Tool · audit

The four-step claim audit

1. IDENTIFY the claim — return, timeframe, strategy

   Pin the claim down until it has three parts: what return, over what period, produced by what mechanism. "Strong performance" is not a claim; "12 percent a year for five years from long-short equity" is. A claim that resists being pinned down has already failed the audit at step one — vagueness is a design choice, not an accident.

2. BENCHMARK against a simple alternative

   Ask what the same money would have done over the same period in the simplest available alternative — a Singapore T-bill, the STI ETF, or an S&P 500 index fund — after all costs. Complexity is a cost, and it must beat boring after fees and after risk or it has not justified itself. Watch for benchmark-shopping: the pitch will offer the flattering ruler, so the auditor chooses the ruler that matches the claim's real risk.

3. DECODE every term into plain English

   Translate each term until nothing technical remains: "absolute return" means no benchmark is offered; "capital-protected" means protected by a specific balance sheet that can itself fail; "enhanced yield" often means upside has been sold away or capital is being returned in instalments. Honest language simplifies under questioning; fog thickens. If a term cannot be translated, treat it as a cost that has not yet been named.

4. IDENTIFY what is NOT being told

   Run the silence audit: the all-in fee as one number, the maximum drawdown, the exact exit terms, and whether the record is live or built after the fact. A pitch is engineered to foreground its best number, so the missing number is usually the decision-relevant one. The absence is the disclosure.

A financial claim is a manufactured object: it has a producer, a cost of production, and an incentive gradient that points away from full disclosure. No conspiracy is required — claims that lead with drawdown convert worse and get made less, so the ones that arrive have survived a filter that rewards gloss. The audit reverses the filter in four steps, and the whole pass costs minutes.

The transparency score

After the four steps, assign a score from 1 to 5. It measures one thing only: how much of the decision-relevant picture the claim volunteered.

• A score of 2 or below is a decline.
• A 3 proceeds only once the gaps are closed in writing.
• A 4 or 5 has earned a deeper look — and nothing more, because transparency measures honesty of disclosure, not quality of investment.

A worked example: the AI-fund email

A hypothetical marketing email arrives: “Our AI-driven fund has returned 24 percent a year. Capacity is limited — reply this week.”

• Step one pins the claim and immediately finds the soft spot: 24 percent a year over what period? The fine print answers — a three-year simulation.
• Step two benchmarks it: the question is what a plain index fund did over those same three years after costs, because in strong equity years a double-digit headline can be a modest edge over boring, purchased with far more risk and fee.
• Step three decodes “AI-driven”: unless the record is live and audited, it means “a backtested model” — the label is a costume, not a mechanism.
• Step four lists the silence: no fee, no drawdown, no exit terms, and a simulated record presented with live confidence.

Score: 2 of 5. The urgency device (“capacity is limited”) is a separate tell: a seller with a durable edge has no reason to rush the buyer.

Eagerness is information about the seller, not the opportunity.

Claims the standard audit under-reads

Three tells matter specifically for crypto and token claims.

• First, the yield source: a yield is real only if it traces to a real payer doing a real thing — interest from actual borrowers, fees from actual trading, rewards backed by actual demand. A yield paid in a token the project can mint at will is emissions, not earnings, and the headline rate is high precisely because the token is being inflated.
• Second, custody: whoever holds the private keys owns the asset, a platform-held coin is a promise from that platform, and transfers are irreversible — “is the money actually there, and who can make it disappear” precedes every return question.
• Third, legal status: a token sold on the promise that the team’s work will raise its price walks through all four prongs of the Howey investment-contract test and may be an unregistered security — meaning the protections registration exists to provide have been quietly forfeited, and a regulator’s calendar becomes a risk no price chart contains.

Two further tells matter for private and alternative assets. A return series with eerily low volatility and low correlation usually reflects appraisal-smoothed marks — valuations nobody can actually transact at, which understate risk exactly when liquidity is needed; ask what fraction of the record is realised transactions and what happened to the assets that failed to sell. And a guarantee is a hidden put: “capital-protected” is only as strong as the balance sheet writing the protection, so the first question about any guarantee is the solvency of the guarantor.

For growth claims, add the durability test — the underwriting question is never how big the market gets, but how durable the competitive advantage is, and pricing power that survives competition is the only observable proof.

Where this breaks

The audit reads the document, not the world. It cannot verify that an audited record was genuinely audited, that reserves exist, or that a custodian is solvent — it tells the reader what to demand in writing and verify independently, and does none of the verifying itself. A perfect score is also not comfort in the other direction: a record with no blemishes, no down months, and no drawdown is more alarming than an honest one with scars, because smoothness can be the fraud.

Treat the audit’s output as a list of demands — the start of due diligence, never the end. Whether to act at all belongs to the ten questions; whether the person presenting the claim deserves a hearing belongs to the advisor red-flag checklist. Run all three, in that order of cheapness.