The conventional account of each case centers on deception. Holmes built a mythology around technology that never worked and enforced secrecy through NDAs, a culture of fear, and a board stocked with political legitimacy and no scientific judgment. Javice paid $18,000 to generate four million synthetic user profiles and sold them to JPMorgan as the real thing. Neumann dressed a commercial real estate sublease operation in tech-company multiples, governed by supervoting shares and documented self-dealing, and collected $13.5 billion from SoftBank at a valuation that required ignoring the unit economics in the company’s own filings.
Different industries. Different mechanisms. Different criminal charges.
The same posture pattern.
In each case, what ultimately surfaced was not something concealed in a locked room. It was something that had never been requested in a format that required evidence to answer it. The question had been asked. The answer had been given. What was missing was a structured framework that treated the absence of a verifiable artifact as a finding, not as friction.
Theranos
The board was observable. Kissinger, Shultz, Mattis, Nunn. Zero clinical or diagnostic expertise. This was not hidden; it was in the company’s own materials. A peer-reviewed study, an independent lab audit, and FDA clearance were never produced within any evidence window because none were ever demanded within one. The NDAs that prevented independent validation were disclosed. The refusal to permit an audit was a stated policy, presented as protecting proprietary technology.
When investors asked for financial controls, the request was declined. Not evaded. Declined. The response, that the information was competitively sensitive, was accepted rather than treated as a posture signal.
The core technology claim: a drop of blood, tested on a proprietary device, producing results across hundreds of assays. The operational implication of that claim, that it would require an unprecedented machine producing results at a level that would have appeared in published science, was never asked about in binary form with an artifact requirement attached.
Board composition verifiable. FDA clearance requestable. Lab audit refusal documented. Financial controls denied on record. NDAs used to block third-party validation. Zero peer-reviewed evidence in the public record for the core technology claim. All of this was knowable.
Frank
The claim was 4.25 million users of a student financial aid platform. The artifact that would have tested it: an independently supervised count of the user database, or a reconciliation of the claimed user volume against observable operational signals.
Four million active users on a platform implies server costs, customer service volume, email open rates, and marketing spend at a particular scale. These numbers exist in financial records. The P&L was available. The numbers did not reconcile. That reconciliation was never demanded as a condition of proceeding.
The founder cited user privacy when asked for direct database access. The privacy concern was accommodated. A third-party firm was engaged to verify the user count, but the method they used counted data fields rather than users, which is a different exercise. The engagement produced a number that matched the claim. The question of how the number was derived was not interrogated as a posture signal.
A competitive offer from Bank of America was cited to accelerate the timeline. The urgency was accepted. The evidence window, the period during which a clear, structured request for verification could have been made, was compressed rather than extended.
User volume claim unreconciled against P&L. Database verification refused. Third-party count methodology unaudited. Competitive urgency applied to compress timeline. A single unverified number anchored the entire valuation.
WeWork
This case is the starkest. The evidence was not behind an NDA or inside a database. It was in the S-1 that the company filed for its own IPO.
The CEO leased personal real estate to the company. He sold a personal trademark to WeWork for $5.9 million. He held supervoting shares that made board oversight structurally impossible. His wife was named in governance documents as the selector of his replacement in the event of his removal. The 2018 net loss was $1.9 billion on $1.8 billion in revenue. The S-1 contained no unit economics that would support a technology-company multiple applied to a business that signed long-term leases and subleased the space at short-term rates. Every one of these facts was in the document.
The company was valued at $47 billion before the S-1 was filed. The S-1 caused the IPO to collapse. The valuation dropped to $9 billion. The CEO was removed with a $1.7 billion exit package. SoftBank wrote down more than $10 billion. WeWork filed for Chapter 11 bankruptcy in November 2023.
Every risk that caused the collapse was visible in a public document the company itself prepared. What was missing in the years before that document was someone treating the same category of signals as findings requiring a priced response, rather than as the acceptable costs of backing a founder with vision.
Self-dealing across four documented vectors. Supervoting governance with no board remedy. Net loss exceeding revenue. Tech multiples applied to a lease-arbitrage business. Unit economics absent from investor materials. None of this was confidential. It was just never priced as a finding before the capital was committed.
The problem was not access to information. The problem was that no one treated the absence of verifiable evidence as a finding with a dollar value attached.
The lesson these cases offer is not about fraud detection. Fraud is, by definition, difficult to detect before it is known. What these cases demonstrate is something more precisely useful: the posture gaps that allow fraud to survive diligence are themselves priceable, before the fraud is known, using only information that was available at the time.
In each case, the evidence was either producible or it was not. The artifacts either existed or they did not. The contradictions between claims and observable operational signals were either visible or they were not. In each case, they were visible. What was missing was the framework to price that visibility before the check cleared.
The board of Theranos could not have been unknowingly assembled. The user count of Frank could not have been unknowingly unreconciled against the financials. The self-dealing at WeWork could not have been unknowingly present in the S-1. These were not oversights. They were observations that were never formalized into findings.
Pre-LOI intelligence does not catch fraud. It prices the absence of evidence. In each of these cases, the absence was categorical. No peer review. No database access. No governance that could survive scrutiny. Each absence, requested in binary form within a defined evidence window, would have produced the same result: a memo with a number attached to what was missing, delivered while the capital was still uncommitted.
That is a different outcome. Not because the fraud would have been discovered. Because the price of proceeding without evidence would have been explicit, and the burden of closing without resolving it would have fallen on someone who chose to proceed anyway, with their eyes open, rather than on an institution that thought it had done its diligence.