Hypothesis Forensics
The $4 Billion Question Nobody Asked Before Close
How biotech deals can close on scientific claims no one has independently traced to their sources — and what an independent forensic audit is designed to catch.
Mike Demko · Last updated 10-15-2026, 04:47 PM PT
Executive Summary
- The service. A citation-chain forensic audit walks the scientific claims behind a biotech deal backward to the primary papers they rest on, flagging every break in the evidence chain before the deal closes.
- The problem is documented. Published research points the same way: in Amgen's own attempts only 6 of 53 "landmark" cancer papers were confirmed; a 2005 essay argues a published research claim is more likely false than true in most study designs and settings; a 2015 analysis estimates about $28 billion a year of US preclinical research is not reproducible; a pooled analysis of studies of medical journal articles estimated a total quotation error rate of 25.4% (major errors 11.9%); and in a study of 100 oncology press releases from 8 large pharmaceutical companies, only 31% gave the size of the study findings.
- The gap, in our view. We found no firm marketing this as a named, reproducible service for M&A diligence (our review, April 2026; not exhaustive). Our hypothesis is that acquirers have not typically been asked to pay for it, and that founders have little incentive to look.
- The cost can be large. AbbVie reported an impairment of about $4 billion tied to Stemcentrx’s lead drug (AbbVie 8-K, January 2019). The case is set out from AbbVie’s filings and the original papers here. It illustrates diligence risk; this page does not claim the loss was caused by a break in a citation chain.
The problem, at industry scale
11%
of 53 "landmark" cancer papers were confirmed in Amgen's own attempts
Begley & Ellis, Nature, 2012 (a commentary; the authors chose papers describing something new)
20–25%
of Bayer’s 67 in-house target-validation projects had published data completely in line with Bayer’s own findings
Prinz et al., NRDD, 2011, as summarized in open-access reviews; Begley & Ellis give about 25%
25.4%
total quotation error rate in a pooled analysis of studies of medical journal articles (major errors 11.9%, minor 11.5%; minor includes secondary citations)
Jergas & Baethge, PeerJ, 2015
$28B
per year (estimate) — US spend on preclinical research that is not reproducible
Freedman et al., PLOS Bio, 2015
These studies document reproducibility and citation-accuracy problems across research. We did not find a study among them that applies that evidence to a specific acquisition target’s citation chain before a deal closes, and we found no commercial service that does (our review, April 2026). That is the gap this service addresses.
The body of work
Each page below is a separate piece of the picture. The first case studies are linked below; further cases will be added once checked against primary sources.
Landing
Why biotech's biggest deals close on unverified hypotheses
The core thesis in one page. Problem statement, service description.
Positioning
Where few are looking
A 2x2 map of the diligence firms we reviewed, and the quadrant we found empty.
Case study
AbbVie and Stemcentrx
A $5.8 billion deal and what AbbVie’s filings and the key papers show, with the questions an audit would ask.
Case study
Celgene and mongersen
A $710 million license, a positive Phase 2, a failed Phase 3, and a 2022 batch analysis offering a possible explanation.
Case study
AbbVie and Cerevel
An $8.7 billion deal, a sister deal for $14.0 billion, and two Phase 2 trials that missed 102 days after close.
Case study
The CETP inhibitor class
Four large outcome trials of HDL-raising drugs from four companies: three negative, one modestly positive.
Case study
The amyloid hypothesis
Two antibodies failed in Phase 3, one approval was later discontinued, and two later antibodies slowed decline modestly.
Workflow
The workflow has a missing stage
The eight stages of a biotech acquisition, what each checks, and where citation auditing would fit.
Why humans at every stage miss the citation chain
The errors aren't caused by people doing their job badly. They're caused by structural gaps that no role in the current workflow is built to close.
It isn't anyone's job.
In the diligence workflows we have seen, no role is assigned to backward citation auditing, and each stage tends to assume the prior stage checked it.
Time pressure.
Scientific DD timelines are designed to produce a plausibility judgment, not a lineage audit. Walking citations backward through multiple levels of primary literature is focused work the deal clock doesn't accommodate — and the deliverable expected at the end of the process doesn't require it.
Expertise mismatch.
Senior scientific advisors are domain experts, not citation-forensics experts. Judging whether a mechanism is plausible is a different skill from verifying whether the papers supporting it actually say what they're cited for.
Entity-conflation blind spot.
The same protein name can appear across multiple organisms, glycoforms, and research groups — all treated as the same molecule in downstream slides and citations. Spotting this requires systematic entity matching, not expert intuition.
Cross-area transfer blind spot.
When a mechanism validated in one disease area gets transferred to another, the underlying evidence doesn't always transfer with it. It is rarely checked whether it did.
Forward bias.
A CSO builds the scientific story forward — hypothesis, experiments, results. Auditing requires reading the same story backward — claim, citation, source, and whether the source supports the claim. The two motions aren't interchangeable.
These gaps can compound. Any single deal can survive one of them.
What the service actually does
Six-part deliverable, per audit
Scope: public literature, filings and registries. The audit does not cover non-public company data, IND-enabling studies, manufacturing records or FDA correspondence. A claim is load-bearing when the deal value or development plan would change if it were false.
- Citation chain trace. Every load-bearing scientific claim walked backward to primary sources. Each link tagged supports, partial, contradicts, or missing.
- Entity verification. Protein, gene, and drug identifiers are cross-referenced against public databases such as UniProt and ChEMBL, where the database covers the question, to check for conflation across studies.
- Mechanism verification. Mechanism-of-action claims are checked against the primary binding data in the cited papers.
- Cross-area transfer check. Any mechanism validated in one disease area and applied to another is independently re-verified against primary data in the new context.
- Published-critique surface. Independent peer-reviewed critiques that challenge the foundational assumption are surfaced, with publication date relative to deal close.
- Evidence-chain map. The final deliverable names the specific foundational assumption, the functional validation that does NOT exist for it, and the alternatives available to the acquirer — restructure with milestone payments, walk, or accept the risk with eyes open.