We Built a Due Diligence Engine and Pointed It at 124 Names
- Patrick Duggan
- Feb 22
- 13 min read
title: "We Built a Due Diligence Engine and Pointed It at 124 Names"
subtitle: "The CARVER Calculator: Cross-index risk scoring across 7.4 million documents in 556 milliseconds"
date: 2026-02-22
author: Patrick Duggan
category: deep-dive
tags:
- CARVER
- due diligence
- Epstein Files
- ICIJ Offshore
- risk scoring
- OSINT
- Butterbot
- compliance
- AML
- KYC
# We Built a Due Diligence Engine and Pointed It at 124 Names
*The CARVER Calculator: Cross-index risk scoring across 7.4 million documents in 556 milliseconds*
We took the CARVER methodology -- a military targeting framework used by US Army Special Operations Command to prioritize targets for direct action -- and we built an automated due diligence engine out of it. Six dimensions. Each scored 1-5. Total out of 30. The kind of thing that used to take a team of analysts six weeks and a retainer that starts with a comma.
Then we pointed it at 124 names. People, companies, shell entities. We let it search 7.4 million documents across 8 indexes. Epstein files. ICIJ offshore leaks. Federal court decisions. Threat actor databases. Institutional abuse records. Media coverage. Cyber indicators. Phishing campaigns.
It took 556 milliseconds.
Here's what it found.
What CARVER Is
CARVER stands for Criticality, Accessibility, Recuperability, Vulnerability, Effect, and Recognizability. The US military uses it to decide which targets to hit first. A bridge that's critical to supply lines, accessible to a strike team, hard for the enemy to rebuild, vulnerable to a specific weapon system, would produce cascading effects if destroyed, and is easy for a pilot to identify at night -- that bridge gets a high CARVER score. That bridge gets hit.
We're not hitting bridges. We're doing due diligence.
Here's how we adapted each dimension:
**C -- Criticality**: How critical are the findings? We weight Tier 1 indexes (Epstein Files, ICIJ Offshore Leaks) at 3x because if your name appears in both a trafficking network's documents AND a database of hidden offshore structures, that's not a coincidence. That's a pattern. Tier 2 indexes (Federal Decisions, Threat Actors, Institutional Abuse) weight at 2x. Tier 3 (Media, Cyber Indicators, Phishing) weight at 1x. A weighted score above 200 earns a 5/5. Below 5 earns a 1/5.
**A -- Accessibility**: How much evidence exists? Pure volume plus source diversity. If you have 500+ documents spread across 4 or more indexes, you score a 5. If you have 3 documents in one index, you score a 1. More paper means more accessible evidence surface. Simple math.
**R -- Recuperability**: Can you recover from this exposure? This is the defense metric. How many distinct exposure *categories* are you in? If your name appears in trafficking network documents AND financial crime records AND public awareness campaigns AND institutional abuse databases -- that's four categories. Multi-dimensional exposure is the kind you don't come back from. Ask Ghislaine Maxwell.
**V -- Vulnerability**: Evidence of lies, cover-ups, contradictions? This is the cross-index smoking gun detector. If you appear in the Epstein files AND the ICIJ offshore leaks, that's trafficking network proximity plus hidden money. Score: 5/5. If you're in the Epstein files AND federal court decisions, that's network proximity plus official scrutiny. Score: 4/5. The machine is looking for the gap between what the record shows and what you've said publicly.
**E -- Effect**: What happens if the worst-case is true? Based on document volume in the three critical indexes: Epstein Files, ICIJ Offshore, and Federal Decisions. More than 200 documents across those three? The effect of full exposure would be seismic. Score: 5/5. Zero documents? Nothing happens. Score: 1/5.
**R2 -- Recognizability**: Does anyone know who you are? A target nobody recognizes has low due diligence value no matter how dirty the paper trail. Blog hits plus search query frequency. If people are already Googling your name in this context, the story is already in motion.
One more thing: this is NOT a guilt score. I need to say that clearly because some idiot is going to screenshot a number and post it with "GUILTY" in all caps. A high CARVER score means "there is a lot of paper on you and it looks bad." It measures document exposure. The distance between "there's a lot of paper" and "you did it" is where lawyers, judges, and juries live. The CARVER tells them where to look. It does not tell them what to find.
The Engine
The API endpoint is `POST /api/v1/carver/evaluate`. Professional tier, $99/month. You send us a name, we search 8 indexes in parallel, score 6 dimensions, and return a risk classification with evidence citations. The whole thing runs through Meilisearch on a single Azure VM that costs us less than your monthly coffee habit.
Here's what happens under the hood:
**Phase 1: Cross-index phrase search.** This matters more than it sounds. Before we calibrated the engine, we were running keyword search. "Les Wexner" would match "Les" OR "Wexner" -- which returned every document that mentioned any Les and any Wexner separately. Jeff Bezos scored a 28/30 on the uncalibrated engine because "Jeff" appears in approximately every document ever written. Robert Lighthizer scored 28/30. The machine was lying to us with false positives and inflated numbers.
We fixed it. Phrase search wraps the target name in quotes. `"Les Wexner"` matches only the exact phrase. Bezos dropped from 28 to 15. Lighthizer dropped from 28 to 6. The truth was less dramatic and more useful. Intellectual honesty beats inflated numbers every single time, even when the inflated numbers would make for a better headline.
**Phase 2: Tier weighting.** Eight indexes, three tiers:
| Tier | Indexes | Weight | Why |
|------|---------|--------|-----|
| Tier 1 (Critical) | Epstein Files (329K+), ICIJ Offshore (1.7M+) | 3x | Trafficking network + hidden money. If you're in both, that's the combination. |
| Tier 2 (Significant) | Federal Decisions (1.6M+), Threat Actors, Institutional Abuse | 2x | Official scrutiny, state-sponsored threats, institutional cover-ups |
| Tier 3 (Contextual) | Media/Blog, IOCs, Phishing | 1x | Public awareness, cyber exposure, brand weaponization |
**Phase 3: Scoring.** Six dimensions calculated simultaneously. Criticality from weighted hit totals. Accessibility from volume and diversity. Recuperability from exposure category count. Vulnerability from cross-index pattern detection. Effect from critical index volume. Recognizability from media coverage and search frequency.
**Phase 4: Ma'at confidence.** Named after the Egyptian goddess of truth, because we're Irish-American in Minnesota and we can do what we want. The confidence score factors in data coverage, document volume, source diversity, and recency. It caps at 95%. Always. We NEVER claim 100% on anything. O'Toole's Axiom: Murphy was an optimist. Something will be wrong. We guarantee 5% bullshit exists. That's honest. Claiming 100% is either lying or ignorance, and both are worse than admitting uncertainty.
**Phase 5: Response.** The API returns dimension scores, total score, risk classification (LOW through CRITICAL), Ma'at confidence, evidence citations from each index, and processing metadata. The whole round trip runs in under 600 milliseconds on a good day. 556ms on the day we scored all 124 targets.
Butterbot Jr -- our AI assistant -- has the CARVER tool built in. You can ask it "Run CARVER on Deutsche Bank" in plain English and it calls the API, interprets the results, and explains what the numbers mean. That's the product. Due diligence you can talk to.
The Results: Who Scored What
We ran 124 targets through the calibrated engine. People, companies, and shell entities. Sorted them into four risk tiers. Here's the leaderboard.
HIGH (20-24) -- 18 Targets
The red zone. Significant document exposure across multiple indexes, multiple exposure categories, cross-index vulnerability signals. These are the names with the most paper on them.
| Rank | Name | Score | Docs | Key Signal |
|------|------|-------|------|------------|
| 1 | Jeffrey Epstein | 24/30 | 1,135 | 8/8 indexes hit. Maximum breadth. The center of the dataset, obviously. |
| 2 | Ghislaine Maxwell | 23/30 | 1,060 | 7/8 indexes. Primary recruitment operator. Convicted. |
| 3 | Deutsche Bank | 23/30 | 1,043 | 7/8 indexes. 4,400+ check images. 20+ Epstein-affiliated accounts. The institution scores higher than most humans. |
| 4 | Lesley Groff | 23/30 | 1,031 | 6/8 indexes. Executive assistant. 1,008 Epstein file hits. She scheduled everything. She touched everything. |
| 5 | Leon Black | 23/30 | 509 | 6/8 indexes. $158M in documented payments to Epstein. Apollo Global's original sin. |
| 6 | Jes Staley | 22/30 | 1,028 | 6/8 indexes. JP Morgan and Barclays banking relationship. 956 Epstein file hits. |
| 7 | Southern Trust | 22/30 | 1,011 | 5/8 indexes. Epstein's primary financial receiving entity. $83.5M+ flowing through. |
| 8 | Apollo Global | 22/30 | 541 | 5/8 indexes. Leon Black's firm. Epstein had an "advisory role." |
| 9 | Prince Andrew | 22/30 | 343 | 5/8 indexes. Named in civil proceedings. Settlement confirmed. Titles stripped. |
| 10 | Larry Visoski | 21/30 | 1,007 | Epstein's personal pilot. The flight log custodian. 989 Epstein file hits. Grand jury testimony. |
| 11 | Gratitude America | 21/30 | 436 | Epstein-linked charitable entity. Tax and charitable fraud vectors. |
| 12 | Kathy Ruemmler | 21/30 | 336 | Former White House Counsel. Goldman Sachs GC. Legal gatekeeper. |
| 13 | Donald Trump | 20/30 | 196 | Decades-long relationship. Property transactions. Depositions. 1,247 media mentions. |
| 14 | Steve Bannon | 20/30 | 182 | 134 Epstein file hits. 12 ICIJ hits. Fraud convictions and pardons. What is Bannon doing in 182 Epstein documents? |
| 15 | Woody Allen | 20/30 | 156 | Social circle overlap. Pattern of predatory behavior documented. |
| 16 | Bill Clinton | 20/30 | 144 | 26+ flights documented. 112 Epstein file hits. 1,089 media mentions. |
| 17 | Larry Summers | 20/30 | 133 | Treasury Secretary. Harvard president. Epstein access. Harvard donation pipeline. |
| 18 | Alexander Acosta | 20/30 | 102 | "I was told Epstein belonged to intelligence." The man who signed the NPA. |
ELEVATED (15-19) -- 33 Targets
The amber zone. Material findings. Proceed with caution.
This tier includes: Peggy Siegal (19), Michael Wolff (19), Eva Dubin (18), Robert Maxwell (18), Boris Nikolic (18), Alan Dershowitz (18), Plan D LLC (18), Les Wexner (18), Tom Pritzker (18), Brad Karp (18), Geoffrey Berman (18), Harvey Weinstein (17), Ehud Barak (17), Joichi Ito (17), Jean-Luc Brunel (17), Bill Gates (17), Peter Thiel (16), Bill Richardson (16), Peter Mandelson (16), Nicole Junkermann (16), Reid Hoffman (16), Signature Bank (15), John Brockman (15), Elon Musk (15), Jared Kushner (15), Howard Lutnick (15), Kevin Spacey (15), Josh Harris (15), Christopher Dilorio (15), Edmond de Rothschild (15), Glenn Dubin (15), Mort Zuckerman (15), Jeff Bezos (15).
MODERATE (10-14) -- 39 Targets
Some exposure. Standard due diligence sufficient. This tier includes: Wilbur Ross (14), Naomi Campbell (14), Paul Manafort (14), Michael Milken (14), Mark Zuckerberg (14), William Barr (13), Roger Stone (13), Marc Rowan (13), Sergey Brin (13), BV70 LLC (13), Mary Erdoes (13), Amanda Ellison (12), Sarah Kellen (12), Michael Flynn (12), George Mitchell (12), David Copperfield (12), Steve Cohen (12), Mohammed bin Salman (12), Henry Kissinger (12), Philippe Laffont (12), Jean Pigozzi (12), Brett Ratner (12), Doug Band (12), Nadia Marcinkova (12), Tom Barrack (12), Steve Schwarzman (12), Rudy Giuliani (11), David Koch (11), Jamie Dimon (11), David Ellison (11), Safra Catz (11), Steven Mnuchin (11), Lee Bollinger (11), Larry Ellison (10), Sarah Ferguson (10), Phillip Frost (10), Ivanka Trump (10), John Roberts (10), David Urban (10).
LOW (6-9) -- 34 Targets
Minimal findings. The document trail is clean -- in this particular dataset.
This tier includes: Mike Pompeo (9), Janet Yellen (9), Paul Volcker (9), Alberto Gonzales (9), Rich Handler (9), Adriana Ross (9), Valar Ventures (8), Huma Abedin (7), Condoleezza Rice (6), Colin Powell (6), Mette-Marit (6), Ken Moelis (6), Brock Pierce (6), Adam Back (6), Timothy Geithner (6), Ben Bernanke (6), Alan Greenspan (6), Jerome Powell (6), Robert Lighthizer (6), Gary Cohn (6), Rex Tillerson (6), John Bolton (6), Donald Rumsfeld (6), Dick Cheney (6), Paul Wolfowitz (6), George Tenet (6), Karl Rove (6), Scooter Libby (6), Elliott Abrams (6), and several others.
The Surprises
The tier breakdowns are interesting. The stories inside them are better.
Lesley Groff at 23/30
Epstein's executive assistant scores the same as Ghislaine Maxwell. 1,031 documents. She didn't fly the planes, she didn't recruit the girls, she didn't inherit an intelligence network from a dead media baron. She answered the phones and managed the calendar.
1,031 documents.
In every investigation I've ever seen -- corporate fraud, trafficking networks, RICO cases -- the executive assistant is the single most valuable source of documentary evidence. They touch everything. They schedule everything. They know who was where and when because they booked the car, reserved the table, and confirmed the appointment. Lesley Groff didn't need a high public profile. She had the calendar. And 1,008 of those 1,031 documents are in the Epstein files themselves. The operation documented her right back.
Deutsche Bank at 23/30
The institution scores higher than most individual humans in this dataset. 1,043 documents. The bank that maintained 20+ Epstein-affiliated accounts. The bank that processed 4,400+ check images. The bank that knew what was flowing through those accounts because they had a compliance department that was legally required to flag suspicious activity.
Deutsche Bank paid a $150 million fine for its Epstein relationship in 2020. The New York Department of Financial Services found that the bank had processed "hundreds of transactions totaling millions of dollars that, at the very least, should have prompted additional scrutiny." The CARVER score of 23/30 isn't the machine's opinion. It's the document trail's shadow. The institution left more paper than most people.
Steve Bannon at 20/30
This one caught me off guard. 182 documents. 134 hits in the Epstein files. 12 in the ICIJ offshore leaks. 36 in federal decisions.
What the hell is Steve Bannon doing in 134 Epstein documents?
Bannon's public narrative is the populist insurgent, the working-class warrior, Breitbart and border walls. His document trail says he operates in the same financial and political networks as the people he claims to fight against. The fraud convictions. The presidential pardon. The offshore hits. 182 documents is not casual contact. 182 documents is a paper trail.
Southern Trust at 22/30
A shell company scores higher than most named individuals. $83.5 million documented flowing through it. Southern Trust was Epstein's primary financial receiving entity -- the bucket where the money landed before it got moved to wherever it was going. 1,011 documents. 923 in the Epstein files. 78 in the ICIJ offshore leaks.
When a shell company has more document exposure than the people who used it, that tells you the financial architecture was doing more work than the humans. The money was the operation. Southern Trust was the machine.
The Phrase Search Fix
I need to talk about this because it's the most important technical decision we made, and it's the one that made the numbers less dramatic.
Before phrase search calibration, Jeff Bezos scored 28/30. After: 15/30. Robert Lighthizer scored 28/30. After: 6/30. The uncalibrated engine was matching first names and last names independently. "Jeff" appears in thousands of Epstein documents because there's a Jeff on every page of every deposition transcript. "Bill" matches Clinton, Gates, Richardson, and every billing statement in the financial records.
We could have shipped the inflated numbers. They would have gotten more clicks. "Jeff Bezos scores 28/30 on CARVER analysis" is a better headline than "Jeff Bezos scores 15/30, most of it media coverage and search interest." But inflated numbers make the tool useless. If everyone scores 28, the score means nothing. The whole point of due diligence is separating signal from noise.
We fixed it. The numbers are smaller. The numbers are real.
The Neocons Scored 6/30
Bolton, Cheney, Rumsfeld, Rice, Wolfowitz, Tenet, Rove, Libby, Abrams -- the entire Bush-era national security cabinet scores 6/30. Minimum. Zero documents in the Epstein files. Zero in the ICIJ offshore leaks. Zero in federal decisions related to this network.
These are people who authorized the invasion of Iraq on fabricated intelligence. Who ran a torture program. Who outed a CIA agent for political revenge. War criminals by any reasonable standard.
And they score 6/30.
Because the CARVER measures document exposure in these specific indexes, not moral character. Clean hands in this particular dataset doesn't mean clean hands. It means they kept their names out of Epstein's documents and the offshore leaks. The absence of evidence is not evidence of absence -- but it IS a 6/30 on the CARVER. The machine doesn't have opinions. It has indexes.
Christopher Dilorio at 15/30
The SEC whistleblower. 27 documents. He tried to expose the system and the system documented him right back. His presence in 27 documents tells you something about how these networks handle threats: they don't just silence you, they generate paper on you. The documentation is bilateral.
What This Means for Due Diligence
This isn't an Epstein toy. This is a due diligence engine that happens to be calibrated on the largest indexed collection of DOJ trafficking documents, offshore leak records, and federal court decisions in the world. The use cases are straightforward:
**M&A Due Diligence.** Before you acquire a company, run the principals through CARVER. If the CEO scores 20+ across indexes that include trafficking networks and offshore leaks, maybe you want to know that before you sign the LOI. Maybe your board wants to know. Maybe your D&O insurer wants to know.
**C-Suite Hiring.** Background checks run credit scores and criminal records. They don't search 7.4 million documents for hidden connections to sanctioned entities, offshore structures, and institutional abuse networks. The CARVER does. Not as a replacement for the background check -- as the layer that tells you what the background check can't see.
**Vendor Risk Assessment.** Is your supply chain connected to sanctioned entities? Is the company you're about to wire $2 million to owned by a shell entity that shows up in the ICIJ offshore leaks? Standard vendor risk programs check OFAC lists. The CARVER checks 8 indexes across 7.4 million documents.
**Identity Verification.** Who are you really doing business with? KYC compliance requires "reasonable due diligence." What's reasonable when 7.4 million documents are searchable in 556 milliseconds?
The CARVER doesn't replace lawyers. It doesn't replace investigators. It tells lawyers and investigators where to look. That's the value. Due diligence has always been about reading the paper. We just read 7.4 million pages faster than you can read this sentence.
The Tool
The interactive CARVER Calculator is live at:
**[https://epstein.dugganusa.com/carver-calculator.html](https://epstein.dugganusa.com/carver-calculator.html)**
124 targets. Sortable by any column. Filterable by 9 groups (Political, Financial, Tech, Inner Circle, Legal, Academic, Media, Neocon, Fed/Treasury). Click any row to expand dimension-by-dimension scoring with analyst notes for each CARVER factor. Heatmap visualization. Search by name.
The API is available for programmatic access:
Professional tier: $99/month. Enterprise (bulk screening up to 100 targets per request): $499/month. Contact [email protected].
The methodology is public: `GET /api/v1/carver/methodology` -- no auth required. We don't hide how the scoring works. Transparency is the product.
The Numbers
124 targets evaluated. 12,709 documents referenced. 556 milliseconds processing time. 8 indexes searched. 95% confidence cap.
18 scored HIGH (20-24). 33 scored ELEVATED (15-19). 39 scored MODERATE (10-14). 34 scored LOW (6-9).
The highest individual score: Jeffrey Epstein at 24/30. Not 30/30. Even the center of the dataset doesn't max out, because Recuperability and Vulnerability are scored on cross-index pattern complexity, not just volume. Epstein scores 2/5 on both -- only 2 exposure categories hit, even at massive volume. The scoring methodology is honest even when it's scoring the subject of the entire investigation.
The machine doesn't have opinions. It has indexes. Point it at a name and it tells you what the paper says. That's all due diligence ever was -- reading the paper. We just read 7.4 million pages faster than you can.
**DISCLAIMER**: CARVER scores measure document exposure, not guilt. Presence in a dataset does not imply wrongdoing. A high score means there is significant documentary evidence that warrants further investigation, not that a person or entity has committed any crime. All findings require independent verification. This tool is designed for due diligence professionals, journalists, and researchers. It is not a substitute for legal counsel.
*DugganUSA LLC -- Minneapolis, MN -- [www.dugganusa.com](https://www.dugganusa.com)*
*The public record should be searchable by the public.*
*Her name was Renee Nicole Good.*
*His name was Alex Jeffery Pretti.*
Comments