstein wrote on 17 February 2017 at 21:09: “I met fernando last week, he came with nicole to palm beach”. The answer, from “Ugo Brachetti Peretti” over a barred address and stamped Date: Sat, 18 Feb 2017 05:09:16 +0000, reads: “hope everything is well, i knew my brother was around,let’s keep in touch / ugo”. So the release carries Ugo Brachetti Peretti saying in his own words that the man Epstein had just named to him was his brother. What the sheet does not carry is the identification of that man, and this page does not close it from the corpus: the forename on it is spelled “fernando” and not “Ferdinando”, and nothing on it says who its “nicole” is. The desk’s page for that correspondent reads the same sheet, states the brother relation as the documented fact in those same terms, and declines the identification in the same way; the extraction of this sheet prints the forename “femando”, which is the rn-to-m fault, and that is a further reason the reading above was taken at the image. The identification of that brother as this subject’s husband is on the outside record, and is taken from it in two steps, each stated in its own words by a named newspaper. Il Fatto Quotidiano of 12 February 2026, reporting on this release, writes of “Ugo Brachetti Peretti, all’epoca presidente e proprietario con il fratello Ferdinando del gruppo petrolifero” — Ugo Brachetti Peretti, at the time chairman and owner of the oil group with his brother Ferdinando (Nicola Borzi, “Gli italiani negli Epstein file: Brachetti chiese aiuto per la cessione di Api”, https://www.ilfattoquotidiano.it/in-edicola/articoli/2026/02/12/gli-italiani-negli-epstein-file-brachetti-chiese-aiuto-per-la-cessione-di-api/8288928/). Il Tempo at ³² identifies this subject’s husband as “conte Ferdinando Brachetti Peretti, petroliere”, “Presidente e Amministratore Delegato del Gruppo API (primo gruppo petrolifero privato europeo)” — the same man and the same company. So on the outside record Ugo is her husband’s brother and therefore her brother-in-law. That is a two-source chain for the identification and is marked as one here rather than stated flat: a desk file that asserted the same relation without any source was withdrawn, and the withdrawal was right on its own terms because nothing supported it at the time. It was wrong about the release, though, and so was this footnote until now: the sheet above was already on the desk’s own published page for the other correspondent while this page was telling its reader the release was silent on the relation. A correction that leaves the sentence it corrects standing beside it leaves a page saying both things, so the sentence has been removed and not merely qualified. The desk’s brief for this page recorded the split as 252/3/8 across 140 documents; the figures here were re-derived and are 259/2/8 across 122. ↩ ↩²

16. "Infront" returns 124 pages in 115 documents. 48 of the 115 lie inside the name population and every one of the 48 is the company: they comprise the 11 February 2015 thread at ³, the September 2011 “Presse Release Infront” exchange at ²⁶, a recruitment message offering “Small Paris agency / Event driven in uk / Or Infront…”, the profile document at https://epstein-data.com/EFTA00582749 and the interview at ⁴⁷. Of the 67 outside the name population, 65 are the extraction running the two words “in front” together — every one of the 65 was opened and the line carrying the string read — and two are the company: the curriculum vitae at ²¹, whose letterhead the extraction fails to capture, and https://epstein-data.com/EFTA01071778, a two-sheet Infront Sports & Media press page. So: 50 company, 65 false positive, 115 total. "Infront Sports" on its own returns 5 pages in 4 documents. The desk’s brief for this page recorded this trap as “115 of 117 false positives” from a page-level sample; at document level the false-positive share is 65 of 115. ↩

17. "Magnum Photos" returns 66 pages in 50 documents and none of the 50 is in the name population; the context line of each was read, and 47 carry the signature block of the photographer Christopher Anderson in correspondence with Epstein’s assistant in March 2015, and 3 do not: a 2014 invitation to a forum at a members’ club, a dated vendor list, and a list of a retailer’s featured partners. 47 + 3 = 50. "Carbyne" returns 85 pages in 50 documents, 24 inside the name population and 26 outside; the 26 are Ehud Barak’s Qatar presentation thread, Richard Kahn’s 2019 messages, a shareholders’ agreement, a Peter Thiel introduction and company reporting, and they are covered at ehud-barak. The context line of all 50 was read. ↩

18. "Junkerman", with one n, returns 172 pages in 171 documents, of which 66 carry no other spelling of the name. All 66 were opened and the line carrying the string was read in each. All 66 concern this subject. Classified by that line: 37 are messages from Epstein whose To: value is “junkerman” or “Nicole Junkerman”; 6 are Google Calendar alerts and reminders naming her; 5 are copies of the 2 July 2010 message at ⁵⁰; 4 are copies of the exchange at ²⁴; 4 are copies of the Advent thread at ⁵¹; 3 are copies of a 3 May 2012 list; 2 are FedEx billing sheets; and 5 were read individually — among them the message at ⁵, a telephone-message list naming her among the day’s callers, and a Robert Kuhn thread. 37 + 6 + 5 + 4 + 4 + 3 + 2 + 5 = 66. None is a different person. The one-n form is not reachable from ocr_variants.py, which for this surname generates a single form, "Junkerrnann". ↩

19. Damaged spellings, each searched as a phrase: "Junkerrnann" 50 documents (11 outside the "Junkermann" population), "Junkennann" 250 (21 outside), "Nicole Junkerrnann" 37 (8 outside), "lunkermann" 43 (6 outside), "Junlcermann" 4 (0 outside), "Junkermarm" 1 (0 outside). Those six per-form figures sum to 46, but they overlap: the union of the six, measured against the "Junkermann" population, is 34 documents, not 46. Of those 34, two had already been reached by the one-n form, so the increment these spellings contribute to the running union is 32, which is the figure used in the body. "Nikole Junkermann" and "Junkermanri" return 0. The surname carries no non-folding diacritic. ↩

20. "Junkermanngroup" is a single FTS token and therefore does not match a search for "Junkermann". It returns 1,265 documents, 16 of them outside the "Junkermann" population; five of those 16 had already been reached by a damaged spelling, so the increment to the running union is 11, which is the figure used in the body. "Junkennanngroup" returns 44 documents and "Junkerrnanngroup" 36; 2 and 1 of those respectively lie outside the "Junkermann" population, but both sets sit entirely inside "Junkermanngroup" and neither adds anything to the union. ↩

21. Curriculum vitae, four sheets, undated, running in reverse chronological order from “2012 Current Investor in Real Estate”, which heads its first sheet, to “1998 – 2001 Co-Founder & CEO, WINAMAX” at the foot of its third. Read at the image at 400 dpi, printed page 3, which prints “2005 – 2012 Founder and Executive Partner & Investor, United in Sports, Luxembourg”, “Since inception in 2005, the fund has raised €242.2m and with the successful sale of Infront it generated 21% ROI for original investors”, “2008 – 2012 Vice Chairman, Infront Holding & Infront Sports & Media AG, Switzerland” and “2002 – 2005 Private equity manager for Robert Louis Dreyfus”. https://epstein-data.com/EFTA01919765 p.3; the “2012 Current Investor in Real Estate” heading is on p.1 of the same run. The corpus text for these sheets carries no readable letterhead: the image prints “Nicole Junkermann” in light grey type at the head of the sheet, above barred contact lines and the business name “Elkstone Capital International Services Ltd.”; on the first sheet the extraction emits only the “NJ” monogram, and on the second, third and fourth it emits the monogram preceded by a failed capture of the same letterhead — ;( (Il l' luu6ri n:,: Fin on two of them and \I( Mil, ill on the last. The operative point is unchanged and was verified separately: no page of this four-sheet document carries the string Junkermann in the corpus text, which is why the document is reachable by "Winamax" and "Infront Sports" and not by her name. The corpus text also renders the euro sign as E and the date range “2008 – 2012” as “200R 2012”; both are read from the image above. Two further profile documents are at https://epstein-data.com/EFTA00582749 (four sheets, text layer only, letterhead extracted) and https://epstein-data.com/EFTA02725548 (two sheets, text layer only, the surname extracted as “IUNKERMANN”). The Harvard line quoted in the body is on p.4 of https://epstein-data.com/EFTA01919765 (text layer). ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

22. FedEx invoice 4-397-07556, invoice date 30 September 2002. Read at the image at 400 dpi, printed page 4 of 10: picked up 18 September 2002, FedEx Standard Overnight, tracking ID 835377097351, sender “J EPSTEIN / JEFFREY E EPSTEIN / 457 MADISON AVE”, recipient “NICOLE JUNKERMANN / HARVARD BUSINESS SCHOOL”, 10.0 lbs, delivered 19 September 2002 at 10:00, signed by “R.ALMEIDA”, total transportation charges USD 22.83. https://epstein-data.com/EFTA00217863 p.4. ↩ ↩²

23. FedEx invoice 1-907-98996, invoice date 2 August 2004. Read at the image at 400 dpi, printed page 4 of 5: picked up 26 July 2004, reference “LESLEY”, FedEx Intl Priority, tracking ID 790219140020, sender “CECILIA STEEN / JEFFREY E. EPSTEIN / 457 MADISON AVENUE / 4TH FLOOR”, recipient “NICOLE JUNKERMAN / NIRO”, 0.3 lbs, delivered 28 July 2004 at 09:50, signed ”.RIBEIRO”, total USD 31.28. https://epstein-data.com/EFTA00216942 p.4; a second copy at https://epstein-data.com/EFTA01313004 p.4 (text layer only). Three further shipments, all of them from that office to her, are billed across three invoices, and the four identifiers below were all read at the image at 400 dpi. On every one of them the Junkermann block prints “JEFFREY E. EPSTEIN / 457 MADISON AVENUE / 4TH FLOOR / NEW YORK NY 10022 US” in the Sender column and “NICOLE JUNKERMANN” in the Recipient column, with her address barred, “Payor: Shipper” and “Orig./Dest. JRB/PAR”; not one is a shipment from her. Invoice 4-495-05371, invoice date 9 December 2002, printed page 6 of 6: tracking ID 791989127138, dropped off 4 December 2002, delivered 6 December 2002 at 11:05, signed “G.HUGES”, USD 28.25 — https://epstein-data.com/EFTA01317964 p.6. Invoice 4-540-22984, invoice date 30 December 2002, printed page 5 of 6: tracking ID 790171891179, picked up 24 December 2002, 3.0 lbs, printed on the sheet as “Undeliverable Package”, delivered 26 December 2002 at 08:07 against a signature field reading “RETURN” and the airbill reference, USD 61.30 — https://epstein-data.com/EFTA00218115 p.5 and https://epstein-data.com/EFTA01313902 p.5, which are two copies of that single sheet — the same invoice, the same page 5 of 6 and the same tracking ID — and not two separate billings. Invoice 4-540-55582, whose own invoice date is 6 January 2003, printed page 4 of 5: tracking ID 791501798458, picked up 26 December 2002, 4.0 lbs, delivered 30 December 2002 at 10:59, signed “E.HUGUES”, USD 109.88 — https://epstein-data.com/EFTA01318048 p.4. The corpus text of all four sheets interleaves the two address columns row by row, emitting her name before his; read in the text layer alone the blocks appear to run the other way, and the image says they do not. "Junkermann" AND "invoice" returns 9 pages in 9 documents, all of them FedEx billing sheets on this account, and all 9 were opened. ↩ ↩² ↩³

24. Thread of 28 September 2009, ending 23:45:00 UTC. Read at the image at 400 dpi: the exchange runs to seven messages — 12:08 a.m., 6:08 a.m., 1:21 p.m., 19:16:28, 7:17 p.m., 19:34 and 23:45 — of which the 19:34 is Epstein’s, on an unrelated domestic subject, and is not quoted here. Every gmax sender address in the thread is barred; that is what makes the identification at ²⁵ rest on a court record rather than on this sheet. https://epstein-data.com/EFTA00770639 p.1. "do yu mean nicole junkerman" returns 4 pages in 4 documents and all four were opened: this one, https://epstein-data.com/EFTA00884447, https://epstein-data.com/EFTA02438195 and HOUSE_OVERSIGHT_022245. ↩ ↩²

25. The account name is printed as “gmax” and “Gmax” on the sheet at ²⁴ and the address beside it is barred. This page does not identify the account from that sheet. The basis for associating the handle with Ghislaine Maxwell is a separate document in the same release: a sentencing transcript in United States v. Maxwell, in which the court states, at lines 20 to 22 of the sheet, “Metadata indicated that the computer was registered to “GMax” and the document was saved under the user name “Ghislaine."" Read at the image at 300 dpi on https://epstein-data.com/EFTA02844213 p.1168, which is the page-aligned twin of DOJ-OGR-00020385; both copies’ text layers are scrambled by the transcript’s two-column capture — in each, the words “Metadata indicated” are displaced several lines away from the clause they open, with the transcript’s line numbers interleaved between — so the quotation is taken from the image and from neither text layer. The identification is offered here as a documented court finding about the handle, not as a fact the 2009 sheet establishes. ↩ ↩²

26. Jeffrey Epstein to Nicole Junkermann, “Re: Fw: Presse Release Infront,” 3 September 2011, 09:39:15 UTC, and her reply of 4 September 2011, 6:25:06 a.m. Text layer only. https://epstein-data.com/EFTA00919009 p.1; https://epstein-data.com/EFTA02024158 p.1; https://epstein-data.com/EFTA01809343 p.1. "Presse Release Infront" returns 3 pages in 3 documents and all three were opened. ↩ ↩²

27. "infront? is that your former co?" returns 24 pages in 24 documents. Two were opened — those cited at ³ — and the other 22 were counted and not read; nothing is said here about their contents beyond the count. ↩

28. FTS5 conjunctions over text_content, run 11 September 2026, each scoped to the surname. Returning zero pages: "Junkermann" AND "consulting agreement", "Junkermann" AND "salary", "Junkermann" AND "engagement letter", "Junkermann" AND "wire", "Junkermann" AND "retainer". Returning pages, all of which were opened: "Junkermann" AND "employment" 2 pages in 2 documents, both copies of a Carbyne board minute authorising “Execution of employment agreements of up to USD150,000 per employee” and concerning that company’s staff, not her; "Junkermann" AND "invoice" 9 pages in 9 documents, all FedEx billing sheets and all cited at ²³; "Junkermann" AND "payment" 22 pages in 22 documents, of which 9 are FedEx remittance boilerplate, 12 are copies of the Rome-to-Berlin car errand at ³⁰, and 1 is the curriculum vitae, where “Payment” names a market sector in a fund’s investment thesis. No document among these records any employment of her by Epstein, any fee, or any payment by him to her. The claim is bounded to the documents read for this page. ↩

29. Google Calendar alerts and reminders in Epstein’s own account. "Nicole Junkerman to join quant meetings" returns 2 pages in 2 documents, both opened: https://epstein-data.com/EFTA02148857 (alert, 5 February 2013, 1:56:02 p.m.) and https://epstein-data.com/EFTA02150117 (reminder, 5 February 2013, 10:20:05 a.m.). The 14 February 2013 reminder is at https://epstein-data.com/EFTA02148439 and the 26 February 2013 alert at https://epstein-data.com/EFTA02146140; the 20 June 2013 pair is at https://epstein-data.com/EFTA02135529 and https://epstein-data.com/EFTA02135540. All six read in the text layer only. "quant meetings" returns 136 pages in 134 documents corpus-wide; that wider set was not read and nothing is said about it beyond the count. ↩

30. “Rome to Berlin” thread, 20–25 June 2013, between Lesley Groff, Nicole Junkermann and a third coordinator. Text layer only. https://epstein-data.com/EFTA02135243 pp.1–3, which carries the whole thread including Groff’s “how I can get payment to you or the company”, Junkermann’s replies of 20 June, the confirmation of 21 June and Groff’s follow-up of 25 June offering a reimbursement cheque. The travellers’ names are barred on this copy. "Rome to Berlin" returns 40 pages in 22 documents; one document was opened in full and the other 21 were counted and not read. ↩ ↩²

31. Ehud Barak to “Junkermann Nicole”, copied to “Elichai Amir”, “Reporty Advisory Board NJ,” 11 February 2016, 6:05 p.m. Text layer only. https://epstein-data.com/EFTA02474299 p.1. "Reporty Advisory Board" returns 1 page in 1 document. ↩

32. Outside record, not the release, and cited as that. Silvia Sfregola, “In Vaticano il battesimo dell’anno”, Il Tempo, 24 March 2019, https://www.iltempo.it/roma-capitale/2019/03/19/news/in-vaticano-il-battesimo-dell-anno-1118997/, reporting the baptism of the couple’s daughter and saying of the marriage: “La coppia si era sposata due anni fa in gran segreto in una cerimonia molto intima a cui erano presenti soltanto i testimoni” — the couple had married two years earlier in great secrecy, in a very intimate ceremony at which only the witnesses were present. The year is reached by subtraction from a report dated March 2019 and is given here as a year, not as a date; March 2017 and Rome are reported elsewhere and are not asserted here. No document in the release states the marriage, its date or its place, and nothing about it is taken from a sheet. What the release itself shows, independently, is the married name in use by January 2019, at ¹³ and ³³. The section heading above, the infobox row “Contact after the 2017 marriage” and the clause “three years before this subject’s marriage” in the opening section all rest on this footnote and on nothing in the release. ↩ ↩²

33. "Nicole Brachetti Peretti" returns 2 pages in 2 documents, https://epstein-data.com/EFTA02628055 and https://epstein-data.com/EFTA02628068, both opened and both read at the image. A third copy of the same thread exists at https://epstein-data.com/EFTA02628104 and is not returned by that search, because the extraction of that copy carries quoted-printable transmission damage through the name and through several other words of the same sentence. It was found by reading the Bates neighbourhood, not by searching, and it is the reason the figure here is 2 and not 3. ↩ ↩² ↩³

34. Nicole Junkermann to Ehud Barak, “News/ Questions,” 11 June 2018, 2:20:26 GMT+2, forwarded by Barak to Epstein at 00:27:35 UTC the same day with the covering line “FYI. / Best / EB”, and carrying the attachment named “Carbyne_questions_on_behalf of Nicole_Junkermann.docx”. Text layer only. https://epstein-data.com/EFTA01036472 p.1. ↩

35. Ehud Barak to Nicole Junkermann, “Re: News/ Questions,” 11 June 2018, 20:48:27 UTC, with two attachments. Text layer only. https://epstein-data.com/EFTA01013264 pp.1–2. "Carbyne_questions_on_behalf" returns 3 pages in 2 documents and both were opened. ↩

36. Nicole Junkermann to Jeffrey Epstein, “FW: Carbyne,” 17 February 2019, 1:03 p.m., forwarding her chief financial officer’s message of 15 February 2019, 18:38, headed “Execution of warrants only calculation” and carrying a four-row sensitivity table; with Epstein’s replies of 5:04 p.m. (“Yes”) and 7:31 p.m. Text layer only. https://epstein-data.com/EFTA02630978 pp.1–3; https://epstein-data.com/EFTA02631272 pp.1–3. "Execution of warrants only calculation" returns 2 pages in 2 documents and both were opened. ↩

37. Jeffrey Epstein to Nicole Junkermann, 14 February 2019, 19:52:37 UTC, carrying her “Sure tmr?” of 2:52 p.m. and his “can we speak about carbyn?” of 18:58. Text layer only. https://epstein-data.com/EFTA01029469 p.1; second copy https://epstein-data.com/EFTA02629670 p.1. "i am letting ehud do the warrants" returns 2 pages in 2 documents and both were opened. ↩

38. Nicole Junkermann to Robert Kuhn, copied to Epstein, 17 December 2018, 18:40:51 UTC, carrying Kuhn’s reply of 14 December 2018, 23:48:48 UTC in its quoted block. Text layer only. https://epstein-data.com/EFTA01013299 p.1; Kuhn’s message alone at https://epstein-data.com/EFTA01013560 p.1. ↩

39. Jeffrey Epstein to Nicole Junkermann, 1 January 2019, 3:10 p.m., whose entire body is the one line quoted. Read at the image at 400 dpi. https://epstein-data.com/EFTA02608775 p.1. Her replies of 6:17 p.m. and 6:18 p.m. and his of 6:53 p.m. are at https://epstein-data.com/EFTA02608222, https://epstein-data.com/EFTA02608150 and https://epstein-data.com/EFTA02608016 (text layer). ↩

40. A document refuting the statement in the body would be a message from her to Epstein, between 1 January 2019 and 7 April 2019 when the population ends, supplying a Chinese name in answer. "you didnt send me the chinese guys name" returns 7 pages in 7 documents and all seven were opened: they are six copies of the message itself and its replies, and one further copy of the same thread. What was read, stated as work done and not as a count of the record. A list was built of the documents in the union whose own first sheet yielded a header date between 1 December 2018 and 7 April 2019, the last date in the population, using the parser described at ⁷; that list ran to 276 documents — 174 in December, January and February, and a further 102 in March and the first week of April — and the body of each of those first sheets was printed and read, deduplicated by body so that the copies collapse to the distinct messages behind them. None of them contains her sending him a Chinese name. The qualifier is load-bearing and the unqualified form, which this footnote carried until now, was false: at least one document inside this very window is a message from her to Epstein naming a member of her own staff, which is why the criterion this footnote tests is the one stated in its first sentence — a Chinese name, in answer to the request of 1 January — and not the sending of a name as such. The window was then enumerated exhaustively rather than scoped, and the enumeration can be stated in full. Under the four readings of the specification at ⁷ the window holds 275 documents on both stopping readings and 277 on both scanning ones; the intersection is 275 and the margin between them is two documents, EFTA01010516 p.1 of 25 December 2018 and EFTA02631233 p.1 of 4 March 2019, each inside the window when the parser scans on and outside it when it stops. The 276 documents printed and read are the 275 of the intersection plus the first of those two, which is where the 174 and the 102 come from; the second was read as a margin document. A separate class then had to be tested, because it is invisible to the window and not to the record: the specification at ⁷ cannot date a first sheet whose labels print as a block, nor one whose date value the extraction has truncated, and such a document is outside the window however the sheet itself is dated. Of the 193 documents that specification leaves undated, 98 print some parseable date somewhere on their own first sheet, and exactly four print one inside 1 December 2018 to 7 April 2019: EFTA02277089 p.1 of 1 February 2019, EFTA02630742 p.1 of 18 February, EFTA02631097 p.1 of 10 March and EFTA02636919 p.1 of 27 March. The remaining 189 were swept for a Dec-to-April month word standing beside a 2018 or 2019 year in any other written form, which returned five sheets, all of them 2018 dates outside the window. So the count of documents whose first sheet carries a date in this window, by the specification or against it, is 281, and all 281 have been read. Of those four, three are block-stacked sheets and the fourth, EFTA02631097, is a sheet whose Sent value the extraction cuts off mid-month. All six of the documents named in this paragraph and the last, being the two at the stopping-and-scanning margin and the four the specification cannot date, were read, in the text layer and not at the image, which is the provenance an enumeration of first sheets carries and is stated as the weaker thing it is. None of the six contains her sending him a Chinese name: they are a 25 December 2018 forward of Robert Kuhn’s message and its Chinese press attachments; a further copy of the 18 February thread at ¹⁰; a two-line exchange of 4 and 5 March about a New York Times letters page; the 10 March conference-call dial-in sent by Lesley Groff to her, copied to Robert Kuhn and Epstein; a three-line “Can i come on Monday?” of 27 March; and, the sixth, a 1 February 2019 message on Epstein’s side of the correspondence to a third party named Daphne Wallace, carrying beneath it an introduction her own chief financial officer had sent that party the same day about an unrelated company. That sixth is not from her, not addressed to him, and carries no Chinese name; it was found by the enumeration above and not by the list this footnote previously published, which is the test an absence claim has to be able to fail. So the conclusion holds across the whole of the window and across every reading of the parser, and is no longer scoped to the readings that happened to be run. The enumeration is also first-sheet-only in both directions: a document whose first sheet is dated February is inside the window even if a March message sits on its third sheet, and those later sheets were not read. The corpus-wide "Junkermann" AND "Wu" arm below closes that gap for the name Wu and for no other name. The March and April traffic is the conference call with Robert Kuhn on 10 March, the Carbyne warrant exercise, travel arrangements, and the exchange of 4 and 7 April on which the population ends; the only Chinese names in it are Wu and Zhang, which he already had from the 2 January forward, and Adam Zhu, whom Kuhn names and speaks for on 10 March and who enters the thread from Kuhn’s side and not from hers. Documents whose first sheet yields no header date are outside the window as the specification draws it, and are counted at ⁷; they are not outside this enumeration, because the paragraph above searches the first sheet of all 193 for a date in any of the four forms and reads every one that prints a date in the window. "Junkermann" AND "Wu" returns 22 pages in 22 documents and all 22 were opened. By header date the earliest is EFTA01009648 of 2 January 2019, the memorandum forward at ¹², which carries the 31 December 2018 staff exchange about passport copies; four documents of 6 January carry the interleaved reply; four of 9 January carry the arbitration result; Epstein’s message of 22 January at ⁴² is the first in which he writes the name himself. None of the 22 is a reply by her to the request of 1 January. ↩ ↩² ↩³

41. Nicole Junkermann to Jeffrey Epstein, 9 January 2019, 12:41 p.m., with his reply of 5:47 p.m. (“no suprise. do nothing for a week or two”) and her answer of 6:01 p.m. Text layer only. https://epstein-data.com/EFTA02605926 p.1. "Arbitration results Wu claims were fully rejected" returns 6 pages in 6 documents and all six were opened. ↩

42. Jeffrey Epstein to Nicole Junkermann, “do not pass on,” 22 January 2019, 00:59, quoted here in full, with her reply of 9:45 a.m. Text layer only. https://epstein-data.com/EFTA02626801 p.1. "Wenzhou" returns 15 pages in 13 documents corpus-wide; this is the only one in the name population, and the context line of the other 12 was read. ↩ ↩²

43. Jeffrey Epstein to Nicole Junkermann, 11 February 2019, 11:50 a.m. Text layer only. https://epstein-data.com/EFTA02629583 p.1. "should i stand down on really sport" returns 14 pages in 14 documents; two were opened and the other 12 were counted and not read. ↩

44. Nicole Junkermann to Jeffrey Epstein, 1 March 2019, 4:49 p.m., carrying his question of 28 February 2019, 13:25 in its quoted block, with his reply of 22:05:44 UTC. Text layer only. https://epstein-data.com/EFTA01032071 pp.1–2. "I would like to progress with China" returns 3 pages in 3 documents and all three were opened; the other two are https://epstein-data.com/EFTA02633157 and https://epstein-data.com/EFTA02633194. ↩ ↩²

45. Department of Health and Social Care, “Health technology expert panel meets for the first time”, published 19 November 2018, https://www.gov.uk/government/news/health-technology-expert-panel-meets-for-the-first-time. The notice states “Dr Ben Goldacre will chair the board” and, under “Dr Goldacre will be joined by:”, lists “Nicole Junkermann – founder of NJF Holdings, an international finance and investment company” among the members. This is the outside record and is cited as such; no document in the release states it, and the searches run to test that are at ⁴⁹. ↩ ↩²

46. This is a statement issued on her behalf to a named outlet, not a document in the release, and is cited as that. Erin Malik, “Lancaster University visiting professor named 3,475 in the Epstein files leaves position”, The Tab, 11 February 2026, article:published_time 2026-02-11T15:12:22+00:00, https://thetab.com/2026/02/11/lancaster-university-visiting-professor-who-was-named-3475-in-epstein-files-leaves-position. It prints a statement from a spokesperson for her in five sentences across three paragraphs, and all five are reproduced in the body exactly as that outlet prints them, in its own paragraphing, the first sentence included, which is missing a word: “The scale Jeffrey Epstein’s crimes is appalling.” Nothing is repaired. An earlier draft of this page quoted the first two sentences only, which left out both the account the statement gives of why the correspondence reads as it does and everything it says about his victims; the whole of it is now given, because a partial defence beside the record is worse than none. The same report gives Lancaster University’s statement of 10 February 2026 ending a visiting professorship she had held since September 2025. The second sentence is quoted again by Neerja Mishra, “Who is Nicole Junkermann? German Countess Resigns from Royal Marsden Charity Over Jeffrey Epstein Email Controversy”, The Sunday Guardian, 16 February 2026, https://sundayguardianlive.com/world/who-is-nicole-junkermann-german-countess-resigns-from-royal-marsden-charity-over-jeffrey-epstein-files-email-controversy-170701/, which reports her resignation as a trustee of the Royal Marsden Cancer Charity and prints “misled by him, and deeply regrets” with a comma after “him” where The Tab prints none; the body follows The Tab, and neither form is repaired. That outlet prints fragments rather than the whole statement — the word “appalling”, the second sentence, and the closing sentence in paraphrase rather than in quotation — so it corroborates The Tab’s text in part and is not a second full copy of it. No statement by her or on her behalf about Epstein’s offences appears in any document read for this page. An earlier version of this footnote called both outlets “secondary reporting of a statement whose own text this page has not seen”. That was wrong about the first of them, and it understated the page’s own best evidence. The Tab is the party the statement was given to. Its earlier piece by the same author introduces the statement with the words “A spokesperson for Nicole Junkermann told The Lancaster Tab”: Erin Malik, “Lancaster University visiting professor mentioned 3,475 times in the Epstein files”, The Tab, 9 February 2026, article:published_time 2026-02-09T14:13:48+00:00, https://thetab.com/2026/02/09/lancaster-university-visiting-professor-mentioned-3475-times-in-the-epstein-files. Its text is therefore the best copy of the statement available outside the spokesperson’s own paper, and not a re-report of somebody else’s copy. Both Tab pieces were fetched raw, not through a summarising reader, and the statement in each was compared with the block quote above word by word: under quotation-mark and whitespace normalisation, 675 characters on this page against 677 in each piece, the whole difference being the two quotation marks that outlet re-opens at its two paragraph breaks, which a block quote does not carry. The two pieces print the statement identically. The missing word in the first sentence is in the statement and not in the outlet, and that is now corroborated and not merely asserted. Virtual Lancaster, a local outlet independent of The Tab, which writes that the spokesperson “told media” rather than naming a recipient and so is not repeating The Tab’s framing, prints the same first sentence with the same word missing, in the same three paragraphs: John Freeman, “Lancaster University wipes ties with entrepreneur Nicole Junkermann named in Epstein Files”, Virtual Lancaster, 20 February 2026, https://virtual-lancaster.net/lancaster-university-wipes-ties-with-entrepreneur-nicole-junkermann-named-in-epstein-files/. It was fetched raw and compared the same way, with the same 675-against-677 result and no other difference. The only text found anywhere that carries the missing word is an aggregator, https://epsteintransparencyact.com/people/nicole-junkermann/, which prints “The scale of Jeffrey Epstein’s crimes is appalling” and silently harmonises the sources it collects. Two independent outlets print the defect and one aggregator repairs it. This page does not repair it, and the reader who notices the missing word can now tell from this footnote that it is not a typographical error of The Tab’s. ↩ ↩²

47. “Handelsblatt / INTERVIEW / ‘Al can have a disturbing effect’ — Investor Junkermann on the pros and cons of artificial intelligence”, by Peter Brors, datelined 27/11/2018 16:47, five sheets, an English translation, standing alone in the production with no covering message on an adjacent Bates number. Read at the image at 400 dpi, printed page 2, which carries the Hancock question and her answer, the “Introducing NJF Capital” panel and the 2018 transaction summary. https://epstein-data.com/EFTA00803367 p.2. The Magnum passage quoted in the body is on p.5 and the Winamax, Neuf Cegetel and Infront passages on p.1, both read in the text layer. "Handelsblatt" returns 32 pages in 24 documents corpus-wide, of which 3 are in the name population and all three were opened; the other two are copies of the Robert Kuhn thread at ³⁸, in which Kuhn mentions the newspaper. ↩ ↩² ↩³

48. A document refuting the statement in the body would be a message to or from a UK government address, or naming the Department of Health, the Health Secretary or the HealthTech Advisory Board, in which Epstein takes some part in her appointment. Searches run 11 September 2026, each scoped to the surname: "Junkermann" AND "NHS" 0 pages; "Junkermann" AND "HealthTech" 0 pages; "Junkermann" AND "Hancock" 1 page in 1 document, which is the interview at ⁴⁷; "Junkermann" AND "Department of Health" 1 page in 1 document, the same one; "Junkermann" AND "advisory board" 9 pages in 9 documents, all nine opened — the interview at ⁴⁷; the two copies of the Ben Goldacre exchange at ⁴⁹; the Reporty invitation at ³¹; four further Ehud Barak messages about the Reporty advisory board, at EFTA00680244, EFTA02487155, EFTA02704430 and EFTA02704699; and one message at EFTA02347006 about a different Israeli venture. 1 + 2 + 1 + 4 + 1 = 9. "Junkermann" AND "stepped down", "Junkermann" AND "resigned" and "Junkermann" AND "terminated" each return 0 pages. Nothing among the nine records any contact between Epstein and the UK government. ↩

49. Nicole Junkermann to Jeffrey Epstein, “What do you think about him?,” 3 November 2018, and his reply the same morning. Both sheets read at the image at 400 dpi. https://epstein-data.com/EFTA02616680 p.1 (her message) and https://epstein-data.com/EFTA02616649 p.1 (his reply, carrying hers in the quoted block). The clock. Each sheet’s own header field is a bare Sent: carrying no zone — hers “Saturday, November 3, 2018 10:36 AM”, his “Saturday, November 3, 2018 11:10 AM” — and each also prints an iPhone property-list trailer whose date-received value is a Unix epoch: 1541241372 on her sheet, which decodes to 10:36:12 UTC, and 1541243425 on his, which decodes to 11:10:25 UTC. Each epoch agrees with its own printed header to the minute, so the zone here is established by the sheets and not inferred. The release disagrees with itself on her stamp: the quoted attribution inside his reply reads “On Sat, Nov =, 2018 at 11:36 AM”, an hour later than her own sheet’s 10:36 and later than the 11:10 reply it sits beneath, which would put her message after the answer to it; the day digit in that same attribution is damaged to =. The top-level headers and the epochs are the reading taken here. "He will be my chairman" returns 2 pages in 2 documents and both were opened. The = characters in the quotation are quoted-printable transmission damage carried onto the face of the document and are reproduced as printed rather than repaired. The name Goldacre was searched across the whole corpus. "Goldacre" returns 10 pages in 5 documents, "Ben Goldacre" 7 pages in 5, and "Junkermann" AND "Goldacre" 2 pages in 2 documents, which are the two sheets above. All five documents were opened. The other three lie outside her name population and none is used on this page: an undated five-sheet transcript of an interview with Goldacre about Bad Pharma, publication bias and the AllTrials petition at https://epstein-data.com/EFTA01100068, which stands alone in the production with no covering message and names neither her nor Epstein, and whose internal references date it to 2013; a mass-mailed weekend reading circular of 7 April 2013, sent to undisclosed recipients, at https://epstein-data.com/EFTA00958500, whose attachment list includes “Dr_Ben_Goldacre.pdf” and whose own text introduces Goldacre in almost the same words the transcript opens with; and a book-jacket blurb quoting him at https://epstein-data.com/EFTA00640724. That overlap makes it likely the transcript reached these papers as an attachment to the circular rather than as a file kept on Goldacre; the release does not settle which, and nothing read here connects any of the three to the November 2018 exchange. The absence test for the board. A document refuting the statement in the body would be a sheet in the release that names the board she meant, or that puts Goldacre and a chairmanship in the same place. Run corpus-wide on 11 September 2026, not scoped to her surname: "Healthtech Advisory Board" returns 0 pages; "HealthTech Advisory" 0 pages; "Goldacre" AND "chair" 0 pages; and "advisory board for health" returns 2 pages in 2 documents, which are the two sheets above and nothing else. The conjunctions at ⁴⁸ cover the same ground scoped to the surname and also return nothing. ↩ ↩² ↩³

50. A comparable message of 2 July 2010, 12:21:12 -0400, to a different correspondent, opens “nciole junkerman , the tal german that i used to go out with is also”; it survives at https://epstein-data.com/EFTA00736222, https://epstein-data.com/EFTA00744072, https://epstein-data.com/EFTA00894501, https://epstein-data.com/EFTA01814983 and https://epstein-data.com/EFTA02410825, none read at the image. It is cited here only to show that the one-n spelling is Epstein’s own and recurs. ↩

51. Ehud Barak to James Brocklebank of Advent International, 20 March 2018, 19:28, forwarded into Epstein’s mailbox and answered by Epstein on 24 March 2018: “I was informed by Mrs. Junkerman of your interest in having Advent’s man in Israel and that my name was raised as a potential candidate.” The same thread carries Brocklebank’s line, “Felicia and I enjoyed meeting you at Nicole’s dinner a few months ago.” Text layer only, four copies: https://epstein-data.com/EFTA00853313 pp.1–3; https://epstein-data.com/EFTA00852946; https://epstein-data.com/EFTA00852997; https://epstein-data.com/EFTA00853028. Barak’s side of this belongs to ehud-barak. ↩