Epstein Revealed

Lawrence Krauss: 29,399 Mentions and the Origins Project

2026-05-11T00:00:00.000Z

TLDR

Lawrence M. Krauss, former Foundation Professor at Arizona State University and founding director of the ASU Origins Project, appears 29,399 times across the Epstein corpus — more than any other identified academic (PAPER TRAIL Project, 2026a). The Origins Project received Epstein-linked funding during Krauss's directorship (Aldhous et al., 2018); Krauss publicly defended Epstein after his 2008 conviction (Wolfe, 2011)^¹ and retired from ASU in May 2019 following a July 31, 2018 university finding that he had violated the sexual-harassment policy (Rapanut, 2018; Aldhous, 2018b).

The Foundation Professor and Origins

Krauss held the Foundation Professor title at ASU, directed the Origins Project from its 2008 launch, and maintained an office in the School of Earth and Space Exploration. Co-occurring entity matches in the corpus corroborate each role: "Arizona State University" (3,400), "asu" (5,025), "The Origins Project" (2,636), and "Project Co-Director" (1,931). BuzzFeed News reported Epstein as "one of the Origins Project's major donors."

At the time of his most heavily documented Epstein contacts, Krauss held the Foundation Professor title at ASU, directed the ASU Origins Project, and maintained an office in the Sciences School of Earth and Space Exploration (PAPER TRAIL Project, 2026a). The corpus corroborates each of these roles: co-occurring entity names include "Arizona State University" (3,400 co-document matches), "asu" (5,025), "The Origins Project" (2,636), "Sciences School of Earth and Space Exploration" (1,925), "Project Co-Director" (1,931), and "Dir, Beyond Center" (1,924).

Krauss launched the Origins Project in 2008 to host public science conferences featuring high-profile scientists. The project's funding came from private donations — some routed through ASU's Research Office, some through an external Origins Project Foundation. BuzzFeed News reported that Epstein was "one of the Origins Project's major donors" (Aldhous et al., 2018).

What the Corpus Shows

Krauss appears in 8,786 dated documents across the corpus, with 3,075 shared with "Jeffrey Epstein" — rising to approximately 10,400 when OCR variants are aggregated. The highest-mention documents concentrate in 2018-2019 iMessage extracts from Epstein's seized forensic phone: EFTA00786667.pdf (90 mentions), EFTA00785782.pdf (86), and EFTA00786509.pdf (84). Related name fragments include "Lawrence M. Krauss" (14,826) and "Krauss" (9,723).

Krauss is present in 8,786 dated documents across the corpus (PAPER TRAIL Project, 2026a). His strongest co-occurrence with "Jeffrey Epstein" spans 3,075 shared documents, rising to approximately 10,400 when OCR variants of Epstein's name ("JEFFREY," "jeffrey E," "jeffrey E. <") are aggregated. The highest-mention documents are concentrated in iMessage extracts from Epstein's seized forensic phone image dated 2018–2019: EFTA00786667.pdf (90 mentions), EFTA00785782.pdf (86), EFTA00786509.pdf (84), and EFTA00781867.pdf (80) (PAPER TRAIL Project, 2026a).

The co-occurring entity "Research Office" (2,393 shared documents) suggests significant ASU institutional correspondence in the corpus. This volume is compatible with — though not proof of — the 2018 sexual-harassment investigation (Rapanut, 2018). Krauss was released under a cluster-size of 1 by Script 19 entity resolution, meaning OCR fragmentation yielded multiple related entity records ("Lawrence Krauss," "Lawrence M. Krauss" with 14,826 mentions, "Krauss" with 9,723) that remain unconsolidated but are confidently the same person.

Epstein Text Messages

Forensic phone image NYC024362.aff4 yielded iMessages between Krauss and "Self" (resolved to jeeitunes@gmail.com). On August 4, 2018, Krauss wrote: "You will be integral during all my decisions." On September 17, 2018, as the ASU investigation intensified, he asked Epstein for "the name and phone number of one or two of the lawyers you have spoken to regarding action down the line." On October 10, 2018: "Thank you for everything."

The highest-mention Krauss documents (EFTA00785764.pdf, EFTA00785782.pdf, EFTA00786509.pdf, EFTA00786667.pdf, EFTA00781867.pdf) are extracted messages from a forensic phone-device image labeled NYC024362.aff4, containing iMessages between "Lawrence Krauss" and "Self" — where "Self" resolves to the Epstein-associated account jeeitunes@gmail.com. Krauss to Epstein, EFTA00785782.pdf, August 4, 2018: "You will be integral during all my decisions" (PAPER TRAIL Project, 2026a). On the day of the BuzzFeed News exposé, EFTA00785764.pdf, August 3, 2018: "Anyway more later. No more now. About to board." On September 13, 2018, as the ASU investigation intensified, EFTA00786509.pdf: "hi.. on plane home. lets talk tomorrow.. got a call last night" — followed later the same day by "Positive? Positive? they are all sympathetic it looks like." On September 17, 2018 (EFTA00786667.pdf), Krauss asked Epstein: "can you perhaps give me the name and phone number of one or two of the lawyers you have spoken to regarding action down the line. It would really help relieve Nancy's anger at the system if we could just talk to a few of them to see what the issues are." On October 10, 2018, EFTA00781867.pdf: "Thank you for everything" (PAPER TRAIL Project, 2026a).

The Public Record

In 2011, three years after Epstein's 2008 Florida guilty plea on solicitation charges, Krauss told The Daily Beast: "I don't feel tarnished in any way by my relationship with Jeffrey; I feel raised by it." On February 22, 2018, BuzzFeed News published harassment accounts from multiple women. ASU's July 31, 2018 finding concluded Krauss violated the sexual-harassment policy; he retired effective May 2019.

In 2011, Krauss told The Daily Beast, "I don't feel tarnished in any way by my relationship with Jeffrey; I feel raised by it" (Wolfe, 2011). He made the statement three years after Epstein's 2008 Florida guilty plea on solicitation charges.

On February 22, 2018, BuzzFeed News published detailed accounts from multiple women alleging sexual harassment by Krauss across more than a decade (Aldhous et al., 2018). ASU placed Krauss on paid administrative leave and prohibited him from being on campus during the subsequent investigation (Banks, 2018). In a letter dated July 31, 2018, ASU Executive Vice President and Provost Mark Searle concluded that Krauss had violated the university's sexual-harassment policy (Rapanut, 2018). ASU did not renew Krauss's Origins Project directorship (Rapanut, 2018). Krauss retired from ASU effective May 2019 (Aldhous, 2018b).

After ASU

Following his May 2019 ASU departure, Krauss reorganized public-facing operations through The Origins Project Foundation, an independent 501(c)(3) unaffiliated with Arizona State University. He has continued writing and public speaking but has not held a full faculty position since 2019 and has consistently denied the harassment allegations. Among academics resolved by direct seed-list matching, Krauss carries the highest mention count in the corpus.

Following his ASU departure, Krauss reorganized public-facing operations through The Origins Project Foundation, an independent 501(c)(3) that is not part of Arizona State University (Arizona State University, n.d.). He has continued writing and public-speaking but has not held a full faculty position since 2019. Krauss has consistently denied the harassment allegations (Aldhous et al., 2018).

Of the academics the corpus identifies by direct seed-list resolution, Krauss has the highest mention count (PAPER TRAIL Project, 2026a). His case is the clearest example of a professor whose financial, institutional, and rhetorical relationship with Epstein is documented across multiple evidence domains — and whose professional separation from his host institution was documented in near-real-time in the corpus itself.

CTA

Every EFTA document cited above is publicly released under the Epstein Files Transparency Act (Pub. L. No. 119-38, 2025). Readers can pull EFTA00785782.pdf, EFTA00786667.pdf, or any other identifier through the U.S. House Oversight Committee's Epstein disclosure portal and verify quoted passages in original context. The next profile in the PAPER TRAIL Academic Network series applies the same methodology to additional academics.

Every EFTA document cited above is publicly released under the Epstein Files Transparency Act (Pub. L. No. 119-38, 2025) — pull EFTA00785782.pdf, EFTA00786667.pdf, or any of the other identifiers through the U.S. House Oversight Committee's Epstein disclosure portal and verify the quoted passages in their original context. Once you have, the next profile in the PAPER TRAIL Academic Network series documents the same methodology applied to academics found in the Epstein Files.

About the Methodology

This profile cross-referenced the full-text EFTA corpus against a seeded list of publicly documented academic associates. Mention counts and co-occurrence statistics were computed from PostgreSQL queries against entity-resolution output applying Splink probabilistic record linkage to 2.38 million extracted entities. Direct quotes were verified verbatim against on-disk OCR markdown at partial_ratio >= 95. Journalism citations were independently WebFetch-verified for existence, date, authorship, and quoted text.

This profile was produced by cross-referencing the full-text Epstein Files corpus — released under the Epstein Files Transparency Act (Pub. L. No. 119-38, 2025) — against a seeded list of publicly documented academic associates. Mention counts, co-occurrence statistics, and top-document rankings were computed from PostgreSQL queries against the project's entity-resolution output, which applied Splink probabilistic record linkage to 2.38 million extracted entities. Every direct quote attributed to an EFTA document was verified verbatim against the on-disk OCR markdown using fuzzy-string matching at a partial_ratio ≥ 95 threshold. Every journalism citation was independently WebFetch-verified for article existence, exact publication date, correct authorship, and that the quoted text appears in the article body. Interpretive characterizations are attributed to their originating source; all factual statements are grounded in either a released EFTA document, an official ASU proceeding, or a named press report listed in the References section below.

Note on "publicly defended" ↩

The TLDR sentence "Krauss publicly defended Epstein after his 2008 conviction" resolves into three atomic factual claims, each grounded in the primary source.

1. "defended" — Wolfe (2011) quotes Krauss verbatim, on the record, in four distinct registers:

"The unfortunate period he suffered." (reframing Epstein's 13-month sex-offense sentence as hardship)

"They're not as young as the ones that were claimed." (disputing the age of accusers)

"My presumption is that I would believe him over other people." (credibility endorsement over victims)

"I don't feel tarnished in any way by my relationship with Jeffrey; I feel raised by it." (personal endorsement)

Wolfe's own editorial frame in the same article introduces Krauss as the example scientist for the sentence: "Renowned scientists whose research Epstein has generously funded through the years also stand by him" (Wolfe, 2011).

2. "publicly" — Wolfe (2011) is an on-the-record interview in The Daily Beast, a national news outlet; Krauss is quoted by name with no "off the record" or "background" caveat. The statements were republished by third parties within days and re-surfaced in 2018 by Aldhous et al. (2018). Krauss has not in fifteen years claimed the quotes were misattributed, taken out of context, or not intended for publication.

3. "after his 2008 conviction" — Epstein entered a Florida guilty plea on June 30, 2008 (two state counts including solicitation of a minor), completed a 13-month sentence on July 22, 2009, and the Wolfe interview was published on April 1, 2011 — two years and nine months after the plea, one year and eight months after release. Wolfe (2011) situates the statements as post-conviction in its own opening paragraphs.

Legal posture. Under Milkovich v. Lorain Journal Co., 497 U.S. 1 (1990), characterizing someone's disclosed statements as "defending" another person is protected opinion when the underlying statements are disclosed so a reader can evaluate them independently — which this note does. Under New York Times Co. v. Sullivan, 376 U.S. 254 (1964), Krauss as a public figure would have to prove actual malice, which cannot lie where the characterization tracks his own verbatim words.

Defensibility summary. The Daily Beast, April 1, 2011, quotes Professor Krauss saying, of convicted sex offender Jeffrey Epstein, "I don't feel tarnished in any way by my relationship with Jeffrey; I feel raised by it" and "my presumption is that I would believe him over other people" — published two years and nine months after Epstein's June 30, 2008 guilty plea. The TLDR sentence summarizes that interview accurately.

References

Aldhous, P. (2018b, October 21). Celebrity physicist Lawrence Krauss will retire from Arizona State in wake of harassment investigation. BuzzFeed News. https://www.buzzfeednews.com/article/peteraldhous/lawrence-krauss-retires-arizona-state-settlement

Aldhous, P., Ghorayshi, A., & Hughes, V. (2018, February 22). A star physicist has been accused of sexual misconduct. BuzzFeed News. https://www.buzzfeednews.com/article/peteraldhous/lawrence-krauss-sexual-harassment-allegations

Arizona State University. (n.d.). Origins Project at Arizona State University. Retrieved April 16, 2026, from https://origins.asu.edu

Banks, M. (2018, March 7). Lawrence Krauss banned from Arizona State University campus following misconduct allegations. Physics World. https://physicsworld.com/a/lawrence-krauss-banned-from-arizona-state-university-campus-following-misconduct-allegations/

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025). https://www.congress.gov/bill/119th-congress/house-bill/4405

PAPER TRAIL Project. (2026a). Professor evidence packet: Lawrence Krauss [Data set]. _exports/academics/professor_evidence/lawrence_krauss.json

PAPER TRAIL Project. (2026b). Academic identification: Seeded resolution of 30 university professors [Computer software]. app/scripts/34_academic_identification.py

Rapanut, K. (2018, August 7). Lawrence Krauss violated ASU sexual harassment policies, investigation shows. The State Press. https://www.statepress.com/article/2018/08/spscience-asu-investigation-states-lawrence-krauss-groped-woman-while-on-asu-funded-trip

Wolfe, A. (2011, April 1). Katie Couric, Woody Allen: Jeffrey Epstein's society friends close ranks. The Daily Beast. https://www.thedailybeast.com/katie-couric-woody-allen-jeffrey-epsteins-society-friends-close-ranks

This investigation is part of the SubThesis accountability journalism network.

Lithuania: The Thinnest Thread, Confirmed

2026-05-11T00:00:00.000Z

TLDR

Lithuania was initially the thinnest financial thread connecting Epstein to any investigating country — approximately EUR 90,000 in confirmed wire transfers to two entities directed by one person. On March 17, 2026, it became the first of five countries and thirty congressional offices to respond: Prosecutor Aurelijus Navickas confirmed materials were attached to an active criminal case file (Panevezys Regional Prosecutor's Office, 2026). Both wires are now T2 — confirmed in the DOJ-released corpus with Bates numbers, IBAN codes, and a bank statement.

Two Wires, Now Confirmed

Two Lithuanian wires are now T2-confirmed in the DOJ corpus. EUR 75,000 to UAB Fors Projektai exists in three copies (EFTA01365100, EFTA01365793, EFTA01409890), authorized by Darren Indyke for J. Epstein Virgin Islands Foundation Inc., dated January 20, 2017, routed to SEB Bank Vilnius. The Baleto Teatras wire from Gratitude America Ltd. (EFTA01362207) executed April 3, 2018: EUR 15,000 at 1.23618562, totaling $18,542.78.

The EUR 75,000 wire to UAB Fors Projektai exists in three corpus copies (EFTA01365100, EFTA01365793, EFTA01409890), authorized by Darren Indyke on behalf of J. Epstein Virgin Islands Foundation Inc., dated January 20, 2017, to SEB Bank Vilnius (PAPER TRAIL Project, 2026a). The wire to NPO Baleto Teatras exists as a Deutsche Bank Authorization for Asset Movement from Gratitude America Ltd. (EFTA01362207), confirmed by a bank statement showing execution on April 3, 2018: EUR 15,000 at an exchange rate of 1.23618562, totaling $18,542.78 (PAPER TRAIL Project, 2026a).

The original blog (March 10, 2026) cited the Baleto Teatras wire as $28,000 based on LRT reporting (LRT, 2026). Corpus validation on March 18, 2026, contradicted this figure. The bank statement shows $18,542.78 — a $9,457.22 discrepancy that the corpus documents do not explain. The LRT figure is superseded by the Deutsche Bank record.

Both transfers route through USVI foundations — J. Epstein Virgin Islands Foundation Inc. and Gratitude America Ltd. — to Lithuanian organizations directed by Valdas Petreikis. The authorization chain is documented: Indyke authorized Wire 1; Wire 2 was initiated after Petreikis emailed Richard Kahn (HBRK Associates) with Baleto Teatras banking details on April 2, 2018 (PAPER TRAIL Project, 2026a).

The $3 Million Bequest

Epstein's last will — signed August 8, 2019, forty-eight hours before his death — bequeathed $3 million to Simona Petreike, the single largest named bequest in the estate. Her family has publicly stated the bequest was refused. A bequest is a stated intention, not a completed transaction, and a refused bequest has no financial effect. But the $3 million combined with EUR 90,000 in business entity transfers indicates a relationship beyond casual acquaintance.

Epstein's last will and testament — signed August 8, 2019, forty-eight hours before his death — bequeathed $3 million to Simona Petreike (PAPER TRAIL Project, 2026b). This is the single largest named bequest in the estate. Her family has publicly stated the bequest was refused. A bequest is a stated intention, not a completed transaction — and a refused bequest has no financial effect. But the $3 million figure, combined with EUR 90,000 in business entity transfers, indicates a relationship that extended beyond casual acquaintance.

The First Response

On March 17, 2026, the Panevezys Regional Prosecutor's Office (Organized Crime and Corruption Investigation Division) formally acknowledged receipt of citizen-submitted corpus analysis. Prosecutor Aurelijus Navickas, signing with a qualified electronic signature, confirmed materials were "attached to the criminal case materials and forwarded for evaluation to the pre-trial investigation institution." Five countries were contacted on March 5, 2026 — Lithuania, the smallest economy and thinnest financial thread, is the only entity that responded.

On March 17, 2026, the Panevezys Regional Prosecutor's Office (Organized Crime and Corruption Investigation Division) formally acknowledged receipt of citizen-submitted corpus analysis. The response, signed by Prosecutor Aurelijus Navickas with a qualified electronic signature, confirmed that materials were "attached to the criminal case materials and forwarded for evaluation to the pre-trial investigation institution" (Panevezys Regional Prosecutor's Office, 2026).

The division assignment — Organized Crime and Corruption — classifies the Epstein network as organized crime of an international nature, aligning with Poland's Investigation Team No. 5 classification. Five countries were contacted on March 5, 2026. Lithuania, the smallest economy and the thinnest financial thread, is the only entity that responded.

What Changed

Three things changed since the original "thinnest financial thread" assessment based on T3 LRT journalism. First, both wires were independently confirmed in the DOJ-released corpus at T2, with Bates numbers, IBAN codes, SWIFT codes, and a Deutsche Bank execution record. Second, the LRT-reported $28,000 figure was contradicted by corpus data showing $18,542.78 — a $9,457 discrepancy. Third, a named prosecutor confirmed materials are part of an active criminal file.

The original blog called Lithuania "the thinnest financial thread" with T3 evidence drawn from LRT journalism. Three things changed. First, both wires were independently confirmed in the DOJ-released corpus (T2), with Bates numbers, IBAN codes, SWIFT codes, and a Deutsche Bank execution record. Second, the LRT-reported $28,000 figure was contradicted by corpus data showing $18,542.78. Third, a named prosecutor confirmed that the analytical materials are part of an active criminal file (Panevezys Regional Prosecutor's Office, 2026).

The thread is no longer thin. It is documented with primary source banking records from Deutsche Bank's grand jury production, and a government has acted on it.

References

LRT. (2026). Reports on Epstein-related bank transfers to Lithuanian entities since 2008. LRT (Lithuanian National Broadcaster). https://www.lrt.lt

Panevezys Regional Prosecutor's Office. (2026). RE: Proposal [Official correspondence]. Organized Crime and Corruption Investigation Division. communications/responses/2026-03-17_lithuania_panevezys_acknowledgment.md

PAPER TRAIL Project. (2026a). Lithuanian wire transfer corpus validation: 18-search results [Search exports]. _exports/search/search_Fors_projektai.csv, _exports/search/search_Baleto_Teatras.csv

PAPER TRAIL Project. (2026b). Last Will and Testament: $3M Petreike bequest [Corpus document]. .tmp/oversight_downloads/estate_records/Request No. 2.pdf

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025).

This research is sponsored by Subthesis.

EP02: The Pipeline — 1.18 Million Entities and the Missing Third

2026-05-11T00:00:00.000Z

TLDR

Episode 2 of PAPER TRAIL introduces the 16-script processing pipeline that converted 2.1 million raw government documents into structured, searchable data — extracting 1.18 million entities at the time of filming, revealing a 1.55:1 organization-to-person ratio, and estimating via Chao1 that over 468,000 entities remained unseen (PAPER TRAIL Project, 2026a).

What the Episode Covers

EP02 walks through a directed acyclic graph of 16 Python scripts organized into 8 processing stages. Stage 1 ingests raw metadata into PostgreSQL. Stage 2 converts PDFs and images to searchable markdown via OCR. Stage 3 runs NER using spaCy's en_core_web_lg across 8 parallel workers. Later stages deduplicate entities, build co-occurrence graphs, compute temporal density curves, and reprocess degraded documents through a vision-language model on a consumer GPU.

EP02 walks through the computational backbone of the project: a directed acyclic graph of 16 Python scripts organized into 8 processing stages. Stage 1 ingests raw metadata into PostgreSQL. Stage 2 converts PDFs and images to searchable markdown via OCR. Stage 3 runs named entity recognition using spaCy's en_core_web_lg model across 8 parallel workers. Later stages deduplicate entities, build co-occurrence graphs, compute temporal density curves, and reprocess degraded documents through a vision-language model on a consumer GPU (PAPER TRAIL Project, 2026b).

At the time of filming, the pipeline had processed 40.8% of the corpus — 857,000 documents — extracting 1.18 million entities linked by 29.5 million relationship pairs. The entity composition showed a striking inversion: 57% organizations, 37% persons, 6% locations, yielding a 1.55:1 organization-to-person ratio that reflects Epstein's reliance on corporate structures (PAPER TRAIL Project, 2026a).

Two critical gaps were identified at filming. Data Sets 9 and 11 — containing 863,000 email documents — sat at 0% NER coverage, the largest unprocessed sub-corpus, later closed by EP10. The second gap is structural: 197,945 entities appeared exactly once across the entire corpus — singletons representing 24.1% of all extracted entities. Some are OCR errors, some aliases, some real individuals in a single document — the pipeline cannot distinguish without additional context.

EP02 identifies two critical gaps in the processing state. Data Sets 9 and 11 — containing 863,000 email documents — sat at 0% NER coverage, the largest unprocessed sub-corpus. The episode flags these as "the richest veins unmined," a gap later closed when NER completed across all email documents by EP10 (PAPER TRAIL Project, 2026c).

The second blind spot is structural: 197,945 entities appeared exactly once in the entire corpus — singletons representing 24.1% of all extracted entities. Some are OCR errors. Some are aliases. Some are real individuals mentioned in a single document. The pipeline cannot distinguish between these categories without additional context, making the singleton population a persistent source of uncertainty (PAPER TRAIL Project, 2026a).

The Chao1 Estimate

The Chao1 species richness estimator — developed in ecology to estimate total species from partial samples — projected 1,290,141 total entities from 821,633 observed, meaning 63.7% of the entity population had been surfaced and 468,000 remained unseen. The three-step formula counts singletons (f1 = 197,945), counts doubletons (f2 = 41,816), and computes the estimate. Every finding derived from this corpus is drawn from roughly two-thirds of available evidence.

The episode's most consequential number is not what the pipeline found but what it estimated was missing. The Chao1 species richness estimator — originally developed in ecology to estimate the total number of species from partial samples — projected 1,290,141 total entities from 821,633 observed, meaning 63.7% of the entity population had been surfaced and 468,000 remained unseen (Chao, 1984; PAPER TRAIL Project, 2026d).

The three-step formula is presented visually: count singletons (f1 = 197,945), count doubletons (f2 = 41,816), compute the estimate. The result reframes the entire project: every finding derived from the corpus is drawn from roughly two-thirds of the available evidence. The remaining third could confirm, contradict, or transform any conclusion.

Why This Episode Matters

EP02 establishes the evidentiary foundation for every episode that follows. The pipeline is not a black box — every script, parameter, and threshold is named and explained. Processing gaps are disclosed before findings are presented. The Chao1 estimate sets the project's epistemic ceiling: no finding from this corpus can claim completeness. The pipeline knows what it has found, and it knows approximately how much it has not.

EP02 establishes the evidentiary foundation for every episode that follows. The pipeline is not a black box — every script, every parameter, every threshold is named and explained. The processing gaps are disclosed before the findings are presented. And the Chao1 estimate sets the project's epistemic ceiling: no finding from this corpus can claim completeness. The pipeline knows what it has found. It also knows, approximately, how much it has not.

References

Chao, A. (1984). Nonparametric estimation of the number of classes in a population. Scandinavian Journal of Statistics, 11(4), 265–270.

PAPER TRAIL Project. (2026a). EP02 slide content: Entity composition, singleton analysis, Chao1 estimation [Presentation]. communications/ep02_slides/

PAPER TRAIL Project. (2026b). Processing pipeline: 16 scripts in 8 stages [Computer software]. app/scripts/

PAPER TRAIL Project. (2026c). NER processing status: DS9 and DS11 email completion [Data set]. app/scripts/04_extract_entities.py

PAPER TRAIL Project. (2026d). Chao1 species richness estimation [Data set]. _exports/validation/chao1_summary.json

This research is sponsored by Subthesis.

EP03: Ghosts in the Machine — When Machines Hallucinate

2026-05-11T00:00:00.000Z

TLDR

Episode 3 of PAPER TRAIL confronts the pipeline's most dangerous failure mode: phantom entities generated by OCR hallucination. Of the top 25 highest-mention entities, 12 were phantoms — artifacts of blank forms, repeated boilerplate, and degraded handwriting that the pipeline mistook for real people, organizations, and locations (PAPER TRAIL Project, 2026a).

What the Episode Covers

EP03 categorizes four OCR hallucination vectors: blank government forms with pre-printed field labels (e.g., "Persons/Non" extracted as an entity), repeated headers/footers mangled across hundreds of pages, handwritten Deutsche Bank KYC text producing unreadable output, and image degradation from multi-generation photocopies. Two case studies anchor the episode: OBS-5 ("Poland" from a blank KYC form's "Persons/Non" label) and OBS-6 ("Krakow" from 190 pages of mangled "Resource Management Account" UBS headers).

EP03 opens with the 42.7% singleton rate and works backward to its cause. The pipeline's OCR stage — converting scanned PDFs and images to searchable text — introduces systematic errors that propagate through every downstream analysis. The episode categorizes four hallucination vectors: blank government forms with pre-printed field labels (e.g., "Persons/Non" extracted as an entity), repeated document headers and footers mangled across hundreds of pages, handwritten text on Deutsche Bank KYC forms producing unreadable output, and general image degradation from multi-generation photocopies (PAPER TRAIL Project, 2026a).

Two case studies anchor the episode. OBS-5 documented a finding that the word "Poland" appeared in Deutsche Bank compliance records, suggesting a geographic connection. Investigation revealed the source was EFTA01268833.pdf page 17 — a blank KYC form where the pre-printed field label "Persons/Non" was OCR-misread as "Poland." The form was stamped CLOSED 12/17/19 and contained no handwritten data. OBS-6 reported "Krakow" appearing in UBS account statements. The source was EFTA01275697.pdf — 190 pages of Ghislaine Maxwell's UBS records where the repeated header "Resource Management Account" was mangled into "Krakow" by the OCR engine (PAPER TRAIL Project, 2026b).

The Retractions

Both OBS-5 and OBS-6 were publicly retracted with full disclosure of how each error occurred and why it was not caught earlier. The episode presents retraction as a methodological feature, not a failure — a pipeline that cannot detect its own hallucinations is unreliable, while one that detects, discloses, and retracts them demonstrates exactly the self-correction the Daubert standard requires. A morphing animation shows "Persons/Non" transforming character by character into "Poland."

Both observations were retracted — publicly, with full disclosure of how the error occurred and why it was not caught earlier. The episode presents the retraction as a feature of the methodology, not a failure. A pipeline that cannot detect its own hallucinations is unreliable. A pipeline that detects them, discloses them, and retracts them is demonstrating exactly the kind of self-correction that the Daubert standard requires (PAPER TRAIL Project, 2026b).

The morphing animation — showing "Persons/Non" transforming character by character into "Poland" — visualizes how a single OCR misread cascades through entity extraction, deduplication, and graph construction. By the time the phantom reaches the co-occurrence network, it appears as a real geographic entity with dozens of relationships to legitimate entities.

The VLM Solution

EP03 introduces Qwen2.5-VL-7B running on a consumer RTX 4070 with 8 GB VRAM as the mitigation layer. Unlike traditional OCR, a VLM processes documents as images, understanding layout, distinguishing filled from blank fields, and reading handwritten text with higher fidelity. Script 08 reprocesses documents flagged as high-risk for OCR artifacts. The episode is explicit about limitations: VLM runs at roughly 1/10th batch OCR speed, requires GPU resources, and introduces its own error modes.

EP03 introduces the vision-language model (Qwen2.5-VL-7B running on a consumer RTX 4070 with 8 GB VRAM) as the mitigation layer. Unlike traditional OCR, a VLM processes the document as an image, understanding layout, distinguishing filled from blank fields, and reading handwritten text with higher fidelity. Script 08 reprocesses documents flagged as high-risk for OCR artifacts, providing a second opinion that can override the original transcription (PAPER TRAIL Project, 2026c).

The episode is explicit about the VLM's limitations: it runs at roughly 1/10th the speed of batch OCR, requires GPU resources, and introduces its own error modes. It is a correction layer, not a replacement. The 29.5 million entity relationships computed in EP02 still rest on the original OCR foundation — the VLM catches errors but does not retroactively fix every downstream computation.

Why This Episode Matters

EP03 establishes the project's intellectual honesty standard: retractions before findings, pipeline failure modes named and demonstrated. The audience learns to distrust high-mention entities concentrated in contiguous document ranges — a heuristic applicable to any large-scale document analysis, not just this corpus. Every subsequent episode inherits this disclosure: the data is imperfect, the pipeline makes errors, and the methodology includes mechanisms to catch them.

EP03 establishes the project's intellectual honesty standard. The retractions are presented before the findings, not after. The pipeline's failure modes are named, categorized, and demonstrated. The audience learns to distrust high-mention entities concentrated in contiguous document ranges — a heuristic that applies to any large-scale document analysis, not just this corpus. Every subsequent episode inherits this disclosure: the data is imperfect, the pipeline makes errors, and the methodology includes mechanisms to catch them.

References

PAPER TRAIL Project. (2026a). EP03 slide content: OCR hallucination vectors, singleton analysis, phantom entity detection [Presentation]. communications/ep03_slides/

PAPER TRAIL Project. (2026b). Observations log: OBS-5 and OBS-6 retracted [Research document]. OBSERVATIONS.md

PAPER TRAIL Project. (2026c). VLM reprocessing pipeline [Computer software]. app/scripts/08_vlm_reprocess.py

This research is sponsored by Subthesis.

EP10: 863,000 Emails — The Post-Conviction Network

2026-05-11T00:00:00.000Z

TLDR

Episode 10 of PAPER TRAIL covers the 863,000-email sub-corpus spanning Data Sets 9 and 11 — scanned printouts processed through Relativity litigation software, extending the operational timeline to 2017 and revealing that Epstein's network continued functioning after his 2008 conviction. NER coverage jumped from 0% (EP02) to 100%, adding 19.2 million entity mentions from 1,048,234 documents (PAPER TRAIL Project, 2026a).

What the Episode Covers

DS9 contains 531,256 email PDFs (94.5 GB); DS11, originally described by DOJ as "financial ledgers and USVI flight manifests," actually contains 331,655 seized email correspondence PDFs — a misidentification resolved as OBS-7. These are scanned printouts, not native email files: physical pages printed, collected during search warrants, then digitized. Each PDF is an island with no Message-ID headers, no In-Reply-To fields, no threading metadata. Relativity Bates stamps confirm pre-release litigation review.

EP10 returns to the largest blind spot identified in EP02: Data Sets 9 and 11 sitting at 0% NER coverage. DS9 contains 531,256 email PDFs (94.5 GB). DS11, originally described by the DOJ as "financial ledgers and USVI flight manifests," actually contains 331,655 PDFs of seized email correspondence — a misidentification resolved as OBS-7. Together they constitute the largest sub-corpus by document count (PAPER TRAIL Project, 2026a).

These are not native email files. They are scanned printouts — physical pages that were printed, collected during search warrants, and later digitized. Each PDF is an island: no Message-ID headers, no In-Reply-To fields, no threading metadata. The Relativity load file format (Bates-stamped EFTA_R1_xxxxxxxx identifiers) confirms they were processed through litigation review software before release, meaning someone — the FBI, prosecutors, or a contractor — reviewed them before they became public (PAPER TRAIL Project, 2026b).

Thread Reconstruction Without Headers

Without native email headers, the pipeline uses four heuristic dimensions to rebuild conversation trees. First, temporal sequence — replies must postdate originals (hard constraint). Second, subject normalization — strip Re:, Fwd:, FW:, AW: and match base subject lines. Third, participant intersection — Jaccard similarity of sender/recipient arrays. Fourth, body text quotation — TF-IDF cosine similarity at tau = 0.85, with the high threshold accounting for OCR degradation. The methodology is designed but not yet executed at scale.

The episode's technical centerpiece is the thread reconstruction methodology. Without native email headers, the pipeline uses four heuristic dimensions to rebuild conversation trees from unlinked documents. First, temporal sequence: a reply must postdate the message it replies to (hard constraint). Second, subject normalization: strip Re:, Fwd:, FW:, AW: prefixes and match on the base subject line. Third, participant intersection: Jaccard similarity of sender/recipient arrays measures whether the same people are involved. Fourth, body text quotation: TF-IDF cosine similarity (tau = 0.85) detects quoted text in replies, with the high threshold accounting for OCR degradation of quoted passages (PAPER TRAIL Project, 2026c).

The methodology is designed but not yet executed at full scale across all 863,000 emails. The episode is transparent about this distinction: NER entity extraction is complete, but structured thread reconstruction and communication network analysis remain pipeline steps to be run.

The Communication Network Design

EP10 defines four social network analysis metrics for converting extracted email headers into a weighted directed graph: degree centrality (communication volume), betweenness centrality (information flow control), Burt's constraint (brokerage capacity), and Louvain community detection (functional sub-groups). The episode predicts actor patterns before results: Lesley Groff as administrative hub, Darren Indyke as strategic gatekeeper, Richard Kahn as financial gatekeeper, Epstein as central node — predictions the actual network structure will confirm or refute.

EP10 presents the framework for converting extracted email headers into a weighted directed graph. Four social network analysis metrics are defined: degree centrality (communication volume), betweenness centrality (information flow control), Burt's constraint (brokerage capacity), and Louvain community detection (functional sub-groups). The episode predicts actor patterns: Lesley Groff as administrative hub (high degree, high constraint), Darren Indyke as strategic gatekeeper (high betweenness, low constraint), Richard Kahn as financial gatekeeper, and Epstein as central node (PAPER TRAIL Project, 2026c).

These are predictions stated before results — the methodology design phase that EP12 later frames as essential to the Daubert standard. The actual network structure will confirm, refine, or contradict these predictions.

Why This Episode Matters

EP10 closes the largest data gap in the series — the jump from 0% to 100% email NER coverage transforms the pipeline's completeness. The jeevacation@gmail.com handle anchors Epstein's personal communications. The post-conviction timeline extending to 2017 documents continued network operations during the period when public attention had faded. The thread reconstruction methodology, even unexecuted, establishes how scanned printouts can be converted into conversation structures — applicable to any litigation document production.

EP10 closes the largest data gap in the series. The jump from 0% to 100% email NER coverage transforms the pipeline's completeness. The jeevacation@gmail.com handle anchors Epstein's personal communications. The post-conviction timeline (extending to 2017) documents continued network operations during the period when public attention had faded. And the thread reconstruction methodology — even in its designed but unexecuted state — establishes how scanned printouts can be converted back into conversation structures, a technique applicable to any large-scale litigation document production.

References

PAPER TRAIL Project. (2026a). EP10 slide content: Email corpus analysis, NER completion, DS9/DS11 composition [Presentation]. communications/ep10_slides/

PAPER TRAIL Project. (2026b). DS11 content resolution (OBS-7) [Observation log]. OBSERVATIONS.md

PAPER TRAIL Project. (2026c). Email network analysis methodology: Thread reconstruction and communication network design [Research document]. research/EMAIL_NETWORK_ANALYSIS.md

This research is sponsored by Subthesis.

EP12: The Method — 0.84 and the Line Between Computation and Evidence

2026-05-11T00:00:00.000Z

TLDR

Episode 12 of PAPER TRAIL explains how computational findings become evidence: the Daubert standard for legal admissibility, five quality gates the pipeline must pass, a bootstrap calibration protocol using 50 verified dates and 46 anchor entities, and the corrections made when the pipeline's own outputs were wrong — including 2 retractions, 5 date corrections, and 1 refutation (PAPER TRAIL Project, 2026a).

The Daubert Standard

The Daubert standard, established in Daubert v. Merrell Dow Pharmaceuticals (1993), sets four pillars for expert testimony: testable methodology (deterministic pipeline produces same output for same input), known error rate (F1, precision, recall at every stage), peer-reviewable methods (all algorithms published in scientific literature), and general acceptance (Bayesian probability, Leiden, Splink, PELT are standard tools). EP12 applies these pillars to a computational pipeline as a minimum standard for intellectual honesty.

The episode opens with 0.84 — the minimum F1 score required for entity resolution before any finding is presented. Below this line, the pipeline's outputs are not trusted. The number derives from the Daubert standard, established by the Supreme Court in Daubert v. Merrell Dow Pharmaceuticals (1993), which sets four pillars for expert testimony: the methodology must be testable (deterministic pipeline, same input produces same output), the error rate must be known (F1, precision, recall computed at every stage), the methods must be peer-reviewable (all algorithms published in scientific literature), and the methodology must be generally accepted (Bayesian probability, Leiden, Splink, PELT are standard tools) (PAPER TRAIL Project, 2026a).

EP12 applies these pillars to a computational pipeline rather than a human expert. The project was not designed for court, but the episode argues that the Daubert standard is a minimum standard for intellectual honesty: findings that cannot survive legal scrutiny should not survive public scrutiny either.

Five Quality Gates

The pipeline enforces five checkpoints with named thresholds. Gate 1: entity resolution F1 > 0.84 (designed). Gate 2: network modularity Q > 0.30 (tested — 125,620 communities). Gate 3: document retrieval recall 75–80% (designed). Gate 4: calibration anchor alignment > 85% (50 dates, 5 corrected, 89.8% achieved). Gate 5: compound error propagation within bounds (worst-case 0.9162). The transparency about tested versus designed gates is itself part of the methodology.

The pipeline enforces five checkpoints, each with a named threshold. Gate 1: entity resolution F1 > 0.84 (designed, not formally tested at production scale). Gate 2: network modularity Q > 0.30 (testable — 125,620 communities computed). Gate 3: document retrieval recall 75–80% (designed). Gate 4: calibration anchor alignment > 85% (50 dates calibrated, 5 corrected, 89.8% alignment achieved). Gate 5: compound error propagation within bounds (model designed, worst-case ceiling 0.9162) (PAPER TRAIL Project, 2026a).

The episode is explicit about which gates have been tested and which remain pending. This transparency — stating "designed but not formally validated" alongside "tested and passed" — is itself part of the methodology. A quality gate that claims to have been passed when it has not been tested is worse than no gate at all.

Calibration and Corrections

The calibration protocol uses 50 verified dates from primary government and court sources as fixed reference points. Five were wrong. PELT temporal analysis initially produced 889 breakpoints; recalibration with optimal penalty 0.2069 reduced this to 610 with 89.8% anchor alignment, passing the 85% threshold. Two observations retracted (OBS-5, OBS-6). One hypothesis refuted (OBS-1, Robert Crumb FedEx). The N212JE tail number recycling was caught by external corroboration, not internal validation.

The calibration protocol uses 50 verified dates from primary government and court sources as fixed reference points. Five were wrong. The PELT temporal analysis initially produced 889 breakpoints; recalibration with an optimal penalty of 0.2069 reduced this to 610, with 89.8% anchor alignment — passing the 85% threshold (PAPER TRAIL Project, 2026b).

The corrections are presented as evidence of methodological integrity. Two observations retracted (OBS-5 and OBS-6, OCR hallucinations). One hypothesis refuted (OBS-1, the Robert Crumb FedEx finding). The N212JE tail number recycling — two different aircraft using the same registration — was caught by external corroboration research, not by the pipeline itself. The episode uses this as a case study: no pipeline is complete. External verification catches what internal validation misses (PAPER TRAIL Project, 2026c).

Why This Episode Matters

EP12 is the methodological backbone — every finding in every other episode depends on the standards established here. The Chao1 estimate (63.7% complete) means every conclusion is drawn from partial evidence. The compound error model means individually valid stages accumulate uncertainty. The retraction standard — publish errors with the same prominence as original findings — sets the project's credibility floor. A claim without an error rate is not evidence, it is a guess.

EP12 is the series' methodological backbone. Every finding in every other episode depends on the standards established here. The Chao1 estimate (63.7% complete) means every conclusion is drawn from partial evidence. The compound error model means even individually valid stages accumulate uncertainty. And the retraction standard — publish the error with the same prominence as the original finding — sets the project's credibility floor. A claim without an error rate is not evidence. It is a guess.

References

Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).

PAPER TRAIL Project. (2026a). EP12 slide content: Daubert standard, quality gates, calibration protocol [Presentation]. communications/ep12_slides/

PAPER TRAIL Project. (2026b). PELT diagnostic: Optimal penalty calibration and anchor alignment [Data set]. _exports/temporal/anchor_alignment_report.csv

PAPER TRAIL Project. (2026c). Corroboration report: External source verification [Research document]. research/CORROBORATION_REPORT.md

This research is sponsored by Subthesis.

EP13: Convergence — Six Domains, 232,083 Events, Zero Findings

2026-05-11T00:00:00.000Z

TLDR

Episode 13 of PAPER TRAIL aligns six evidence domains — bank documents, wire transfers, FedEx shipments, flight logs, corporate filings, and institutional findings — into a unified event registry of 232,083 events with 143,791 entity-event bridge rows. Nine cross-domain leads were identified, three ACH hypotheses scored positive, and zero findings were declared because all leads fell below the 0.75 adjusted confidence threshold imposed by Chao1's 63.7% completeness estimate (PAPER TRAIL Project, 2026a).

The Silo Problem

Script 25b built a unified event registry of 232,083 events with 143,791 entity-event bridge rows mapping 898 entities verified across multiple domains. For twelve episodes, each evidence domain — Deutsche Bank compliance, FedEx shipments, aircraft, emails — was analyzed in isolation. EP13 connects the silos, asking what happens when a wire, a shipment, and a corporate filing involve the same entity within 48 hours.

For twelve episodes, each evidence domain was analyzed in isolation. EP04 examined Deutsche Bank compliance. EP06 mapped FedEx shipments. EP08 traced aircraft. EP10 processed emails. Each domain revealed patterns, but patterns within a single domain can be coincidental. EP13 asks what happens when the silos connect — when a wire transfer, a FedEx shipment, and a corporate filing all involve the same entity within the same 48-hour window (PAPER TRAIL Project, 2026a).

Script 25b built the unified event registry. Each event — a wire transfer, a shipment, a bank document, a flight, a corporate filing, an institutional finding — receives a standardized record with timestamp, entity identifiers, domain label, and source document reference. The entity-event bridge maps 898 entities verified across multiple domains, creating cross-domain profiles that no single data source could produce (PAPER TRAIL Project, 2026b).

ACH and Monte Carlo

Three ACH matrices scored positive: Spoliation +5.20 (FedEx October 2005 cutoff), Willful Blindness +6.40 (Deutsche Bank), Asset Concealment +4.70 (Butterfly Trust). Monte Carlo simulation with 5,000 iterations confirmed all three verdicts at 100% stability. Bayesian posteriors corroborated: P=0.90, P=0.77, P=0.69. Ten cross-domain contradictions were detected; zero were genuine — all explained by the FedEx data cutoff.

Three Analysis of Competing Hypotheses matrices test the dominant explanations for three observed patterns. Why did FedEx shipments stop in October 2005? Spoliation scores +5.20, ahead of channel migration at +1.90. Why did Deutsche Bank compliance fail for over a decade? Willful blindness scores +6.40, ahead of revenue capture at +4.50. What was the Butterfly Trust used for? Asset concealment scores +4.70, ahead of structured payments at +3.60 (PAPER TRAIL Project, 2026a).

Monte Carlo simulation (5,000 iterations with Beta distributions) confirmed all three ACH verdicts at 100% stability — no iteration reversed the winning hypothesis. Bayesian Belief Network posteriors corroborated: Spoliation P=0.90, Willful Blindness P=0.77, Asset Concealment P=0.69. Ten cross-domain contradictions were detected; zero were true contradictions — all ten are explained by the FedEx October 2005 data cutoff (PAPER TRAIL Project, 2026b).

Zero Findings

Zero findings is the episode's headline number. Despite nine cross-domain leads, three positive ACH verdicts, and 898 multi-domain entity profiles, every lead fell below the 0.75 adjusted confidence threshold because Chao1's 63.7% completeness penalty reduces all scores proportionally. With 36.3% of the entity population unobserved, the pipeline distinguishes leads (patterns worth investigating) from findings (patterns meeting confidence thresholds with known error bounds).

The episode's most important number is zero. Despite nine cross-domain leads, three positive ACH verdicts, and 898 multi-domain entity profiles, the pipeline declares zero findings. Every lead falls below the 0.75 adjusted confidence threshold because the Chao1 completeness penalty reduces all confidence scores proportionally. When you have observed only 63.7% of the entity population, the missing 36.3% could change any conclusion (PAPER TRAIL Project, 2026a).

This is the distinction between a lead and a finding. A lead is a pattern worth investigating. A finding is a pattern that meets the confidence threshold with known error bounds. The pipeline identifies patterns. The determination of what those patterns mean belongs to authorized investigators with access to the remaining 42% of unreleased documents.

Why This Episode Matters

EP13 demonstrates honest cross-domain synthesis: real patterns identified, none promoted to findings. 575 Lexington Avenue appears across three independent domains. Indyke and Kahn thread through financial and shipping records. The Butterfly Trust spans wires, bank documents, and corporate filings. The pipeline refuses to call these findings until the math supports it — a refusal that demonstrates the methodology working as designed, not failing.

EP13 demonstrates what cross-domain synthesis looks like when applied honestly. The patterns are real — 575 Lexington Avenue appears in three independent domains, Indyke and Kahn thread through financial and shipping records, the Butterfly Trust spans wires, bank documents, and corporate filings. But the pipeline refuses to call them findings until the math supports it. That refusal is the methodology working as designed.

References

PAPER TRAIL Project. (2026a). EP13 slide content: Cross-domain synthesis, ACH matrices, convergence profiles [Presentation]. communications/ep13_slides/

PAPER TRAIL Project. (2026b). Cross-domain synthesis engine output [Data set]. _exports/synthesis/

This research is sponsored by Subthesis.

EP14: What Remains — 42% and the Pipeline That Waits

2026-05-11T00:00:00.000Z

TLDR

Episode 14, the series finale, confronts the 42% gap: the DOJ identified over 6 million potentially responsive pages and released approximately 3.5 million. The pipeline that processed 2.1 million documents — extracting 2,383,898 entities, mapping 29.5 million relationships, detecting 125,620 communities — is built, tested, and ready to process whatever comes next (PAPER TRAIL Project, 2026a).

The 42% Gap

DOJ identified over 6 million potentially responsive pages and released approximately 3.5 million — a 42% gap. The Epstein Files Transparency Act (P.L. 119-38) passed 427 to 1, the most bipartisan vote of the 118th Congress. Specific known gaps include 50+ withheld FBI Trump-related interview pages, three of four missing FBI 302 summaries, six names redacted as "likely incriminated," and $1.08 billion in Treasury wire transfers absent from the public corpus.

The number defines the series' conclusion. The DOJ's own figures document the disparity: six million pages identified as potentially responsive, three and a half million released. The Epstein Files Transparency Act (P.L. 119-38) passed 427 to 1 — the most bipartisan vote of the 118th Congress — mandating release. Compliance is the question (Epstein Files Transparency Act, Pub. L. No. 119-38, 2025).

The known gaps are specific. NPR reported 50+ pages of FBI interviews related to Trump accusations withheld or removed. Three of four FBI 302 summaries are missing. Representatives Khanna and Massie reviewed unredacted files at the DOJ and identified six men whose names remain redacted — described as "likely incriminated." Senator Wyden's Treasury investigation documented $1.08 billion in 4,725 wire transfers through Epstein accounts, but those Treasury records are not in the public corpus (PAPER TRAIL Project, 2026a).

At the same time, 31+ child victims' identities were exposed through redaction failures — the exact inverse of the statute's intent. Victims exposed. Perpetrators protected.

The Series in Numbers

The series produced 2,383,898 entities at 97.4% NER coverage, 519,000 Splink clusters, 125,620 Leiden communities, and 535,318 structural brokers. Financial findings include 224 wire transfers totaling $24.1 million, $47.3 million in a single TD Bank SAR, 2,894 FedEx shipments, 863,000 emails, 53 corporate entities, and 8 aircraft. Corrections: 2 retractions (OBS-5, OBS-6), 5 calibration corrections, 1 refutation (OBS-1).

EP14 recaps what 14 episodes and 6 evidence domains produced. 2,383,898 entities extracted at 97.4% NER coverage. 519,000 entity clusters resolved via Splink probabilistic linkage. 125,620 Leiden communities detected. 535,318 structural brokers identified via Burt's constraint. 224 wire transfers totaling $24.1 million parsed. $47.3 million in suspicious activity documented by a single TD Bank SAR. 2,894 FedEx shipments tracked. 863,000 emails indexed. 53 corporate entities mapped. 8 aircraft traced through FAA records (PAPER TRAIL Project, 2026a).

The corrections: 2 observations retracted (OBS-5 and OBS-6, OCR hallucinations). 5 calibration dates corrected. 1 hypothesis refuted (OBS-1, the Robert Crumb FedEx finding). Every error disclosed in the episode where it was discovered.

Ready to Process

The pipeline has already run across every released document and can process unreleased files the moment they arrive — by DOJ compliance, congressional subpoena, or court order. No new infrastructure is needed. Chao1 projects 468,000 entities in the unseen corpus portion. The German Tank Problem flags sequential document gaps consistent with spoliation. These are statistical projections from observed data with known confidence intervals, not speculation.

The episode's title — "Ready To Process The Unreleased Epstein Files When Released" — is a statement of capability. The pipeline is not theoretical. It has already run across every released document. The same entity resolution, the same cross-domain synthesis, the same ACH scoring, the same error disclosure can process new documents the moment they are released — whether by DOJ compliance, congressional subpoena, or court order. No new infrastructure is needed. The unreleased files are the input. The pipeline is the machine (PAPER TRAIL Project, 2026b).

The Chao1 estimator projects 468,000 entities in the unseen portion of the corpus. The German Tank Problem flags sequential document gaps consistent with spoliation. These are not speculations — they are statistical projections from the observed data, with known confidence intervals.

Why This Episode Matters

EP14 closes the series by refusing to close the investigation. Across 14 episodes, the pipeline identified patterns, tested hypotheses, disclosed errors, and declared zero findings because the confidence threshold was not met. It states explicitly it is ready to run again. The remaining question is not whether the pipeline works, but whether the unreleased 42% of documents will be released — a question for Congress, the courts, and DOJ.

EP14 closes the series by refusing to close the investigation. The pipeline identified patterns. It tested hypotheses. It disclosed errors. It declared zero findings because the confidence threshold was not met. And it states, explicitly, that it is ready to run again. The question the series leaves is not whether the pipeline works — 14 episodes demonstrated that it does. The question is whether the remaining 42% of documents will ever be released for it to process. That question belongs to Congress, the courts, and the Department of Justice.

References

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025).

PAPER TRAIL Project. (2026a). EP14 slide content: Series summary, known gaps, pipeline capabilities [Presentation]. communications/ep14_slides/

PAPER TRAIL Project. (2026b). Processing pipeline: 33 scripts, complete processing state [Computer software]. app/scripts/

This research is sponsored by Subthesis.

Lithuania First Response: Corpus Analysis Attached to Criminal Case File

2026-05-11T00:00:00.000Z

TLDR

On March 17, 2026, the Panevezys Regional Prosecutor's Office (Organized Crime and Corruption Investigation Division) formally acknowledged receipt of citizen-submitted corpus analysis, attaching materials to criminal case [case number redacted] and forwarding them for evaluation (Panevezys Regional Prosecutor's Office, 2026). This is the first government response from any of the five countries and thirty congressional offices contacted by the PAPER TRAIL project.

The Letter

The response is one sentence, signed by Prosecutor Aurelijus Navickas with a qualified electronic signature under Lithuanian public key infrastructure. Translated: "We inform you that your proposal has been attached to the criminal case materials and forwarded for evaluation to the pre-trial investigation institution conducting the pre-trial investigation." Every phrase matters procedurally — attached materials become citable in motions and warrant applications, and the Organized Crime and Corruption Division assignment classifies the network accordingly.

The response is one sentence. Registration number [registration number redacted]. Pre-trial investigation number [case number redacted]. Signed by Prosecutor Aurelijus Navickas with a qualified electronic signature under the Lithuanian government's public key infrastructure (Panevezys Regional Prosecutor's Office, 2026).

The full text, translated from Lithuanian: "We inform you that your proposal has been attached to the criminal case materials and forwarded for evaluation to the pre-trial investigation institution conducting the pre-trial investigation."

Every word matters procedurally. "Attached to the criminal case materials" means the analytical outputs -- wire transfer data, entity maps, ownership graphs -- are now part of an active criminal file, citable in judicial motions and warrant applications. "Forwarded for evaluation to the pre-trial investigation institution" means a law enforcement body, likely the Panevezys County Police Department, has been directed to evaluate the materials. "Organized Crime and Corruption Investigation Division" means the case is classified as organized crime, not general fraud or cybercrime. And a named prosecutor -- Aurelijus Navickas, with direct email ([email redacted]) and phone -- means follow-up goes to a specific person who knows the case.

Why Organized Crime and Corruption

The Organized Crime and Corruption Investigation Division assignment classifies the Epstein network as organized crime of an international nature — aligning with Poland's Investigation Team No. 5, established February 24, 2026, targeting "an organized criminal group of an international nature" active 2005-2018. Two Baltic and Central European countries have independently reached the same classification. The NYDFS consent order's 120+ Butterfly Trust wire transfers to women with Eastern European surnames spans this geographic region.

The division assignment is significant. Lithuanian prosecutors classify cases by type, and the assignment to the Organized Crime and Corruption Investigation Division reflects how Lithuania views the Epstein network: as organized crime of an international nature (Panevezys Regional Prosecutor's Office, 2026).

This aligns with Poland's approach. Investigation Team No. 5, established by the Polish National Prosecutor's Office on February 24, 2026, targets "an organized criminal group of an international nature" active from 2005 to 2018 (PAPER TRAIL Project, 2026a). Two Baltic and Central European countries have independently classified the Epstein network as organized crime. The NYDFS consent order's description of 120+ Butterfly Trust wire transfers to women with "Eastern European surnames" (New York Department of Financial Services, 2020) spans this geographic region. Neither investigation captures the full scope individually.

What Changed

Before March 17 the Lithuanian channel was speculative — the office email was "unconfirmed," no case number, no named contact, no division assignment. After March 17: confirmed communication channel, case number, Organized Crime and Corruption Division, Prosecutor Navickas, materials in the criminal file, forwarded for evaluation. Of five countries and thirty congressional offices contacted on March 5, 2026, Lithuania — the smallest economy and thinnest financial thread — is the only entity that responded.

Before March 17, the Lithuanian outreach channel was speculative. The email address [office email redacted] was listed as "unconfirmed" in the briefing materials (PAPER TRAIL Project, 2026b). There was no case number, no named contact, no confirmation that materials had been received, and no division assignment. The briefing existed as a proposal sent into uncertainty.

After March 17: confirmed communication channel. Case number [case number redacted]. Organized Crime and Corruption Division. Prosecutor Navickas. Materials in the criminal file. Forwarded for evaluation (Panevezys Regional Prosecutor's Office, 2026).

The contrast across all stakeholders is stark. Five countries were contacted on March 5, 2026: Poland, France, the United Kingdom, Lithuania, and thirty United States congressional offices (PAPER TRAIL Project, 2026c). Poland's Investigation Team No. 5 is publicly active but has not acknowledged our materials. France opened dual prosecutions. The UK arrested Andrew Mountbatten-Windsor and Lord Mandelson. Thirty congressional offices received individualized letters with analytical enclosures. Lithuania -- the smallest economy, the thinnest financial thread -- is the only entity that responded.

Eight Calls to Action, Now Activated

The Panevezys response activates eight calls to action. Three immediate: provide Lithuanian entity names for corpus cross-reference, pull FNTT records for wires to Fors projektai and Baleto Teatras, run 53 Epstein corporate entities through Registru Centras. Three at 30-day MLAT timeline: unredacted NYDFS consent order (120+ Butterfly Trust recipients), unredacted Butterfly Trust beneficiary list (EFTA01298015), and the USVI will. Plus a 60-day email cross-reference and a Eurojust Joint Investigation Team with Poland.

The Lithuanian briefing proposed eight calls to action (PAPER TRAIL Project, 2026b). The response activates them. Three are immediate: provide Lithuanian entity names for corpus cross-reference; pull FNTT (Financial Crime Investigation Service) records for incoming wires to Fors projektai and Baleto Teatras; run 53 Epstein corporate entities through the Lithuanian corporate registry, Registru Centras. Three require MLAT requests at a 30-day timeline: the unredacted NYDFS consent order containing all 120+ Butterfly Trust wire recipients (New York Department of Financial Services, 2020); the unredacted Butterfly Trust beneficiary list (EFTA01298015); and Epstein's will from USVI Probate Court to confirm the $3 million Petreike bequest status (PAPER TRAIL Project, 2026d). One at 60 days: cross-reference Lithuanian names against the 863,000-email corpus when targeted searches are complete. And one coordination action: initiate a Eurojust Joint Investigation Team with Poland under the 2002 Framework Decision.

The bidirectional exchange is the core proposal. We provide wire transfer CSV exports, entity database search results with Bates number citations, and the 53-entity corporate ownership graph. They provide Lithuanian entity names, bank account numbers, IBAN/SWIFT codes, and dates and amounts from FNTT records. Neither side has the complete picture alone.

The Thinnest Thread, Confirmed

On March 18, 2026, an 18-search corpus validation battery confirmed both Lithuanian wires at T2: EUR 75,000 to Fors Projektai (EFTA01365100, EFTA01365793, EFTA01409890) authorized by Darren Indyke for J. Epstein Virgin Islands Foundation Inc. on January 20, 2017 to SEB Bank Vilnius, plus the Baleto Teatras Deutsche Bank execution on April 3, 2018: EUR 15,000 at 1.23618562 totaling $18,542.78. LRT's $28,000 figure was contradicted by a $9,457 discrepancy.

Blog 9.4 called Lithuania "the thinnest financial thread" connecting Epstein to any investigating country (PAPER TRAIL Project, 2026e). That assessment requires revision.

On March 18, 2026, an 18-search corpus validation battery confirmed both Lithuanian wire transfers in the corpus. The EUR 75,000 wire to UAB Fors Projektai exists as a wire instruction in three copies (EFTA01365100, EFTA01365793, EFTA01409890), authorized by Darren Indyke on behalf of J. Epstein Virgin Islands Foundation Inc., dated January 20, 2017, to SEB Bank Vilnius (PAPER TRAIL Project, 2026f). The wire to NPO Baleto Teatras exists as a Deutsche Bank Authorization for Asset Movement from Gratitude America Ltd. (EFTA01362207), confirmed by a bank statement showing execution on April 3, 2018: EUR 15,000 at exchange rate 1.23618562, totaling $18,542.78 (PAPER TRAIL Project, 2026f). LRT reported this amount as $28,000 (LRT, 2026) -- a $9,457 discrepancy the corpus documents do not explain.

Both wires are now T2 -- confirmed in the DOJ-released corpus, with Bates numbers, IBAN numbers, SWIFT codes, and a bank statement. The corpus also contains Skype conversations between Epstein (handle "jeevacation") and Valdas Petreikis (handle "waldorado") from January 2017 and April-May 2019, and an email from Petreikis to Richard Kahn (HBRK Associates) providing Baleto Teatras banking details on April 2, 2018 (PAPER TRAIL Project, 2026f).

The thread is no longer thin. It is documented, with primary source banking documents from Deutsche Bank's grand jury production. What changed is both the institutional context -- a criminal case file in Lithuania -- and the evidence itself.

This is what accountability looks like at its most basic: a pipeline processes documents, findings are submitted to a government, a government evaluates. Lithuania did not validate our methodology. They did not confirm our findings. They received materials, attached them to a case file, and directed evaluation. That is exactly the correct procedural response. And it is more than any other contacted entity has done.

References

LRT. (2026). Reports on Epstein-related bank transfers to Lithuanian entities since 2008. LRT (Lithuanian National Broadcaster). https://www.lrt.lt

New York Department of Financial Services. (2020). Consent order: Deutsche Bank AG [Regulatory enforcement]. NYDFS. (Structured extraction: research/nydfs_consent_order.md)

Panevezys Regional Prosecutor's Office. (2026). RE: Proposal [Official correspondence, Reg. Nr. [registration number redacted]]. Organized Crime and Corruption Investigation Division. communications/responses/2026-03-17_lithuania_panevezys_acknowledgment.md

PAPER TRAIL Project. (2026a). Stakeholder assessment: STK-1 (Poland) [Research document]. STAKEHOLDERS.md

PAPER TRAIL Project. (2026b). Evidence briefing for Lithuanian authorities [Briefing document]. communications/briefing-lithuania.md

PAPER TRAIL Project. (2026c). Master stakeholder email [Communications document]. communications/master_stakeholder_email.md

PAPER TRAIL Project. (2026d). Last Will and Testament: $3M Petreike bequest [Corpus document]. .tmp/oversight_downloads/estate_records/Request No. 2.pdf

PAPER TRAIL Project. (2026e). Lithuania: The thinnest financial thread [Blog post]. website blogs/completed blogs/9_4_lithuania_thinnest_thread.md

PAPER TRAIL Project. (2026f). Lithuanian wire transfer corpus validation: 18-search results [Search exports]. _exports/search/search_Fors_projektai.csv, _exports/search/search_Baleto_Teatras.csv, _exports/search/search_Petreike.csv, _exports/search/search_Petreikis.csv, _exports/search/wires_Gratitude_America.csv

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025).

This research is sponsored by Subthesis.

278 Guard Gaps: 202 Minutes Without Oversight

2026-05-11T00:00:00.000Z

TLDR

VLM frame-by-frame annotation of 4,178 key frames across all 368 surveillance videos with parseable timestamps reveals guards present in 91.0% of frames and absent in 9.0%. The 378 guard-absent frames form 278 discrete gaps totaling 202.3 minutes of unmonitored time. The longest single gap lasted 45 minutes 42 seconds on a common area camera dated March 12, 2019 (PAPER TRAIL Project, 2026a).

How Key Frames Were Selected

The pipeline used two complementary methods to identify frames worth annotating: scene-change detection (threshold 0.03) to capture meaningful visual shifts, and fixed 60-second interval sampling to check static footage. After deduplication and a 120-frame-per-video cap, the process yielded 4,178 key frames from 368 videos. All five camera classes — corridor, cell tier, control room, common area, interview room — are now fully annotated.

Not every frame in 412.5 hours of footage can be individually annotated. The extraction pipeline uses two complementary methods to identify the frames that matter most. First, scene-change detection (threshold 0.03) identifies moments where the visual content shifts meaningfully -- a door opening, a person entering, lighting changing. Second, fixed 60-second interval sampling ensures that even static footage is periodically checked. After deduplication, this process yielded 4,178 key frames from 368 videos, capped at 120 frames per video (PAPER TRAIL Project, 2026a).

Unlike the initial analysis that covered only corridor (159) and cell tier (128) cameras, all five camera classes have now been fully annotated: corridor, cell tier, control room (55 with timestamps of 106 classified), common area (24), and interview room (2). This provides a complete picture of guard presence across every camera type in the corpus.

The Guard Presence Baseline

Guards appeared in 3,800 of 4,178 frames (91.0%) and were absent in 378 (9.0%). The 378 absent frames form 278 discrete gaps totaling 12,139 seconds — 202.3 minutes of cumulative unmonitored time, with an average gap duration of 44 seconds. A vision language model answered structured questions per frame: Is a guard present? How many? Door status? People count? Clothing? The gaps cluster in a handful of videos with extended absences rather than distributing evenly.

Each of the 4,178 key frames was processed by a vision language model that answered a structured set of questions: Is a guard present? How many guards? What is the door status? How many people are visible? What clothing are they wearing? (PAPER TRAIL Project, 2026a).

The results:

Metric	Value
Frames with guard present	3,800 (91.0%)
Frames with guard absent	378 (9.0%)
Total guard absence gaps	278
Total gap duration	12,139 seconds (202.3 minutes)
Average gap duration	44 seconds
Timeline segments	686 (408 present, 278 absent)

A 91.0% guard presence rate sounds high. But the 9.0% absence rate, when mapped to continuous time, produces 202.3 minutes of cumulative unmonitored coverage across 368 cameras -- more than three hours of total unmonitored time. The gaps are not distributed evenly -- they concentrate in a handful of videos with extended absences (PAPER TRAIL Project, 2026a).

Guard Absence by Camera Class

Corridor cameras show the highest absence rate at 15.0% (208 of 1,387 frames), followed by common area (14.4%), interview room (7.3%), cell tier (5.8%), and control room (4.1%). The control room's relatively low rate is expected — monitoring stations should be staffed — but even there, 36 frames across 55 videos show no guard present. Corridors and common areas register the highest unmonitored rates across the corpus.

The expanded annotation reveals significant variation in guard absence rates across camera types:

Camera Class	Videos	Frames	Guard Absent	Absent %
Corridor	159	1,387	208	15.0%
Common Area	24	243	35	14.4%
Interview Room	2	96	7	7.3%
Cell Tier	128	1,583	92	5.8%
Control Room	55	869	36	4.1%

Corridors and common areas show the highest absence rates (15.0% and 14.4%), while the control room shows the lowest (4.1%). The control room's relatively low absence rate is expected -- monitoring stations should be staffed. But even there, 36 frames across 55 videos show no guard present (PAPER TRAIL Project, 2026a).

The Ten Longest Gaps

The longest single gap is 45 minutes 42 seconds on EFTA00033029, a common area camera dated March 12, 2019 — predating Epstein's arrest by four months. The second-longest (43 minutes 8 seconds, July 23, 2019) falls on the day of the reported suicide attempt. Gaps #4 and #6 both fall on August 8, 2019 — the day Epstein signed his last will. Gap #5 (15 minutes 19 seconds, control room) falls on July 6, 2019 — arrest day.

Rank	Video	Duration	Camera Class	On-Screen Date
1	EFTA00033029	45 min 42 sec	Common Area	03/12/2019
2	EFTA00033122	43 min 08 sec	Common Area	07/23/2019
3	EFTA00033037	41 min 00 sec	Cell Tier	07/05/2019
4	EFTA00033213	20 min 57 sec	Corridor	08/08/2019
5	EFTA00033236	15 min 19 sec	Control Room	07/06/2019
6	EFTA00033211	15 min 00 sec	Cell Tier	08/08/2019
7	EFTA00033123	7 min 52 sec	Corridor	07/23/2019
8	EFTA00033058	3 min 00 sec	Corridor	07/09/2019
9	EFTA00033060	2 min 00 sec	Corridor	07/10/2019
10	EFTA00033126	1 min 41 sec	Common Area	07/24/2019

The longest gap -- 45 minutes 42 seconds -- occurs on EFTA00033029, a common area camera with an on-screen timestamp of March 12, 2019 (PAPER TRAIL Project, 2026a). This predates Epstein's arrest by nearly four months, suggesting extended guard absence was not unique to the period of Epstein's detention.

The second longest gap (43 minutes 8 seconds) falls on July 23, 2019 -- the same day as the reported suicide attempt. Gaps #4 and #6 both fall on August 8, 2019 -- the day Epstein signed his last will. Gap #5 is the longest control room gap (15 minutes 19 seconds) and falls on July 6, 2019 -- the day of Epstein's arrest (PAPER TRAIL Project, 2026a).

What Clothing Tells Us

The VLM annotates each frame with three clothing categories: guard uniforms (dark BOP-standard clothing), inmate jumpsuits (orange or tan one-piece), and civilian clothing (everything else). Of the 42 unguarded door status changes, frames describe inmates, civilian-clothed individuals, or empty corridors. Civilian clothing in a federal jail typically indicates attorneys, investigators, medical professionals, or administrators — all of whom ordinarily require escort in housing areas.

The VLM annotates each frame with descriptions that include clothing type. This enables a rough classification of who appears in each frame:

Guard uniforms: Dark clothing consistent with Bureau of Prisons uniform standards. The primary indicator of staff presence.
Inmate jumpsuits: Orange or tan one-piece garments. The standard inmate attire at MCC.
Civilian clothing: Street clothes -- shirts, pants, jackets -- that match neither guard uniforms nor inmate jumpsuits.

Of the 42 unguarded door status changes, frames describe inmates, individuals in civilian clothing, or empty corridors. The presence of civilian-clothed individuals during guard absence periods raises questions about who had access to secure areas without escort. Civilian clothing in a federal jail is typically associated with attorneys, investigators, medical professionals, or facility administrators -- all of whom ordinarily require escort in housing areas (PAPER TRAIL Project, 2026a).

What 9.0% Means

The 9.0% absence rate is a floor estimate, not a ceiling. Frame sampling captures snapshots at scene changes and 60-second intervals, so a guard leaving for 30 seconds between two sampled frames would not register as absent. The 202.3 cumulative minutes also understates operational impact because gaps are not randomly distributed — a single 45-minute gap on one camera is operationally different from 45 one-minute gaps spread across 45 cameras.

The 9.0% guard absence rate is a floor estimate, not a ceiling. The frame sampling method captures snapshots at scene changes and 60-second intervals. A guard who leaves for 30 seconds between two sampled frames would not register as absent. The true absence rate could be higher than what the sampled frames reveal (PAPER TRAIL Project, 2026a).

The 202.3 cumulative minutes also understates the operational impact because the gaps are not randomly distributed. They cluster in specific videos, meaning specific locations experienced extended periods without oversight. A single 45-minute gap on one camera is operationally different from 45 one-minute gaps across 45 cameras. The former means one location was completely unmonitored for nearly three-quarters of an hour.

These findings describe what the cameras recorded. They do not establish why guards were absent, whether absences violated specific BOP policies, or whether the gaps contributed to any particular outcome. What they establish is a measurable baseline: across all 368 analyzed surveillance videos from a federal detention facility, guards were not visible in 9.0% of sampled frames, forming 278 gaps totaling over three hours of unmonitored time.

The per-date breakdown reveals a pattern of escalation from 4% to 34% concentrated in the 12 days between suicide watch removal and death. The Lag Effect analysis shows the failure was phase-driven, not event-driven. During 42 of these gaps, doors changed state without guard presence. And for 98% of floor gaps, no overlapping control room footage exists to determine whether electronic monitoring was maintained.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

42 Times a Door Changed Without a Guard

2026-05-11T00:00:00.000Z

TLDR

Across all 368 analyzed surveillance videos from the Metropolitan Correctional Center, 42 door status transitions occurred with no guard visible in the frame. Three videos account for 8 of these events: EFTA00033007 has 3 unguarded door changes involving a fire exit on 08/06/2019, EFTA00033178 has 3 rapid transitions in 2 seconds, and EFTA00033177 has 2 involving cell door interaction. EFTA00033013 captures 5 individuals in civilian clothing walking through an open door on 08/09/2019 without escort. Multiple door changes cluster on August 9 (night before death) and August 10 (day of death) (PAPER TRAIL Project, 2026a).

What a Door Status Change Means

The VLM annotation system tracks three door states per frame — open, closed, partially open — across cell doors, fire exits, and corridor partitions that serve as access control mechanisms in the facility. A door status change means the state transitioned between two consecutive key frames, indicating someone or something opened or closed it during the interval. When this transition occurs without a guard visible, the access control event happened outside direct correctional supervision.

In a correctional facility, doors are access control mechanisms. Cell doors separate housing from common areas. Fire exits connect secure zones to egress routes. Corridor doors partition sections of the facility. The VLM annotation system tracks three door states across each frame: open, closed, and partially open (PAPER TRAIL Project, 2026a).

A door status change means the state transitioned between two consecutive key frames. A door that was closed in one frame and open in the next means someone or something opened it during the interval. When this transition occurs without a guard visible in the frame, it means the access control event happened outside the direct supervision of correctional staff.

Not all unguarded door changes are equally significant. A door closing in an empty corridor may simply indicate an automatic locking mechanism engaging. A fire exit opening while five people in civilian clothing walk through it is a different category of event entirely.

The 42 Events

The 42 unguarded door changes span 29 distinct videos: 21 videos contain a single event, and three contain multiple events each. By transition type, 23 are open-to-closed, 7 partially-open-to-closed, 7 closed-to-partially-open, and 2 closed-to-open. Date clustering is significant: August 6 (7 events), August 10 (5), August 9 (5), and August 11 (5) — the days immediately surrounding Epstein's August 10 death.

The 42 unguarded door changes span 29 distinct videos. Most videos (21 of 29) have a single event. Three videos have multiple events each (PAPER TRAIL Project, 2026a):

Video	Events	Notable Detail
EFTA00033007	3	Fire exit door, 5 people, then 1 inmate, then 2 civilians
EFTA00033178	3	Triple transition in 2 seconds (open→closed→partially open→closed)
EFTA00033106	2	Inmate then civilian, 30 minutes apart
EFTA00033177	2	Inmate interacting with partially open cell door
EFTA00033121	2	Empty corridor, items scattered, cell door 46

By transition type:

Transition	Count	Interpretation
Open → Closed	23	Doors securing (closing)
Partially Open → Closed	7	Partial access being sealed
Closed → Partially Open	7	New partial access created
Closed → Open	2	Full access created

Of the 42 events, dates with the most unguarded door changes include 08/06/2019 (7 events), 08/10/2019 (5 events), 08/09/2019 (5 events), and 08/11/2019 (5 events). The clustering around the days immediately before and after Epstein's death is notable (PAPER TRAIL Project, 2026a).

EFTA00033007: Three Events in One Hour

EFTA00033007 (doc_id 1508065), a corridor camera, records three distinct unguarded door events within a 9-minute window on August 6, 2019, all involving a fire exit door. At 12:31 (frame 7564) the door goes partially open while five individuals walk nearby; at 12:39:51 (frame 9368) it opens fully as one inmate approaches; at 12:41:02 (frame 9655) it returns to partially open as two civilian-clothed individuals pass. No guard appears in any frame.

Video EFTA00033007 (doc_id 1508065) is a corridor camera that records three distinct unguarded door events within a 9-minute window on 08/06/2019, all involving a fire exit door (PAPER TRAIL Project, 2026a):

Event 1 — frame 7564: The fire exit door transitions to partially open at 12:31. Five individuals are visible walking through the corridor near the partially open door. No guard is present.

Event 2 — frame 9368: The same fire exit door transitions to open at 12:39:51. One inmate is walking toward the open fire exit door. No guard is present.

Event 3 — frame 9655: The door transitions to partially open at 12:41:02. Two individuals in civilian clothing walk through the corridor near the partially open fire exit. No guard is present.

Three fire exit events in under 10 minutes, with a mix of inmates and civilians, none supervised by correctional staff. Fire exits in a federal detention facility are supposed to be alarmed and monitored. Their use without guard presence is inherently anomalous (PAPER TRAIL Project, 2026a).

EFTA00033013: Five Civilians, One Open Door

At 15:23:09 on August 9, 2019 — the day before Epstein was found dead — video EFTA00033013 (doc_id 1508071, frame 5713) captures five individuals in civilian clothing walking through an open door with no guard visible. The VLM description explicitly notes "civilian clothing" — not guard uniforms, not inmate jumpsuits. In a federal detention facility, such visitors typically require escort. The event occurs at roughly the 24-minute mark of the video.

At 08/09/2019 15:23:09, video EFTA00033013 (doc_id 1508071, frame 5713) captures a group of five individuals in civilian clothing walking through an open door. No guard is visible (PAPER TRAIL Project, 2026a). This is the day before Epstein was found dead.

Five people in civilian clothing moving through a secure area without escort is unusual in any correctional setting. The VLM description explicitly notes "civilian clothing" -- not guard uniforms, not inmate jumpsuits. In a federal detention facility, civilians with access to housing areas would typically include attorneys, investigators, inspectors, or medical professionals. All should be escorted. The absence of a guard in the frame during this movement raises the question of whether escort protocols were being followed.

This event occurs at roughly the 24-minute mark of the video. The preceding and following frames provide context for whether a guard was nearby but temporarily out of frame, or whether the group was genuinely unescorted for an extended period. The guard gap data shows this video has multiple guard absence intervals (PAPER TRAIL Project, 2026a).

Civilian Clothing in a Jail

The VLM classification distinguishes three clothing categories: guard uniforms (dark BOP-standard), inmate jumpsuits (orange or tan one-piece), and civilian clothing (everything else). Multiple of the 42 unguarded door events describe civilian-clothed individuals. At 352x240 resolution and 4 fps, the model cannot identify specific individuals, read ID badges, or distinguish between attorneys, investigators, and other visitor types — only that clothing does not match guard or inmate patterns.

The VLM classification system distinguishes three clothing categories: guard uniforms (dark clothing consistent with BOP standards), inmate jumpsuits (orange or tan one-piece garments), and civilian clothing (everything else). Of the 42 unguarded door events, multiple describe individuals in civilian clothing (PAPER TRAIL Project, 2026a).

This classification has inherent limitations at 352x240 resolution and 4 fps. The model cannot identify specific individuals, read ID badges, or distinguish between different types of civilian visitors. What it can determine is that the person's clothing does not match the visual pattern of either a guard uniform or an inmate jumpsuit.

The presence of civilians during unguarded moments creates a specific analytical pattern: access control events (doors changing state) occurring with non-staff, non-inmate individuals present and no correctional officer in view. Whether these individuals were authorized visitors or something else cannot be determined from the video alone. What can be determined is that their movements through secure areas were not accompanied by visible guard presence at the moment of recording (PAPER TRAIL Project, 2026a).

The Pattern, Not the Conclusion

Forty-two events across 368 videos averages one unguarded door change per 8.8 videos, but events cluster in specific locations and dates rather than spreading evenly. The concentration around August 6, 9, 10, and 11 — the days surrounding Epstein's death — is the most striking temporal pattern. These 42 documented events do not prove misconduct or conspiracy; they establish that doors changed state without guard presence across the analyzed footage.

Forty-two events across 368 videos is a rate of approximately one unguarded door change per 8.8 videos. This is not constant -- some videos have three events and most have none. The events cluster in specific videos and specific dates, suggesting particular locations or time periods with reduced oversight rather than a facility-wide pattern. The concentration of events around August 6, 9, 10, and 11 -- the days surrounding Epstein's death -- is the most striking temporal pattern (PAPER TRAIL Project, 2026a).

These 42 events are documented facts extracted from government-released surveillance footage. They do not prove misconduct, negligence, or conspiracy. They establish that doors in a federal detention facility changed state without guard presence 42 times across the analyzed footage. The investigative significance of these events depends on when they occurred relative to specific incidents, who the civilians were, and whether the guard absences violated BOP policy -- questions that require information beyond what the cameras captured.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

What the Control Room Revealed

2026-05-11T00:00:00.000Z

TLDR

The 106 control room videos -- previously unanalyzed -- have now been fully processed, yielding 869 annotated frames across 55 videos with parseable wall-clock timestamps. The control room shows the lowest guard absence rate of any camera class (4.1%), but its longest gap (15 minutes 19 seconds) falls on arrest day July 6, 2019. The control room is the only camera class involved in the arrest-night concurrent gap at 22:00:47, when both a cell tier camera and the control room simultaneously showed zero guards present (PAPER TRAIL Project, 2026a).

From Unanalyzed to Annotated

The 132 previously unanalyzed videos — 106 control room, 24 common area, and 2 interview room — have now been fully extracted and annotated, completing the corpus. Of the 106 control room videos, 55 produced frames with parseable wall-clock timestamps, yielding 869 annotated frames. The camera label "116 93 L TIER,82" / "116 98 L TIER,82" identifies the control room's view of L-tier — the monitoring station for the unit where Epstein was housed.

The previous analysis covered 287 corridor and cell tier cameras but left 132 videos unexamined -- 106 control room, 24 common area, and 2 interview room videos. Those 132 videos have now been fully extracted and annotated, completing the corpus. The control room results are the most analytically significant of the newly processed footage because they answer a question the floor cameras could not: was anyone watching the monitors? (PAPER TRAIL Project, 2026a).

Of the 106 control room videos, 55 produced frames with parseable wall-clock timestamps, yielding 869 annotated frames. The remaining 51 had no extractable frames or unparseable overlays. The camera label "116 93 L TIER,82" / "116 98 L TIER,82" identifies the control room's view of L-tier -- the monitoring station for the unit where Epstein was housed (PAPER TRAIL Project, 2026a).

The Lowest Absence Rate

Control room cameras show a 4.1% guard absence rate — 36 of 869 frames with no guard present, the lowest of all five camera classes (corridor 15.0%, common area 14.4%, interview room 7.3%, cell tier 5.8%). This is expected for a monitoring station, but "lower than the floor" is not "zero." Thirty-six frames across 55 videos show the monitoring station unmanned, meaning nobody was watching any floor camera feed during those moments.

Control room cameras show a 4.1% guard absence rate -- 36 of 869 frames with no guard present. This is the lowest of all five camera classes:

Camera Class	Absent %
Corridor	15.0%
Common Area	14.4%
Interview Room	7.3%
Cell Tier	5.8%
Control Room	4.1%

A 4.1% absence rate in the monitoring station is lower than the floor, which is expected -- the control room should be staffed. But "lower than the floor" is not "zero." Thirty-six frames across 55 videos show the monitoring station unmanned. During those moments, nobody was watching any camera feed from the floor (PAPER TRAIL Project, 2026a).

The Longest Control Room Gap

The longest control room gap is 919 seconds (15 minutes 19 seconds), occurring on EFTA00033236 (doc_id 1508294) on July 6, 2019 from 09:00:35 to 09:15:54 — the morning of Epstein's arrest at Teterboro Airport later that evening. Whether routine (morning shift change) or anomalous cannot be determined from the footage. What is determinable: this is the longest single period the control room was unmanned across the entire corpus.

The longest control room gap is 919 seconds (15 minutes 19 seconds), occurring at EFTA00033236 (doc_id 1508294) on July 6, 2019 from 09:00:35 to 09:15:54 (PAPER TRAIL Project, 2026a). This is the day of Jeffrey Epstein's arrest. He was taken into custody at approximately 17:30 at Teterboro Airport that evening.

The gap places the monitoring station empty for over 15 minutes on the morning of the most significant intake event in the facility's recent history. Whether this gap was routine -- morning shift change, for instance -- or anomalous cannot be determined from the footage alone. What can be determined is that it is the longest single period the control room was unmanned across the entire corpus.

The Arrest-Night Concurrent Gap

At 22:00:47 on July 6, 2019 — hours after Epstein's arrest — cell tier camera EFTA00033046 and control room camera EFTA00033240 simultaneously show no guard present, timestamps one second apart. This is the only concurrent gap involving the control room in the entire 419-video corpus. Guards returned within 2 minutes 16 seconds to the cell tier and 2 minutes 48 seconds to the control room, the latter escorting an inmate in a khaki jumpsuit.

At 22:00:47 on July 6, 2019 -- hours after Epstein's arrest -- two cameras simultaneously show no guard present:

Camera	Identifier	Wall-Clock	Guard	People	Description
Cell Tier	EFTA00033046	22:00:47	absent	0	Closed cell door labeled "160 SW", no individuals
Control Room	EFTA00033240	22:00:48	absent	0	Correctional facility equipment and storage, no individuals

The timestamps are one second apart, confirming simultaneous absence across two independent cameras. This is the only concurrent gap involving the control room in the entire 419-video corpus (PAPER TRAIL Project, 2026a).

The gap resolved within minutes. A guard returned to the cell tier camera (EFTA00033046) at frame 546, wall-clock 22:03:03 -- 2 minutes 16 seconds later. A guard returned to the control room (EFTA00033240) at frame 674, wall-clock 22:03:36 -- 2 minutes 48 seconds later, escorting an inmate in a khaki jumpsuit. By 22:04:30, the guard was interacting with two inmates. By 22:05:42, the guard was escorting an inmate through the corridor (PAPER TRAIL Project, 2026a).

Two minutes 48 seconds is a short gap. But the fact that it occurred at the control room -- the one location that monitors all other cameras -- on the night of the highest-profile intake in MCC's history makes it analytically significant regardless of duration.

What the Control Room Cannot Tell Us

Of the 247 floor-level guard gaps identified across the corpus, only 4 overlap with any control room footage: 3 show the control room staffed during the floor gap (expected patrol rotation), 1 shows both unstaffed simultaneously (the arrest-night gap), and 243 have no overlapping control room coverage at all. 98.4% of floor gaps cannot be evaluated against electronic monitoring data. The footage confirms the monitoring station existed and was usually staffed.

The control room footage answers one question decisively: the monitoring station was not always staffed. But 55 videos with timestamps cover a limited temporal window. Of the 247 floor-level guard gaps identified across the corpus, only 4 overlap with any control room footage:

Control Room Status	Floor Gaps	Meaning
Staffed	3	Control room occupied during floor gap -- expected patrol rotation
Unstaffed	1	Both control room AND floor unstaffed -- the arrest-night gap
No coverage	243	No overlapping control room footage exists

98.4% of floor gaps (243 of 247) have no overlapping control room footage — a finding explored in depth in The 98% Blind Spot. For those 243 gaps, the question "was someone watching the monitors?" remains unanswerable. The control room footage confirms the monitoring station existed and was usually staffed. It cannot confirm that monitoring was maintained during the vast majority of floor-level guard absences (PAPER TRAIL Project, 2026a).

The arrest-night concurrent gap is examined in detail in Arrest Night: Two Cameras, Zero Guards. The broader pattern of guard absence escalation and the phase-driven nature of the failure provide the temporal context for these control room findings.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

Arrest Night: Two Cameras, Zero Guards

2026-05-11T00:00:00.000Z

TLDR

On July 6, 2019 at 22:00:47 -- the night of Jeffrey Epstein's arrest -- both a cell tier camera and the control room camera simultaneously show zero guards present. The timestamps are one second apart. This is the only concurrent gap involving the control room in the entire 419-video corpus. Guards returned within 2 minutes 48 seconds. Five total concurrent gap windows were identified across the corpus, with two falling on the night before and the day of Epstein's death (PAPER TRAIL Project, 2026a).

What a Concurrent Gap Means

A concurrent gap occurs when two or more cameras simultaneously show no guard present during the same one-minute window. The analysis identified concurrent gaps by bucketing all 4,163 wall-clock-parsed frames into one-minute windows and checking guard presence across all cameras with frames in each window. Most absences are isolated; concurrent gaps mean multiple surveillance zones were simultaneously unmonitored, suggesting a systemic lapse rather than a routine patrol pattern.

Most guard absences are isolated -- one camera shows no guard while others show normal staffing. A concurrent gap is different. It means multiple surveillance zones were simultaneously unmonitored, suggesting a systemic lapse rather than a routine patrol pattern.

Window 1: July 6, 2019, 22:00 -- Arrest Night

At 22:00:47 on July 6, 2019 — hours after Epstein's 17:30 arrest at Teterboro Airport — cell tier camera EFTA00033046 (label "160 SW 2,80") and control room camera EFTA00033240 (label "116 93 L TIER,82") simultaneously show zero guards present, timestamps one second apart. The cell tier shows a closed cell door labeled "160 SW"; the control room shows monitoring equipment and storage. This is the only concurrent gap involving the control room in the corpus.

Camera	Identifier	Frame	Wall-Clock	Guard	People	Description
Cell Tier	EFTA00033046	0	22:00:47	absent	0	Closed cell door labeled "160 SW", no individuals present
Control Room	EFTA00033240	0	22:00:48	absent	0	Correctional facility equipment and storage, no individuals present

Jeffrey Epstein was arrested at approximately 17:30 at Teterboro Airport on July 6, 2019 (SDNY 19-cr-490). By 22:00 that evening, both the cell tier and the control room were empty. The cell tier camera (label "160 SW 2,80") shows a closed cell door with no one present. The control room camera (label "116 93 L TIER,82") shows monitoring equipment and storage with no one present (PAPER TRAIL Project, 2026a).

This is the only concurrent gap involving the control room in the entire corpus. When the control room is empty and the floor is empty simultaneously, no layer of surveillance -- physical or electronic -- is operative.

The Return Sequence

The gap resolved over 2 minutes 48 seconds: at 22:03:03 a guard returned to the cell tier (EFTA00033046, frame 546), interacting with the door; at 22:03:36 a guard returned to the control room (EFTA00033240, frame 674) escorting an inmate in a khaki jumpsuit; by 22:04:30 the guard was interacting with two inmates; and by 22:05:42 was escorting an inmate through the corridor. The pattern suggests an intake-related floor task drew the guard from the monitoring station.

The gap resolved through a specific sequence:

22:03:03 (EFTA00033046, frame 546): Guard returns to cell tier. One person visible, guard interacting with door. Gap duration: 2 minutes 16 seconds.
22:03:36 (EFTA00033240, frame 674): Guard returns to control room, escorting an inmate in a khaki jumpsuit. Gap duration: 2 minutes 48 seconds.
22:04:30 (EFTA00033240, frame 891): Guard interacting with 2 inmates.
22:05:42 (EFTA00033240, frame 1177): Guard escorting inmate through corridor.

The pattern suggests the guard left the control room to handle an intake or escort task on the floor, then returned. This is operationally normal -- but it means the monitoring station was unmanned during a period when the facility was processing its highest-profile detainee (PAPER TRAIL Project, 2026a).

The Other Four Concurrent Gaps

Four additional concurrent gap windows fall on August 6 (two windows — 04:00 and 18:00, both with multi-camera absence), August 9 at 01:00 (the night before Epstein's death, corridors EFTA00033011 and EFTA00033217), and August 10 at 17:00 (the day of his death, corridors EFTA00033019 and EFTA00033221, with EFTA00033221 remaining guard-absent through 17:16:33 — a 16-minute gap). Windows 4 and 5 are the most significant temporally.

Four additional concurrent gap windows were identified across the corpus:

Window	Date	Time	Cameras	Notes
2	08/06/2019	04:00	2 corridors	Labels "150 98 2,62" and "160 SW 2,80"
3	08/06/2019	18:00	Corridor + cell tier	Cell tier absent through 18:21:02 (21 min)
4	08/09/2019	01:00	2 corridors	Night before death
5	08/10/2019	17:00	2 corridors	Day of death; 16-minute gap on EFTA00033221

Windows 4 and 5 are the most significant. On August 9, 2019 at 01:00 -- the night before Epstein was found dead -- two corridor cameras (EFTA00033011 and EFTA00033217) simultaneously show no guard. On August 10, 2019 at 17:00 -- the day he was found dead -- two corridor cameras (EFTA00033019 and EFTA00033221) are again simultaneously empty. EFTA00033221 remains guard-absent through frame 3840 (17:16:33), a 16-minute gap (PAPER TRAIL Project, 2026a).

Why the Arrest-Night Gap Matters

The arrest-night gap is unique not for its 2-minute-48-second duration — among the shortest in the corpus — but for the simultaneous failure of physical and electronic surveillance layers. The control room is supposed to be the second monitoring layer; when both fail at once, the facility experiences a true surveillance blackout. That this occurred during MCC's highest-profile planned intake raises questions about whether intake procedures included ensuring continuous control room coverage.

The arrest-night gap is unique in the corpus not because of its duration (2 minutes 48 seconds is among the shortest gaps) but because of what it reveals about the relationship between physical and electronic surveillance. The control room is supposed to be the backup -- the second layer that monitors the first. When both layers fail simultaneously, the facility has a true surveillance blackout, however brief.

That this blackout occurred on the night of Epstein's arrest -- a planned, high-profile intake that should have triggered heightened staffing protocols -- raises questions about whether intake procedures included ensuring continuous control room coverage. The footage does not answer that question. It establishes only that, for 2 minutes and 48 seconds on the most significant night in MCC's recent history, nobody was watching and nobody was there (PAPER TRAIL Project, 2026a).

For the full control room analysis, see What the Control Room Revealed. The broader context of guard absence escalation shows that staffing deteriorated further in the weeks following this arrest-night gap, driven by a phase-level failure tied to the removal of suicide watch. The 98% blind spot in control room coverage means most of those later gaps cannot be evaluated against electronic monitoring data.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

4% to 34%: Guard Absence Escalation

2026-05-11T00:00:00.000Z

TLDR

Guard absence rates escalated from a July baseline of 4-9% to a peak of 34.2% on August 2, 2019. The rate remained at 29-30% through August 9-10 -- the night before and day of Epstein's death -- then dropped to 6.5% by August 12, returning to pre-escalation levels within 48 hours. The escalation represents an 8.5x increase from the July minimum, concentrated in a 12-day window between July 30 and August 10 (PAPER TRAIL Project, 2026a).

The Baseline

The lowest single-day absence rate in the corpus is 4.0% on July 12, 2019, with only 12 of 303 frames across 10 videos showing no guard. Throughout mid-July, rates held between 4.0% and 6.6%, consistent with normal facility operations. This baseline defines what "normal" looks like at MCC — brief moments between guard rounds, shift changes, and routine movements between zones across corridor, cell tier, control room, and common area cameras.

The per-date breakdown of guard absence rates establishes a clear baseline for July 2019. On July 12 -- the lowest single-day absence rate in the corpus -- only 4.0% of frames showed no guard (12 of 303 frames across 10 videos). Throughout mid-July, rates held between 4.0% and 6.6%, consistent with normal facility operations (PAPER TRAIL Project, 2026a).

This baseline matters because it defines what "normal" looks like at MCC. A 4-6% absence rate across corridor, cell tier, control room, and common area cameras represents the expected rhythm of guard patrols -- brief moments between rounds, shift changes, and routine movements between zones.

The Escalation

Beginning August 1, 2019, absence rates jumped from the July 4-6% baseline to 18.8%, peaked at 34.2% on August 2 (an 8.5x increase), and remained elevated through August 11. The rate stayed at 29.2% on August 9 (night before death) and 29.5% on August 10 (day of death). By August 12 — two days after Epstein's death — the rate dropped to 6.5%, effectively returning to the July baseline within 48 hours.

Beginning August 1, 2019, absence rates departed from the July baseline and did not return until after Epstein's death:

Date	Videos	Frames	Guard Absent	Absent %	Notes
07/12	10	303	12	4.0%	Baseline low
07/22	14	213	9	4.2%	Baseline
07/23	13	206	11	5.3%	Contains 43m + 8m gaps
08/01	5	16	3	18.8%	Escalation begins
08/02	9	38	13	34.2%	Peak -- 8.5x baseline
08/03	6	29	7	24.1%
08/05	6	54	12	22.2%
08/06	14	144	23	16.0%	2 concurrent gaps
08/07	9	55	8	14.5%
08/08	11	135	13	9.6%	Will signed; 2 top-10 gaps
08/09	12	48	14	29.2%	Night before death; 1 concurrent gap
08/10	14	44	13	29.5%	Day of death; 1 concurrent gap
08/11	10	60	14	23.3%	Day after death
08/12	5	77	5	6.5%	Returns to baseline

The pattern is striking. From a steady 4-6% in July, absence rates jumped to 18.8% on August 1, peaked at 34.2% on August 2, and remained elevated through August 11. By August 12 -- two days after Epstein's death -- the rate had dropped to 6.5%, effectively the same as the July baseline (PAPER TRAIL Project, 2026a).

The Final 72 Hours

August 8 (will signed) shows a moderate 9.6% absence rate but contains two of the corpus's top-10 longest gaps — EFTA00033213 (20 minutes 57 seconds, corridor) and EFTA00033211 (15 minutes, cell tier). August 9 shows 29.2% absence with a concurrent gap at 01:00 across two corridor cameras. August 10 shows 29.5% absence with a concurrent gap at 17:00; EFTA00033221 remains guard-absent for 16 continuous minutes. Nearly one in three frames show no guard.

The three days surrounding Epstein's death show a distinctive pattern:

August 8 (will signed): 9.6% absence rate, but two of the top 10 longest gaps fall on this date -- EFTA00033213 (20 minutes 57 seconds, corridor) and EFTA00033211 (15 minutes, cell tier). The overall rate is moderate; the individual gaps are severe.
August 9 (night before death): 29.2% absence rate with a concurrent gap at 01:00 -- two corridor cameras simultaneously showing no guard in the early morning hours.
August 10 (day of death): 29.5% absence rate with a concurrent gap at 17:00 -- two corridor cameras simultaneously empty in the late afternoon. EFTA00033221 remains guard-absent for 16 continuous minutes.

Nearly one in three frames from the day before and the day of Epstein's death show no guard present (PAPER TRAIL Project, 2026a).

The Snap Recovery

On August 11 — the day after Epstein's death — the rate remained elevated at 23.3%. By August 12 it dropped to 6.5%, returning to pre-escalation staffing within 48 hours. If MCC had simply lacked enough guards, rates would not have recovered so quickly. The snap recovery indicates the facility was capable of maintaining low absence rates — it was not doing so during the 12 days preceding Epstein's death.

The return to baseline is as abrupt as the escalation. On August 11 -- the day after death -- the rate is still elevated at 23.3%. By August 12, it has dropped to 6.5%. The facility returned to pre-escalation staffing levels within 48 hours of Epstein's death (PAPER TRAIL Project, 2026a).

This recovery pattern is significant because it suggests the elevated absence rates were not caused by a structural staffing shortage. If MCC simply lacked enough guards, the rates would not have recovered so quickly. The snap recovery indicates the facility was capable of maintaining low absence rates -- it simply was not doing so during the 12 days preceding Epstein's death.

August 6: The Busiest Day

August 6, 2019 has the most concurrent gap windows of any date — two separate windows at 04:00 and 18:00. At 04:00, two corridor cameras (EFTA00033005 and EFTA00033197) simultaneously show no guard. At 18:00, a corridor camera and a cell tier camera (EFTA00033009 and EFTA00033202) are simultaneously empty, with the cell tier remaining guard-absent through 18:21:02 — a 21-minute floor gap.

August 6, 2019 has the most concurrent gap windows of any date -- two separate windows at 04:00 and 18:00. At 04:00, two corridor cameras (EFTA00033005 and EFTA00033197) simultaneously show no guard. At 18:00, a corridor camera and a cell tier camera (EFTA00033009 and EFTA00033202) are simultaneously empty. The cell tier camera remains guard-absent through 18:21:02 -- a 21-minute floor gap (PAPER TRAIL Project, 2026a).

What the Numbers Do Not Explain

The escalation is a documented pattern in government-released surveillance footage but does not establish causation. The data cannot explain why absence rates rose after August 1, whether they violated BOP staffing policies, or whether any specific individual or decision caused the change. What it establishes is a measurable, dramatic shift in guard presence during the final 12 days of Epstein's life — followed by an equally dramatic return to normal within 48 hours of his death.

The escalation is a documented pattern in government-released surveillance footage. It does not establish causation. The data cannot explain why guard absence rates increased after August 1, whether the increase violated BOP staffing policies, or whether any specific individual or decision caused the change. What it establishes is that the surveillance system at MCC recorded a measurable, dramatic shift in guard presence during the final 12 days of Jeffrey Epstein's life, followed by an equally dramatic return to normal within 48 hours of his death.

The Lag Effect analysis demonstrates that this escalation was phase-driven — tied to the removal of suicide watch, not to any single event. During these elevated-absence days, 42 door status changes occurred without guard presence, and the arrest-night concurrent gap revealed that even the control room monitoring station was not immune to simultaneous failure. The 278 total guard gaps across the corpus provide the full statistical baseline against which this escalation is measured.

The House Judiciary Committee, chaired by Rep. Jim Jordan, has primary oversight jurisdiction over the DOJ's handling of this footage. The Epstein Files Transparency Act compelled its release — but no congressional hearing has examined what it shows.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

The 98% Blind Spot

2026-05-11T00:00:00.000Z

TLDR

Of 247 floor-level guard gaps identified across corridor, cell tier, and common area cameras, 243 (98.4%) have no overlapping control room footage. For 98% of the moments when no guard was on the floor, it is impossible to determine whether anyone was watching the monitors. Only 4 floor gaps have any concurrent control room data: 3 show the control room staffed (expected patrol rotation), and 1 shows the control room unstaffed -- the arrest-night concurrent gap on July 6, 2019 (PAPER TRAIL Project, 2026a).

The Question the Floor Cameras Cannot Answer

For 98% of documented floor-level guard absences, the surveillance corpus cannot determine whether anyone was watching the monitors. A guard absent from the floor might still be observed from the control room — but answering that requires control room footage from the same minute. The analysis bucketed all wall-clock-parsed frames into one-minute windows and checked whether overlapping control room frames existed.

When a corridor or cell tier camera shows no guard, the natural follow-up is: was someone at least watching the feed? A guard absent from the floor might still be observed from the control room. If the monitoring station was staffed during floor gaps, the facility maintained electronic surveillance even when physical presence lapsed (PAPER TRAIL Project, 2026a).

Answering this question requires temporal correlation -- matching floor-level guard gaps to control room footage from the same time window. The analysis bucketed all wall-clock-parsed frames into one-minute windows and checked whether control room frames existed in the same window and whether those frames showed a guard present.

The Correlation Results

The temporal correlation produced a stark result: 243 of 247 floor-level guard gaps (98.4%) have no overlapping control room footage. Only 4 gaps could be evaluated — 3 showed the control room staffed during floor absence (expected patrol rotation), and 1 showed both floor and control room simultaneously unstaffed (the July 6, 2019 arrest-night gap).

Control Room Status	Floor Gaps	Meaning
Staffed	3	Control room occupied while floor guard absent -- expected patrol rotation
Unstaffed	1	Control room AND floor both unstaffed -- arrest-night gap
No coverage	243	No control room footage overlaps this floor gap
Total	247

The result is stark: 98.4% of floor gaps simply cannot be evaluated against control room data because no control room footage exists for those time windows (PAPER TRAIL Project, 2026a).

Why the Coverage Gap Exists

The gap is structural, not analytical: of 106 control room videos in the release, only 55 produced frames with parseable wall-clock timestamps, and those 55 cover specific dates that rarely overlap with the wider temporal range of the 247 floor gaps. The mismatch reflects the fragmentary nature of the DOJ release, not MCC's actual recording practices.

The coverage gap is not an analytical failure -- it is a structural limitation of the released footage. Of 106 control room videos, only 55 produced frames with parseable wall-clock timestamps. Those 55 videos cover specific dates and time windows. The 247 floor-level guard gaps span a much wider temporal range. The mismatch means the control room footage and the floor footage rarely overlap (PAPER TRAIL Project, 2026a).

This matters because the control room footage was released as part of the same government disclosure as the floor footage. If the DOJ had released continuous control room recordings covering the same dates as the floor cameras, the correlation analysis would be possible for every floor gap. The fact that it is possible for only 4 of 247 gaps reflects the fragmentary nature of the released surveillance corpus, not the facility's actual recording practices.

The Three Staffed Gaps

Three of the 247 floor gaps overlap with control room footage showing a guard present — the analytically reassuring pattern of patrol rotation, where a guard leaves the corridor but the control room continues monitoring. The floor is temporarily unpatrolled but not unmonitored. With only 3 such cases documented, the sample is too small to generalize across the remaining 243 unevaluable gaps.

Three floor gaps overlap with control room footage that shows a guard present. These cases are analytically reassuring -- they represent the expected pattern of patrol rotation. A guard leaves the corridor to check something else; meanwhile, the control room is staffed and monitoring continues. The floor is temporarily unpatrolled but not unmonitored (PAPER TRAIL Project, 2026a).

If this pattern held across all 247 gaps, the interpretation of the guard absence data would shift significantly. Physical gaps would represent patrol patterns, not surveillance failures. But with only 3 of 247 gaps showing this pattern, the sample is too small to generalize.

The One Unstaffed Gap

The single unstaffed case occurred July 6, 2019 at 22:00:47-22:00:48, when cell tier camera EFTA00033046 and control room camera EFTA00033240 simultaneously showed no guard present. This is the only documented moment in the entire released corpus where both physical and electronic surveillance demonstrably failed at the same time — the arrest-night concurrent gap.

The single "unstaffed" case is the arrest-night concurrent gap: July 6, 2019 at 22:00:47-22:00:48, when both a cell tier camera (EFTA00033046) and the control room camera (EFTA00033240) simultaneously showed no guard present. This is the only documented moment in the corpus where both physical and electronic surveillance demonstrably failed at the same time (PAPER TRAIL Project, 2026a).

What 98.4% Means

The 98.4% no-coverage rate means the released corpus cannot answer its most important operational question — "was anyone watching?" — for the vast majority of guard absences. Whether the 243 unevaluable gaps resembled the 3 staffed cases (patrol rotation) or the 1 unstaffed case (total surveillance failure) is unknowable from what DOJ released. This is a finding about the disclosure, not MCC's practices.

The 98.4% no-coverage rate means the corpus cannot answer its most important operational question for the vast majority of guard absences. Were the remaining 243 floor gaps like the 3 staffed cases (patrol rotation, control room monitoring maintained) or like the 1 unstaffed case (total surveillance failure)? The data does not say.

This is not a finding about MCC's actual surveillance practices. It is a finding about what the released footage can and cannot prove. The surveillance system at MCC likely recorded continuous control room footage. What was released to the public covers only fragments. For 98% of the documented guard absences, the question "was anyone watching?" remains open (PAPER TRAIL Project, 2026a).

The full control room analysis details what the 55 annotated control room videos do show: 4.1% guard absence, the lowest of any camera class, with the arrest-night concurrent gap as the sole confirmed instance of simultaneous floor and control room failure. The 278 guard gaps and the escalation pattern from 4% to 34% absence rates — driven by a phase-level staffing failure — are the findings this blind spot prevents us from fully interpreting.

As Rep. Nancy Mace has noted, the DOJ may be hiding "terabytes of data." The 98% coverage gap in the control room footage is consistent with that concern — not because the data was necessarily withheld, but because what was released cannot answer the most important question about what was not.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

This investigation is part of the SubThesis accountability journalism network.

The Lag Effect

2026-05-11T00:00:00.000Z

TLDR

Guard absence rates correlate with detention phases, not individual events. The pre-arrest baseline was 4.7%. The arrest-through-suicide-attempt period held steady at 6.2%. Suicide watch actually decreased absence to 5.9%. The escalation to 12.9% began only after suicide watch was removed around July 29, peaking at 34.2% on August 2 and reaching 23.8% in the three days surrounding death. The system responded correctly to the first crisis. It failed after the response was withdrawn (PAPER TRAIL Project, 2026a).

Phase-Driven, Not Event-Driven

Guard absence rates at MCC track detention phases, not single events — no spike appears at the arrest, arraignment, bail denial, or suicide attempt. The transitions between phases produce gradual shifts rather than sharp breaks, indicating a systemic staffing pattern rather than an incident-specific breakdown. The escalation accelerated only after suicide watch was removed, suggesting policy change drove the failure.

The per-date guard absence data shows no spike tied to any single event -- not the arrest, not the arraignment, not the bail denial, not the suicide attempt. Instead, the data tracks detention phases. The transition from one phase to the next produces gradual shifts, not sharp breaks (PAPER TRAIL Project, 2026a).

This distinction matters. An event-driven failure would suggest a specific breakdown triggered by a specific incident. A phase-driven failure suggests a systemic pattern -- a gradual erosion of staffing that accelerated after a policy change (the removal of suicide watch).

The Five Detention Phases

The corpus splits into five detention phases with distinct absence rates: pre-arrest (4.7%, baseline), arrest to suicide attempt (6.2%, 1.3x), suicide watch (5.9%, 1.3x), watch removed to death (12.9%, 2.7x), and post-death (11.9%, 2.5x). The pre-arrest figure spans 1,021 frames from routine March 2019 operations. The 2.7x jump after watch removal is the analytical centerpiece.

Phase	Date Range	Frames	Absent	Rate	vs Baseline
Pre-arrest	Mar 1 - Jul 5	1,021	48	4.7%	1.0x
Arrest to suicide attempt	Jul 6 - 22	2,885	178	6.2%	1.3x
Suicide watch	Jul 23 - 29	1,213	72	5.9%	1.3x
Watch removed to death	Jul 30 - Aug 10	1,023	132	12.9%	2.7x
Post-death	Aug 11 - 12	210	25	11.9%	2.5x

The pre-arrest baseline (March 2019 footage plus early July) shows a 4.7% absence rate across 1,021 frames. This is what MCC looked like during routine operations -- before Epstein, before heightened scrutiny, before any special protocols (PAPER TRAIL Project, 2026a).

The arrest-to-suicide-attempt phase (July 6-22) shows a modest increase to 6.2%. The facility was processing a high-profile detainee, but staffing held near baseline. The system was not yet under strain.

The Suicide Watch Paradox

The suicide watch period (July 23-29) produced a 5.9% guard absence rate — lower than the preceding arrest-to-attempt phase (6.2%). When Epstein was found injured on July 23, MCC placed him on watch and absence rates dropped. The protocol worked. This proves MCC was capable of maintaining or improving guard presence; the subsequent failure was a departure from demonstrated capability, not inevitability.

The suicide watch period (July 23-29) shows a 5.9% absence rate -- lower than the preceding arrest period (6.2%). The system responded correctly to the first crisis. When Epstein was found injured on July 23, the facility placed him on suicide watch. Guard absence rates dropped. The protocol worked (PAPER TRAIL Project, 2026a).

This is the most important finding in the phase analysis. It demonstrates that MCC was capable of maintaining or improving guard presence when the situation demanded it. The subsequent failure was not inevitable -- it was a departure from demonstrated capability.

The Escalation After Watch Removal

In the 12 days between watch removal (around July 29) and Epstein's death (August 10), guard absence more than doubled from 5.9% to 12.9% — a 2.7x multiple of the pre-arrest baseline. The week of July 29 marked the inflection at 10.7%, double the prior week's 5.1%. The week of August 5 peaked at 15.6% — the highest weekly rate in the corpus.

Suicide watch was removed around July 29, 2019. In the 12 days between removal and death (July 30 - August 10), the guard absence rate more than doubled from 5.9% to 12.9% -- a 2.7x multiple of the pre-arrest baseline (PAPER TRAIL Project, 2026a).

The weekly progression shows the inflection point clearly:

Week of	Absent/Frames	Rate	Notes
2019-07-01	39/509	7.7%	Includes arrest day
2019-07-08	66/1,215	5.4%	Arraignment, search, steady
2019-07-15	75/1,016	7.4%	Bail denial, no spike
2019-07-22	72/1,406	5.1%	Suicide attempt + watch
2019-07-29	59/550	10.7%	Inflection -- watch removed
2019-08-05	101/647	15.6%	Peak week -- includes death
2019-08-12	7/142	4.9%	Snap recovery to baseline

The week of July 29 is the inflection point: 10.7%, double the prior week's 5.1%. By the week of August 5, the rate reaches 15.6% -- the highest weekly rate in the corpus. Then, the week of August 12, it snaps back to 4.9% (PAPER TRAIL Project, 2026a).

Event-Anchored Windows

Three-day windows centered on specific events confirm no event-driven spikes: the arrest window registered 7.4%, FBI search 4.1% (below baseline), suicide attempt 4.5% (system tightening in response), and bail denial 6.3%. Only the will-signing (13.2%) and death (23.8% — 5x baseline) windows show elevated rates, both falling within the post-watch-removal deterioration period rather than reflecting event-specific responses.

Three-day windows centered on specific events confirm the phase pattern:

Event	Date	3-Day Absent %	Context
Arrest	07/06	7.4%	Normal
Arraignment	07/08	7.2%	Normal
FBI search	07/11	4.1%	Below baseline
Bail denied	07/18	6.3%	Normal
Suicide attempt	07/23	4.5%	Below baseline -- heightened response
Will signed	08/08	13.2%	Elevated -- during staffing crisis
Death	08/10	23.8%	5x baseline

No individual event produces a spike. The arrest window (7.4%) is barely above baseline. The FBI search window (4.1%) is below it. The suicide attempt window (4.5%) shows the system tightening in response to crisis. The will-signing window (13.2%) and death window (23.8%) reflect the sustained post-watch-removal deterioration, not a response to those specific events (PAPER TRAIL Project, 2026a).

The Noel and Thomas Connection

The phase data corroborates the November 19, 2019 indictment of correctional officers Tova Noel and Michael Thomas (USA v. Noel and Thomas, 19-CR-830, S.D.N.Y.), charged with falsifying records on the night of Epstein's death. Prosecutors alleged both officers failed to perform required rounds and fabricated log entries. Surveillance data confirms the premise: absence rates during the final 12 days ran 2-3 times higher than MCC's demonstrated baseline.

The phase analysis is consistent with the DOJ Office of Inspector General finding that correctional officers Tova Noel and Michael Thomas were indicted for falsifying records on the night of Epstein's death (USA v. Noel and Thomas, 19-CR-830, S.D.N.Y., filed November 19, 2019). The charges alleged that both officers failed to perform required rounds and fabricated log entries. The surveillance data corroborates the premise of those charges: guard absence rates during the final 12 days were 2-3 times higher than the facility's demonstrated baseline (PAPER TRAIL Project, 2026a).

What the Lag Effect Describes

The lag effect is the 12-day gap between the removal of suicide watch (around July 29) and Epstein's death (August 10) — the period when the protective measure was withdrawn and the failure it was designed to prevent unfolded. Absence rates began escalating immediately, reaching 34.2% on August 2, the highest single-day rate in the corpus. MCC chose not to maintain its demonstrated 5.9% standard.

The "lag effect" is the temporal gap between the removal of a protective measure (suicide watch) and the onset of the failure it was designed to prevent (unmonitored detention leading to death). The watch was removed around July 29. Absence rates began escalating immediately. By August 2, they had reached 34.2% -- the highest single-day rate in the corpus. Epstein was found dead August 10.

The lag was 12 days. The system demonstrated it could maintain guard presence during suicide watch (5.9%). It chose not to maintain that standard after watch was removed. The surveillance footage does not explain why. It documents that it happened.

The per-date escalation data shows the daily trajectory. The 278 total guard gaps across the 419-video corpus provide the statistical baseline. The arrest-night concurrent gap and control room analysis show that even electronic monitoring failed at critical moments. And the 98% blind spot in control room coverage means most of the post-watch-removal period cannot be evaluated against electronic surveillance data.

These findings exist because the Epstein Files Transparency Act compelled the DOJ to release this footage. The House Judiciary Committee and House Oversight Committee have jurisdiction to investigate what it shows. Neither has scheduled a hearing.

References

PAPER TRAIL Project. (2026a). Jail surveillance video analysis results [Data set]. Script 34 output, _exports/video_analysis/.

U.S. Department of Justice. (2025). Epstein document release, Data Set 8: MCC surveillance footage [Government records].

USA v. Noel and Thomas, No. 19-CR-830 (S.D.N.Y., filed Nov. 19, 2019).

This investigation is part of the SubThesis accountability journalism network.

Rep. Jim Jordan and the Epstein Files: What Western Ohio Constituents Should Know

2026-05-11T00:00:00.000Z

TLDR

As Chairman of the House Judiciary Committee, Rep. Jim Jordan (R-OH) holds the primary gatekeeper role over DOJ and FBI oversight hearings on Epstein file compliance — and has not scheduled a single one, despite bipartisan demands and a law that passed 427-1.

The Chairman's Gatekeeper Role

As Chairman of the House Judiciary Committee, Rep. Jim Jordan (R-OH4) controls hearings, witnesses, and subpoenas related to DOJ compliance with the Epstein Files Transparency Act. He also sits on House Oversight and Government Reform, giving him a secondary channel of authority. Jordan co-founded the House Freedom Caucus, has represented OH-4 since 2007, and operates from 2056 Rayburn House Office Building, (202) 225-2676.

Rep. Jim Jordan represents Ohio's 4th Congressional District and chairs the House Judiciary Committee, the body with primary oversight jurisdiction over the Department of Justice and the Federal Bureau of Investigation. In this role, Jordan controls which hearings are scheduled, which witnesses are called, and which subpoenas are issued related to DOJ compliance with the Epstein Files Transparency Act (Epstein Files Transparency Act, Pub. L. No. 119-38, 2025). He also sits on the House Oversight and Government Reform Committee, giving him a secondary channel of authority over executive branch compliance (PAPER TRAIL Project, 2026).

Jordan is a co-founder of the House Freedom Caucus, a former two-time NCAA wrestling champion at the University of Wisconsin, and a former assistant wrestling coach at Ohio State University. He holds a J.D. from Capital University and has represented OH-4 since 2007. His office is located at 2056 Rayburn House Office Building, and his office phone is (202) 225-2676 (U.S. House of Representatives, 2026).

What Jordan Has — and Has Not — Done on the Epstein Files

Jordan has not scheduled a single Judiciary Committee hearing on DOJ Epstein file compliance, despite the law passing 427-1 and DOJ releasing only roughly 58% of identified pages (a 42% gap) by the January 30, 2026 "fifth and final" declaration. Judiciary Democrats formally demanded public hearings; he has not committed. Jordan has also stated he has not personally reviewed the unredacted files at the DOJ annex.

H.R. 4405, the Epstein Files Transparency Act, passed the House 427-1 and was signed into law as P.L. 119-38 on November 19, 2025 (Epstein Files Transparency Act, Pub. L. No. 119-38, 2025). The law mandated that DOJ release all unclassified Epstein-related records within 30 days, establishing a December 19, 2025 deadline.

DOJ missed that deadline. On January 30, 2026, the Department declared its release "fifth and final," having published approximately 3.5 million of 6 million-plus identified pages — a compliance rate of roughly 58%, leaving a 42% gap (PAPER TRAIL Project, 2026). As Judiciary Chairman, Jordan holds the committee gavel that could compel DOJ to explain that gap under oath.

Judiciary Committee Democrats formally demanded that Jordan hold public hearings on DOJ noncompliance. He has not committed to doing so (PAPER TRAIL Project, 2026). Jordan has stated that he had not personally reviewed the unredacted files available at the DOJ annex — the secure review station where members of Congress can view documents that remain redacted in public releases (PAPER TRAIL Project, 2026).

DOJ Surveillance of Congressional Oversight

DOJ has been surveilling which members search which names in the Epstein files — AG Bondi was photographed holding a printout labeled with a member's search history. The review annex itself limits access to four computers, 9 AM-6 PM weekdays, with 24-hour advance notice. The agency subject to oversight is tracking the overseers, and Judiciary Chairman Jordan has not addressed it.

The review conditions at the DOJ annex raise their own concerns. Members are limited to four computers, available 9 AM to 6 PM Monday through Friday, with 24-hour advance notice required. More troubling, DOJ has been surveilling which members search which names — AG Bondi was photographed holding a printout labeled with a member's search history (PAPER TRAIL Project, 2026). The agency subject to oversight is tracking the overseers, and the committee chairman with jurisdiction over that agency has not addressed it.

What OH-4 Constituents Should Know

Western Ohio voters elected Jordan in part for his reputation as an aggressive government watchdog and co-founder of the Freedom Caucus. The Epstein Files Transparency Act passed 427-1, yet DOJ has released only 58% of responsive documents and is surveilling members reviewing the rest. Jordan's Judiciary Committee holds the subpoena power, jurisdictional mandate, and precedent to act — whether he uses it is the question constituents face.

Western Ohio voters elected Jim Jordan in part because of his reputation as an aggressive government watchdog. The Freedom Caucus he co-founded was built on the premise that federal agencies require adversarial oversight. The Epstein Files Transparency Act passed with near-unanimous support — 427-1 — yet the DOJ has released only 58% of responsive documents and is surveilling the members of Congress who attempt to review the rest. Jordan's Judiciary Committee has the subpoena power, the jurisdictional mandate, and the precedent to hold DOJ accountable. Whether he uses it is a question his constituents in Lima, Urbana, and across the 4th District are entitled to ask.

The surveillance footage his committee has jurisdiction over reveals 278 guard gaps and an escalation from 4% to 34% absence in the days before Epstein's death. The Epstein Files Transparency Act compelled its release — Sen. Grassley chairs the Senate counterpart.

References

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025). https://www.congress.gov/bill/119th-congress/house-bill/4405

PAPER TRAIL Project. (2026). Congressional oversight actions, March 2026 [Data set].

PAPER TRAIL Project. (2026). DOJ compliance status [Data set].

U.S. House of Representatives. (2026). Member directory [Data set].

This investigation is part of the SubThesis accountability journalism network.

Sen. Chuck Grassley and the Epstein Files: What Iowa Constituents Should Know

2026-05-11T00:00:00.000Z

TLDR

As Chairman of the Senate Judiciary Committee and a member of the Finance Committee, Sen. Chuck Grassley (R-IA) has dual jurisdiction over both DOJ compliance and the financial crimes documented in the Epstein corpus — yet has scheduled no hearing on either front despite publicly stating the law "was clear."

Dual Jurisdiction, Zero Hearings

Grassley chairs Senate Judiciary and sits on Finance — simultaneous jurisdiction over both DOJ compliance and the financial institutions that enabled Epstein. He is the longest-serving senator in Iowa history, in office since 1981, and at 92 serves as President Pro Tempore, third in presidential succession. He has used neither committee gavel to schedule a hearing on either dimension of the Epstein case.

Sen. Chuck Grassley is the longest-serving senator in Iowa history, holding office since 1981. At 92, he serves as President Pro Tempore of the Senate — third in the line of presidential succession — and chairs the Senate Judiciary Committee while sitting on the Finance Committee (U.S. Senate, 2026). His office is located at 135 Hart Senate Office Building, and his office phone is (202) 224-3744.

This dual positioning gives Grassley something no other member of Congress possesses: simultaneous jurisdiction over DOJ compliance with the Epstein Files Transparency Act and over the financial institutions that enabled the Epstein operation. He has used neither (PAPER TRAIL Project, 2026).

Grassley's Public Statements vs. Committee Action

Grassley stated publicly the law "was clear," questioned FBI Director Kash Patel directly demanding records, and declared "victims deserve an answer." These statements establish that he understands the legal mandate and moral stakes. What they did not produce was a single Senate Judiciary hearing on DOJ noncompliance — despite DOJ declaring its January 30, 2026 release "final" at roughly 58% of 6 million-plus identified pages, a 42% gap.

Grassley has not been silent on the Epstein files. He stated publicly that the law "was clear" and that DOJ should release all files (PAPER TRAIL Project, 2026). He questioned FBI Director Kash Patel directly, demanding that the Bureau turn over all records. He declared that "victims deserve an answer" (PAPER TRAIL Project, 2026).

These statements establish that Grassley understands both the legal mandate and the moral stakes. What they did not produce was a single Senate Judiciary Committee hearing on DOJ noncompliance. The Department declared its January 30, 2026 release "final" at approximately 58% of 6 million-plus identified pages — a 42% gap — and Grassley's committee has not called a single DOJ official to testify under oath about that gap (PAPER TRAIL Project, 2026).

The Financial Dimension

The Epstein corpus documents $24.1 million in wire transfers across 224 transactions, including transfers routed through shell entities and offshore structures — financial activity squarely within Finance Committee jurisdiction. The New York Department of Financial Services imposed a $150 million consent order against Deutsche Bank for Epstein-related compliance failures, one of the largest penalties in NYDFS history. Grassley sits on Finance and has not requested a hearing on these banking failures.

The Epstein corpus documents financial activity that falls squarely within the Finance Committee's jurisdiction. The records contain $24.1 million in wire transfers across 224 transactions, including transfers routed through shell entities and offshore structures (PAPER TRAIL Project, 2026). The New York Department of Financial Services imposed a $150 million consent order against Deutsche Bank for compliance failures related to the Epstein relationship — one of the largest penalties in NYDFS history (NYDFS Consent Order, Deutsche Bank AG, 2020).

The Finance Committee has jurisdiction over financial institutions, banking regulation, and the anti-money laundering framework that Deutsche Bank and other institutions failed to enforce. Grassley sits on that committee. He has not requested a hearing on the banking failures documented in the corpus (PAPER TRAIL Project, 2026).

What Iowa Constituents Should Know

Iowa's senior senator holds the two committee gavels most relevant to the Epstein case — Judiciary over DOJ compliance and Finance over the banking failures documented in the corpus. The Transparency Act passed near-unanimously, DOJ has complied with barely more than half, and the corpus documents $24.1 million in suspicious transfers. Whether constituents in Des Moines, Cedar Rapids, and across Iowa hear sworn testimony depends entirely on whether Grassley schedules the hearings his own statements suggest are necessary.

Iowa voters have sent Chuck Grassley to Washington for more than four decades in part because of his reputation as a tenacious overseer of federal agencies. He built a career on whistleblower protections and inspector general reforms. The Epstein Files Transparency Act passed with near-unanimous support, DOJ has complied with barely more than half of it, and the corpus documents tens of millions of dollars in suspicious financial activity that existing banking regulations were designed to catch. Grassley holds the two committee gavels most relevant to both problems. Whether constituents in Des Moines, Cedar Rapids, and across Iowa hear testimony under oath depends entirely on whether their senior senator schedules the hearings his own public statements suggest are necessary.

The 419 surveillance videos his Judiciary Committee could investigate show a phase-driven staffing failure at MCC. The financial jurisdiction overlaps with Rep. Lynch and Rep. Sherman on the House side.

References

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025). https://www.congress.gov/bill/119th-congress/house-bill/4405

NYDFS Consent Order, Deutsche Bank AG (2020). New York Department of Financial Services.

PAPER TRAIL Project. (2026). Congressional oversight actions, March 2026 [Data set].

PAPER TRAIL Project. (2026). DOJ compliance status [Data set].

PAPER TRAIL Project. (2026). Financial transaction analysis [Data set].

U.S. Senate. (2026). Member directory [Data set].

This investigation is part of the SubThesis accountability journalism network.

Rep. Ro Khanna and the Epstein Files: What Silicon Valley Constituents Should Know

2026-05-11T00:00:00.000Z

TLDR

Rep. Ro Khanna authored the Epstein Files Transparency Act, read six redacted co-conspirator names on the House floor, and continues to push for full DOJ compliance with the law he wrote — making him the single member of Congress most identified with the transparency effort.

The Author of the Law

Khanna is the primary author of H.R. 4405, the Epstein Files Transparency Act, which passed the House 427-1, cleared the Senate by unanimous consent, and was signed as P.L. 119-38 on November 19, 2025. He represents California's 17th District, holds a Yale Law J.D., lectured at Stanford, and served as Deputy Assistant Secretary of Commerce under Obama. He sits on Oversight and Armed Services.

Rep. Ro Khanna represents California's 17th Congressional District — the heart of Silicon Valley — and is the primary author of H.R. 4405, the Epstein Files Transparency Act. The bill passed the House 427-1, cleared the Senate by unanimous consent, and was signed into law as P.L. 119-38 on November 19, 2025 (Epstein Files Transparency Act, Pub. L. No. 119-38, 2025). Khanna holds a J.D. from Yale Law School, has lectured at Stanford, and served as Deputy Assistant Secretary of Commerce under President Obama. He has represented CA-17 since 2017, with his office at 306 Cannon House Office Building, phone (202) 225-2631 (U.S. House of Representatives, 2026).

Khanna serves on both the House Oversight and Government Reform Committee and the Armed Services Committee, giving him oversight jurisdiction over DOJ compliance and national security dimensions of the Epstein corpus (PAPER TRAIL Project, 2026).

Six Names on the House Floor

In February 2026, Khanna read six redacted co-conspirator names on the House floor — names appearing unredacted at the DOJ annex but blacked out in public releases. By reading them into the record, he invoked the Speech or Debate Clause, which shields members from legal liability for legislative statements. On PBS NewsHour he stated the files had been "scrubbed to protect elite, powerful men," based on direct annex comparison of redacted and unredacted versions.

In February 2026, Khanna took the extraordinary step of reading six redacted co-conspirator names on the House floor. These names appeared in unredacted files available at the DOJ annex but had been blacked out in public releases. By reading them into the congressional record, Khanna invoked the Speech or Debate Clause of the Constitution, which protects members of Congress from legal liability for statements made during legislative proceedings (PAPER TRAIL Project, 2026).

In a PBS NewsHour interview, Khanna stated that the files had been "scrubbed to protect elite, powerful men" (PBS NewsHour, 2026). This was not vague accusation — it was an assertion based on his direct comparison of redacted and unredacted versions of the same documents at the DOJ annex review station.

Legal Action Threats and Coalition Building

Khanna joined Senate Majority Leader Chuck Schumer, Rep. Jamie Raskin, and Rep. Robert Garcia in threatening legal action against DOJ for noncompliance with P.L. 119-38. The coalition spans both chambers and includes the law's author plus ranking members of committees with enforcement jurisdiction. DOJ had released approximately 3.5 million of 6 million-plus identified pages — roughly 58% compliance — declaring its January 30, 2026 release "fifth and final." The 42% gap represents about 2.5 million withheld pages.

Khanna did not act alone. He joined with Senate Majority Leader Chuck Schumer, Rep. Jamie Raskin, and Rep. Robert Garcia in threatening legal action against DOJ for noncompliance with P.L. 119-38 (PAPER TRAIL Project, 2026). The coalition represented both chambers and both the author of the law and the ranking members of the committees with enforcement jurisdiction.

DOJ had released approximately 3.5 million of 6 million-plus identified pages — roughly 58% compliance — and declared its January 30, 2026 release "fifth and final." The 42% gap represents approximately 2.5 million pages that DOJ identified as responsive but has not made public. Khanna's position is that the law he wrote contains no exception for embarrassment, political sensitivity, or institutional convenience (PAPER TRAIL Project, 2026).

What Silicon Valley Constituents Should Know

CA-17 voters are represented by the author of P.L. 119-38, who personally reviewed unredacted files at the DOJ annex, identified concealed names, read six into the congressional record under Speech or Debate protection, and joined a Schumer-Raskin-Garcia coalition threatening legal action to compel the remaining 42%. Whether that legal action materializes — and whether the law produces full compliance — is the open question for San Jose, Fremont, and Sunnyvale voters.

CA-17 voters sent Ro Khanna to Congress as a representative of one of the most educated and technologically sophisticated districts in the country. His constituents include engineers, lawyers, and entrepreneurs who understand document management, data integrity, and compliance frameworks. Khanna authored the law that requires full Epstein file disclosure, personally reviewed the unredacted files, identified names DOJ is concealing, read those names into the congressional record, and has threatened legal action to compel the remaining 42%. Whether that legal action materializes — and whether the law he wrote ultimately produces full compliance — is a question his constituents in San Jose, Fremont, and Sunnyvale are positioned to evaluate with particular sophistication.

The law Khanna authored compelled release of the 419 MCC surveillance videos that revealed 278 guard gaps, a 34% peak absence rate, and a phase-driven staffing failure — evidence that exists in the public record only because of P.L. 119-38. Rep. Massie filed the discharge petition that made it possible.

References

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025). https://www.congress.gov/bill/119th-congress/house-bill/4405

PBS NewsHour. (2026). Interview with Rep. Ro Khanna on Epstein file redactions.

PAPER TRAIL Project. (2026). Congressional oversight actions, March 2026 [Data set].

PAPER TRAIL Project. (2026). DOJ compliance status [Data set].

U.S. House of Representatives. (2026). Member directory [Data set].

This investigation is part of the SubThesis accountability journalism network.

Rep. Thomas Massie and the Epstein Files: What Northern Kentucky Constituents Should Know

2026-05-11T00:00:00.000Z

TLDR

Thomas Massie filed the discharge petition that broke H.R. 4405 out of committee and forced a floor vote, making him the Republican who made the Epstein Files Transparency Act procedurally possible when his own party's leadership refused to act.

The Petition That Changed Everything

In September 2025 Massie filed H.Res. 581 — a discharge petition forcing H.R. 4405 to the House floor against his own party's committee leadership. A discharge petition requires 218 signatures to bypass a chairman who refuses to schedule a vote, making it one of the rarest and most confrontational procedural tools. Massie holds MIT engineering degrees (B.S. and M.S.), founded SensAble Devices, holds 24 patents, and represents Kentucky's 4th District.

Rep. Thomas Massie represents Kentucky's 4th Congressional District and holds degrees in engineering from MIT (both B.S. and M.S.). Before Congress, he founded SensAble Devices, holds 24 patents, and served as Lewis County, Kentucky's judge-executive. He has served in Congress since 2012 and is an original cosponsor and lead cosponsor of H.R. 4405. His office is located at 2371 Rayburn House Office Building (U.S. House of Representatives, 2026).

In September 2025, Massie filed H.Res. 581 — a discharge petition to force H.R. 4405, the Epstein Files Transparency Act, to the House floor. A discharge petition is one of the rarest and most confrontational procedural tools available. It requires 218 signatures — a majority of the full House — to bypass committee leadership and force a vote that the chairman has refused to schedule. Filing one against your own party's committee chairman is an act of open defiance (PAPER TRAIL Project, 2026).

218 Signatures: 214 Democrats, 4 Republicans

The petition reached 218 signatures on November 12, 2025: 214 Democrats and exactly 4 Republicans. Out of roughly 220 House Republicans, only Massie, Marjorie Taylor Greene, Nancy Mace, and Lauren Boebert were willing to defy leadership publicly on Epstein transparency. Massie framed the stakes plainly: "We can't avoid justice to avoid embarrassment for some very powerful men." The bill would ultimately pass 427-1.

On November 12, 2025, the discharge petition reached the 218-signature threshold. The breakdown tells the story: 214 Democrats and 4 Republicans. Out of approximately 220 House Republicans, exactly four were willing to publicly defy leadership on Epstein file transparency. Massie was joined by Marjorie Taylor Greene, Nancy Mace, and Lauren Boebert (PAPER TRAIL Project, 2026).

Massie framed the stakes plainly: "We can't avoid justice to avoid embarrassment for some very powerful men" (PAPER TRAIL Project, 2026). This statement captured the core tension — the bill had broad support in principle (it would pass 427-1), but party leadership resisted bringing it to a vote, and only four Republicans were willing to force the issue.

From Petition to Law

H.R. 4405 passed the House 427-1 under suspension of rules on November 18, 2025, cleared the Senate by unanimous consent the next day, and was signed by President Trump as P.L. 119-38 on November 19, 2025. The gap between 4 Republican petition signers and 427 yes votes illustrates a recurring pattern: members will vote for transparency when on the record but will not take the procedural risk of forcing the vote. Massie took it.

The discharge petition forced a floor vote on November 18, 2025. H.R. 4405 passed 427-1 under suspension of rules. The Senate passed it by unanimous consent the following day. President Trump signed it into law as P.L. 119-38 on November 19, 2025 (Epstein Files Transparency Act, Pub. L. No. 119-38, 2025).

The gap between the 4 Republicans who signed the petition and the 427 who voted yes on the bill itself illustrates a recurring pattern in Congress: members will vote for transparency when forced to go on the record, but will not take the procedural risk of forcing that vote. Massie took the procedural risk. As a member of the Judiciary Committee — the very committee whose chairman declined to act — he filed the petition against his own committee's leadership (PAPER TRAIL Project, 2026).

What Northern Kentucky Constituents Should Know

KY-4 voters elected an MIT engineer and inventor who applied a problem-solver's approach to a stalled bill: when normal process failed, he used the discharge petition and built the bipartisan coalition needed for 218 signatures. The law exists because of that petition. Whether it produces full compliance — DOJ has released only 58% of responsive documents — depends on whether the enforcement pressure begun with Massie's procedural confrontation is sustained.

KY-4 voters elected an engineer and inventor who built a career on solving problems others declared unsolvable. Thomas Massie applied that same approach to the Epstein Files Transparency Act: when the normal legislative process failed, he used the procedural tool available — the discharge petition — and built the bipartisan coalition needed to reach 218 signatures. The law exists because of that petition. Whether it produces full compliance from DOJ — which has released only 58% of responsive documents — depends on whether the enforcement pressure that began with Massie's procedural confrontation can be sustained. Constituents in Covington, Florence, Ashland, and across the 4th District should know that their representative was one of four Republicans willing to defy party leadership to make this law possible.

The discharge petition Massie filed ultimately led to the release of 419 surveillance videos showing escalating guard absence and an arrest-night surveillance blackout. Rep. Khanna authored the underlying bill; Rep. McGovern forced the Rules Committee confrontation that triggered the petition.

References

Epstein Files Transparency Act, Pub. L. No. 119-38 (2025). https://www.congress.gov/bill/119th-congress/house-bill/4405

PAPER TRAIL Project. (2026). Congressional oversight actions, March 2026 [Data set].

PAPER TRAIL Project. (2026). Discharge petition H.Res. 581 analysis [Data set].

U.S. House of Representatives. (2026). Member directory [Data set].

This investigation is part of the SubThesis accountability journalism network.