Pattern: Fraud exposé

The most legally hazardous and journalistically high-stakes pattern. Done well, it is the most profitable short position ever struck (Hindenburg / Adani, Muddy Waters / Luckin Coffee, Pershing Square / Herbalife, Einhorn / Allied Capital). Done poorly, it is a defamation lawsuit and a blown thesis.

What it is

You allege — with primary-document evidence — that the target's reported financial or operational reality does not match the underlying truth. The "fraud" can be any of:

Accounting fraud — reported financials are materially false (inflated revenue, hidden debt, fabricated cash, capitalised opex)
Operational fraud — the business described to investors does not exist as described (staged demos, fabricated customers, ghost facilities)
Structural fraud — the business model itself is illegal or fundamentally misrepresented (pyramid scheme, undisclosed related-party transactions, off-balance-sheet vehicles hiding losses)
Regulatory fraud — the company is operating in violation of laws in ways management has concealed from the market

Why it works (when it works)

The payoff is binary and enormous. Successful fraud exposés trigger

70% single-day drawdowns, delisting, SEC action, criminal referral.
It neutralises the "long-term compounding" defence. Growth rates mean nothing if the financials are fake.
It travels across markets. Accounting fraud at one company triggers re-rating of peers suspected of similar practices (the "contagion trade").
It attracts regulators. The SEC, FCA, SEBI, BaFin follow high-quality short reports — you're outsourcing enforcement.

The four forensic angles

Every great fraud exposé is built on at least one of these. The best ones stack two or three.

Angle 1: Primary document contradiction

You find government filings, court records, customs data, satellite imagery, or private documents that directly contradict the company's public statements.

Muddy Waters / Sino-Forest (2011): satellite imagery + forestry-bureau records showing plantations didn't exist
Muddy Waters / Luckin Coffee (2020): 25,000+ hours of recorded transaction data from 620 stores vs. reported same-store sales
Hindenburg / Adani (2023): shell-company network documented from Mauritius registry filings
Muddy Waters / NMC Health (2019): bank statements contradicting reported cash

Difficulty: very high — requires primary-source forensic work. Credibility: maximum. Near-impossible to rebut.

Angle 2: Structural / accounting anomaly

You find a quantitative pattern in the reported financials that cannot be explained by legitimate business operations.

Einhorn / Allied Capital (2002): portfolio carried above fair value + deferred charge-offs reclassified as "unrealized depreciation"
Muddy Waters / NQ Mobile (2013): cash accumulation rate that exceeded plausible operating generation
Muddy Waters / SoFi (2026): inflated EBITDA via charge-off manipulation, off-balance-sheet VIEs, and intra-segment depreciation transfers

Difficulty: moderate — forensic accounting skill required. Credibility: high but disputable; management will hire a Big Four firm to issue a rebuttal.

Angle 3: Whistleblower / insider testimony

A former employee, contractor, or related party provides first-hand testimony about misrepresentation.

Hindenburg / Nikola (2020): former employees describing the staged hydrogen demo, the truck that rolled downhill
Scorpion Capital / QuantumScape, Ginkgo Bioworks (2021-2022): engineers contradicting public technical claims

Difficulty: high — requires investigation + legal protection for sources. Credibility: moderate — always subject to "disgruntled ex-employee" attack. Anonymous sources weaker than named.

Angle 4: Business-model critique

You argue the described business cannot generate the reported economics because the unit economics don't compute, the market isn't big enough, or the product does something different than claimed.

Pershing Square / Herbalife (2012): 1.9mm failed Sales Leaders compensation math proving pyramid structure
Citron / Valeant (2015): "research-driven company" claim vs. price-increase-driven revenue forensics
Pershing Square / MBIA (2002–2009): unit economics showing CDO portfolio losses exceeded statutory capital

Difficulty: moderate — analytical, not forensic. Credibility: high but interpretive; company PR will dispute the model assumptions.

The canonical structure of a fraud exposé

Unlike activist decks which follow the 8-block architecture, fraud exposés follow a distinctive 6-block pattern you should internalise:

1.  Bull-case steel-man           (2–3 slides)
2.  The smoking gun               (1–2 slides)
3.  Evidence stack                (20–60 slides — the bulk)
4.  Management response pre-empt  (5–10 slides)
5.  Valuation bridge              (3–5 slides)
6.  Appendix: primary sources     (30–100 slides)

1 · Bull-case steel-man

Open by describing the bull thesis the market currently believes — in its strongest form. This establishes you're not strawmanning.

"Nikola is a category-defining zero-emissions truck maker with a visionary founder, major strategic partnerships, and a working prototype demonstrated in public tests."

Three slides, no editorialisation. Then you destroy it.

2 · The smoking gun

One slide. The single most damning piece of evidence. It must be immediately comprehensible and visually arresting.

A side-by-side photo (Nikola truck rolling downhill vs. the "driving" claim)
A bank statement showing $0 vs. reported $1bn in cash
A government filing in the opposite direction of the company's disclosure
A CEO's Twitter screenshot contradicting SEC testimony

This slide is what the market screenshots. Design for it.

3 · Evidence stack

20–60 slides of forensic detail. Do NOT optimize for narrative flow here — optimize for exhaustiveness and verifiability. Each claim:

Has a source line with archive / filing / timestamp
Can be independently verified by a skeptical reader
Is numbered so readers can cite "Exhibit 47" later
Links to the appendix for primary documents

Muddy Waters reports are the gold standard here. Learn the style.

4 · Management response pre-empt

Before management rebuts, rebut the likely rebuttals. One slide per anticipated defence:

"They will claim X. Here is why X is inconsistent with Exhibit Y."
"They will hire auditor Z to issue a report. Here is auditor Z's history of failed attestations."
"They will call us short-sellers with conflict of interest. We disclose our position transparently — see cover."

5 · Valuation bridge

What the stock is worth if the thesis is right. Often zero for pure fraud theses, but can be a residual ("the legitimate-looking segment might be worth $X; everything else is a zero").

6 · Appendix

Primary documents, time-stamped screenshots, archived web pages, transcripts. The appendix is often twice the length of the thesis. It protects you legally (evidence) and credibly (any skeptic can reconstruct your work).

Language that works — and language to avoid

Use:

"The reported [metric] is inconsistent with [primary document]."
"We allege, based on the evidence in Exhibits [X–Y], that…"
"We estimate the intrinsic equity value at [$X] under the thesis that [specific claim]."
"The filings state [A]. The [primary source] shows [B]. Readers can verify [B] at [link]."

Avoid:

"[Company] lied" — replace with "the filings are inconsistent with [source]"
"[CEO] is a fraudster" — replace with "[CEO] made statements on [date] that cannot be reconciled with [source]"
Emotional adjectives ("shocking", "disgusting", "criminal") — let the facts do the work
Anonymous-only sourcing for any load-bearing claim

Legal guardrails (non-negotiable)

Everything verbatim quoted must have a primary source. Transcript, filing, court document — never a paraphrase.
Distinguish between "allegations" and "claims". You allege fraud; the filings claim accuracy. Use these words carefully.
Disclose your position. Always. At the top of the report. "We are short [ticker] and will profit from a decline" — this is required by Rule 15Fh and also immunises you from market-manipulation claims.
Have counsel review before publication. A 60-page report that loses a defamation case is worse than no report.
Maintain an evidence lockbox. Before publication, deposit all primary documents with counsel in escrow. If litigation comes, you produce the file instantly.
Never reproduce stolen documents. Hacked emails, exfiltrated databases — stay away. Even if they support your thesis, reproduction exposes you to CFAA and trade-secret liability.

Common failure modes

Relying on one anonymous source. Every load-bearing claim needs either a primary document OR multiple named sources. A single "former employee" won't survive PR counter-attack.
Over-reaching. If your evidence supports "the reported margins are inflated", don't claim "the company is a Ponzi scheme". The stronger the claim, the higher the evidentiary bar.
No appendix. Without the ability for skeptics to reconstruct your work, you are asking the market to take your word. Markets don't.
Emotional language in the narrative section. Reserve rhetorical force for the headline and the smoking gun. Everything else is clinical.
Attacking the wrong person. CEOs move on; companies restate; a fraud thesis targeting an outgoing CEO loses its hero.
Publishing before the short position is established. If your cost basis is bad, your credibility is bad. Public reports work when the fund has capacity to absorb cover-induced volatility.

Exemplars

Greenlight · Allied Capital (Jun 2002) — foundational text. Einhorn documented $65M of deferred charge-offs reclassified as "unrealized depreciation", auditor Arthur Andersen's role, and the Business Loan Express off-balance-sheet vehicle. 27 pages of dense forensics. Outcome: SEC investigation, Allied's credit rating downgraded, stock declined 80%+ over subsequent years.
Muddy Waters · Sino-Forest (Jun 2011) — angle 1 archetype. Satellite imagery + forestry-bureau records proved the plantations didn't exist at claimed scale. Stock halted within weeks; bankruptcy.
Pershing Square · Herbalife (Dec 2012) — angle 4 archetype. 343-page "Who wants to be a Millionaire?" deck arguing the compensation math proves pyramid structure. 1.9mm failed Sales Leaders losing $3.8bn since 1980. Triggered FTC investigation → $200M restructuring.
Muddy Waters · NMC Health (Dec 2019) — angle 1. Bank statements contradicting reported cash. Shares halted, administration filed within 3 months.
Muddy Waters · Luckin Coffee (Jan 2020) — 89-page report + 11,000 hours recorded in stores. Same-store sales fabrication. Delisted 5 months later.
Hindenburg · Nikola (Sep 2020) — angle 3 + 4. Staged hydrogen demo, fabricated partnerships, founder misrepresentations. Trevor Milton resigned 10 days later; criminal conviction 2022.
Hindenburg · Adani Group (Jan 2023) — $100bn+ market cap destruction on publication. Shell-company network from Mauritius registry. Ongoing.
Muddy Waters · SoFi (Mar 2026) — angle 2. Inflated EBITDA via charge-off manipulation + off-balance-sheet VIEs + intra-segment transfers. Real EBITDA $103M vs. reported $1,054M.

Full list: examples/by_pattern.json → fraud_exposure

Find anything