Product Spec

Forensic Audit — What Is Actually in the Audit

04-23-2026

Extracted from ten completed cases, not invented. This document answers the question "what do you actually do in a forensic audit?" by describing the methodology as it was executed — including the parts still codified as working practice rather than formal protocol.

What it is, in one sentence

An independent forensic audit of a biotech target's scientific evidence chain — tracing every citation link behind the mechanism-of-action claim backward to primary sources, classifying each link's confidence, and flagging where standard validation is missing or where downstream claims inflate what the cited papers actually say.

The deliverable — what a completed audit looks like

Every audit produces one artifact with six parts, in this order:

Claim identified. One sentence stating the mechanism-of-action claim the deal, trial, or investment rests on.
Backward citation chain. Start at the clinical or investor-facing claim. Walk backward through each cited paper until you reach primary biology. Typically five to seven links.
Per-link analysis. For each link: what the paper actually says; what downstream papers or press releases claim it says; the gap, if any; confidence level (HIGH / MEDIUM / LOW).
Critical questions. The two to four basic domain facts that would invalidate the composite claim if any were false.
Verdict structure. What holds firmly. What holds with caveats. What does not yet hold. No single-word rating — the audit refuses to collapse to pass or fail because real chains are mixed.
Prediction with kill conditions. A confidence-weighted forward-looking call plus the specific evidence that would disconfirm it.

This structure has been executed on ten targets. See the case-study list at the bottom.

Inputs required from the client

The target's most recent investor deck, press releases, and Phase 1/2 or pivotal-trial publications.
Access to the target's cited references — public literature only. No clean room. No non-public data.
One or two internal scientific personnel for clarification questions (optional but accelerates turnaround).

That is the full input set. The audit does not require access to CMC, clinical operations, or non-public data rooms.

Process — how an audit is performed

Step	What happens	Time
1	Claim extraction — read investor-facing material, isolate the load-bearing mechanism claim	0.5 day
2	Reference harvest — pull every paper cited in support of that claim, plus any review article it links through	0.5 day
3	Backward trace — read each cited paper; note what it actually demonstrates (population, assay, n, endpoint)	2–4 days
4	Gap analysis — compare what each paper says against what downstream papers claim it says; surface inflations, substitutions, missing validations	1–2 days
5	Critical-question check — test composite claim against basic domain facts that, if true, would invalidate it	0.5–1 day
6	Verdict plus writeup — composite judgment, confidence levels, kill conditions, final artifact	1 day

Turnaround: 5 to 10 business days depending on chain length and field. The five-day version is a flash audit with a shorter per-link writeup. The ten-day version is the full forensic artifact suitable for an M&A diligence file.

Outputs delivered

Written audit in Markdown and HTML, ~2,000 to 5,000 words depending on chain length.
Full citation list with confidence tags per link.
One-page executive summary with verdict, kill conditions, and recommended deal actions — proceed, restructure, or walk.
Optional: 30-minute debrief call with the deal team.

Quality standards (what is actually enforced)

Every link traced to primary source. No "per secondary review" citations. If the primary paper cannot be accessed or does not exist, the link is flagged as UNVERIFIED and that status survives into the final deliverable.
Domain-boundary verification. When a claim crosses disciplines — for example, patent law plus clinical timeline equals commercial conclusion — confidence decreases at the boundary. Cross-domain conclusions require explicit verification, not inference.
Conditions preserved. "Works in acute injury" never becomes "works." "n = 8, hypothesis-generating" never becomes "demonstrated."
Verdict refuses binary compression. An audit that can only say pass or fail is doing something other than forensic work. Every verdict carries "holds / holds with caveats / does not hold" structure.
Kill conditions named. Every audit ends with the specific evidence that would disconfirm its verdict. This is what makes the audit falsifiable rather than a narrative.

Not yet codified

Formal inter-reviewer protocol (currently a single-reviewer workflow with machine-assisted citation retrieval); turnaround SLA in a signed engagement letter; legally-defensible adjudication standard (the methodology is consistent but not yet tested as a contract-grade adjudication artifact, which is a requirement if this is ever sold as insurance adjudication infrastructure).

Scope

What it does

Trace citation chains. Flag inflations, gaps, missing validations, domain-boundary crossings. Assign per-link and composite confidence. Name kill conditions. Deliver an acquirer-usable artifact.

What it does NOT do

Run wet-lab experiments. Replicate assays. Audit CMC, manufacturing, or clinical operations. Audit financials or patent validity. Recommend a price. Issue legal opinions.

Pricing anchor — working hypothesis, not committed

Flash audit (5 days, ~2,000 words): anchor range $25K–$50K.
Full audit (10 days, ~5,000 words): anchor range $75K–$125K.

These are directional anchors. Actual pricing depends on the buyer side (acquirer-side M&A diligence, seller-side pre-market audit, or insurance-side adjudication) and needs market validation — scientific diligence is not priced like financial Quality of Earnings and no published benchmark exists in this category.

Case studies executed

Ten traces completed as of 04-23-2026.

#	Target	Claim traced	Verdict
1	AbbVie / Stemcentrx (Rova-T)	DLL3-targeted cancer stem cell model applied to small-cell lung cancer	Missing validation; deal should have been restructured or walked
2	Vera Therapeutics (atacicept in IgAN)	Dual APRIL/BLyS inhibition reduces pathogenic IgA in IgAN	Well-supported; lupus "failure" was combination-therapy safety event, not mechanism failure
3	Corvus (soquelitinib in PTCL)	ITK inhibition preferentially depletes malignant T cells	Weakly supported; chain holds on mechanism, breaks on "preferentially"
4	MindMed (MM120 in GAD)	5-HT2A partial agonism resets pathological fMRI connectivity	Weakly supported; mechanism composite built from three disconnected evidence bases
5	Ocugen (OCU400 in retinitis pigmentosa)	Nuclear-receptor modulator rescues photoreceptor function	See trace
6	EyePoint (Duravyu in wet AMD)	Sustained-release tyrosine kinase inhibitor non-inferior to anti-VEGF	See trace
7	Celgene / GED-0301	SMAD7 antisense oligonucleotide in Crohn's disease	Chain broke at primary target-engagement evidence; Phase 3 failed
8	CETP inhibitors (multiple sponsors)	Raising HDL via CETP inhibition reduces cardiovascular events	Chain broke on HDL-as-surrogate assumption; multiple $1B+ failures
9	Alzheimer amyloid hypothesis (decades)	Amyloid plaque reduction slows cognitive decline	Chain inflated at surrogate-to-clinical translation; $2B+ in failed programs
10	Cerevel / emraclidine (schizophrenia)	M4 muscarinic agonism non-inferior to broad-spectrum antipsychotics	See trace

→ View the full citation-chain audit for AbbVie / Stemcentrx

What is codified versus current practice

Codified

Six-part artifact structure
Per-link analysis template (what-it-says / what-downstream-claims / gap / confidence)
Verdict refusal-of-binary-compression rule
Kill-condition requirement

Current practice, not yet codified

Inter-reviewer protocol
Time-to-turnaround SLA
Adjudication-grade legal standard
Formal pricing tier definitions

All four uncodified items are on the short list to formalize before any engagement letter is signed with a paying client outside the warm-channel introducer.