The Financial Stability Board's latest note to the G20 contains four words that carry more weight than the other four thousand: untested in a crisis. I read it twice. Then I did what I always do with a claim that lacks evidence. I opened a terminal, pulled the reserve attestations for the three largest dollar-pegged assets, and went looking for the stress test the regulator says does not exist.
I found the reserve reports. I found the audit letters. I found the redemption terms, the custody arrangements, the yield disclosures. I did not find the test.
This is not an accusation of fraud. It is an accusation of unprovenness. And in a market that has spent a decade confusing liquidity with solvency, that distinction is the entire story. In a world of noise, code is the only quiet truth. So let us read the code.
The regulator is not describing a product. It is describing a hole.
The FSB is not a consumer-protection agency. It is the coordination body of the G20, staffed by central bankers and treasury officials whose mandate is systemic stability โ the prevention of the kind of cascading failure that turns a single institution's problem into a market-wide event. When an institution with that mandate uses the word "untested," it is not making a philosophical point. It is making a technical one.
Here is the technical claim, translated out of diplomatic language. A stablecoin is a bank that has no central bank, no deposit insurance, and no lender of last resort. It issues a liability that promises par redemption on demand, and it backs that promise with a portfolio of assets that are liquid in normal conditions and illiquid in exactly the conditions that trigger redemption. The FSB is saying: we have never watched this structure absorb a genuine, correlated, system-wide run. We have models. We do not have evidence.
The stablecoin market it is describing is not monolithic. Three technical paradigms exist, and they fail differently.
The first is fiat-reserve, or custodial, stablecoins. A centralized issuer holds cash, Treasury bills, and short-duration paper, and mints a token against it. This is the dominant paradigm by float. Its failure mode is a liquidity mismatch: the reserves are safe in the sense that they do not default, and unsafe in the sense that they cannot be sold at par into a panic.
The second is crypto-overcollateralized stablecoins. A user locks volatile collateral and mints a stable liability against it. Its failure mode is a collateral death spiral: as the collateral falls, liquidations accelerate, and the liquidations push the collateral lower.
The third is algorithmic, or seigniorage, stablecoins. These maintain the peg with a second token and reflexive incentives rather than with reserves. This paradigm has a documented failure record. It is the only one of the three that has been tested to destruction in public, and it failed in the most instructive way possible.
The word "untested" hides two specific mechanisms.
When the FSB says stablecoins are untested in a crisis, it is pointing at two distinct failure surfaces, and it is worth separating them because they demand different mitigations.
The first is the liquidity discount on reserves. A Treasury bill is the most liquid instrument in the world on a normal Tuesday. On a stressed Monday, when everyone is selling and no one is buying, the same bill trades at a discount and settles with a delay. The stablecoin issuer that promised par redemption has now made a promise it can only keep by selling into that discount. The risk-free asset is risk-free only for the holder who never has to sell it.
The second is redemption congestion. Redemption is not a single transaction. It is a queue: an on-chain smart contract call, an off-chain compliance check, a banking rail, a settlement window. Under normal load, the queue clears in hours. Under a correlated run โ thousands of holders redeeming in the same block, through the same contract, during a weekend when the banking rail is closed โ the queue becomes a liability in itself. The contract may be sound. The operation around it may not be.
I have audited enough contract logic to know that the failure is rarely in the line you are staring at. It is in the seam between two systems that were each verified in isolation and never verified together. That is what untested means at the code level: the integration path has no test coverage.
Liquidity is the most abused word in this market.
There are three kinds of liquidity, and stablecoins conflate all of them. Accounting liquidity is whether an asset is liquid in a normal market. Market liquidity is whether you can sell it without moving the price โ and this form is pro-cyclical, meaning it evaporates exactly when you need it most. Funding liquidity is whether you can roll your liabilities as they come due. A stablecoin reserve is accounting-liquid by construction. The peg depends on market liquidity that disappears in a panic. And the redemption promise is a funding-liquidity claim that no balance sheet can honor if every holder calls it at once. In a crisis, all three can vanish together, and the token that looked like cash behaves like a claim on a frozen queue.
Sustainability is a math problem, not a narrative.
In 2022, during the liquidity freeze, I did something that felt cold at the time and looks obvious now. I took three of the largest community-driven tokens that had collapsed, reconstructed their emission schedules, and calculated how long their treasuries could fund their burn rates. The answer was under six months for all three. Eighty percent of the tokens in that cohort failed, and they failed for the same reason: they had no utility outside of speculation, and their economics were mathematically unsustainable on a timeline you could compute in advance.
I told my network to hedge sixty percent into stablecoins. That decision was not a prediction. It was arithmetic.
The same arithmetic applies to stablecoins, and the FSB is asking the market to do it. A stablecoin's sustainability is a ratio: liquid reserves divided by the redemption curve under stress. The numerator is disclosed, roughly, in attestation reports. The denominator is not disclosed at all, because it does not exist as a static number. It is a function of market conditions, holder concentration, and the speed of the run. The numerator is a balance sheet. The denominator is a behavior. One is auditable. The other is not. This is the gap the regulator is pointing at. Proof of reserves proves a snapshot. It does not prove a stress path.
The peg is a coordination equilibrium, not a law of nature.
In 2020, during DeFi Summer, I ran an arbitrage between Curve and Uniswap on a pegged asset. Forty-five thousand dollars, executed algorithmically. It worked. But the trade is not the lesson. The lesson is what the trade revealed about the peg itself.
A peg holds because enough participants believe it holds and act on that belief simultaneously. It is a coordination game with a focal point, and focal points are fragile. When the belief breaks, the peg does not decay โ it snaps. The arbitrage I ran was profitable precisely because, for a few blocks, two venues disagreed about the price of something that was supposed to have one price. The disagreement was the signal. The peg was never a fact. It was a consensus that had not yet been tested.
Now scale that fragility to the base of the entire crypto economy. Stablecoins are not one asset among many. They are the settlement layer. Every DEX pool, every lending market, every exchange's margin system, every payment corridor prices against them. A depeg does not stay in one place. It cascades: the lending protocol liquidates collateral priced in the depegged asset, the liquidation dumps more of it onto the DEX, the DEX pool imbalance drags the peg further, the exchange's margin engine marks positions against the broken price and issues more liquidations.
This is not a hypothetical chain. It is a dependency graph, and I have watched smaller versions of it execute in real time. The FSB's mandate โ systemic stability โ is exactly why it cares about stablecoins and not, say, about a single illiquid altcoin. Stablecoins are the load-bearing wall. When a load-bearing wall is untested, the entire building is a claim rather than a structure.
How to read a reserve attestation like an auditor.
Most readers see a reserve report and read one number: the ratio. That is the least informative figure in the document. Three questions matter more.
First, composition and maturity. A reserve that is one hundred percent Treasury bills with a ninety-day ladder is a different risk object than one holding commercial paper with a twelve-month weighted average maturity, even though both are labeled government-adjacent. The maturity ladder tells you how fast the reserves can be turned into cash without a discount. The composition tells you how much of the book is exposed to credit and rollover risk.
Second, custodian concentration. If the reserves sit at one bank or one prime broker, the reserve is exposed to that counterparty's failure โ which is precisely the event that broke a major custodial stablecoin in 2023. Diversification across custodians is not a nicety. It is the difference between a reserve and a single point of failure.
Third, attestation versus audit. An attestation is a point-in-time snapshot by an accounting firm with a limited scope. A full audit is a retrospective opinion on a past period. Neither is a continuous, permissionless proof. The gap between attested and proven is the gap the FSB is describing, and no amount of third-party letterhead closes it.
The weekend problem.
The stablecoin promises redemption twenty-four hours a day, every day of the year. The reserve assets settle on banking time. Treasury bills settle on a one-day cycle. The mismatch between an always-on liability and a business-hours asset is the structural seam of the entire model. A run that begins on a Saturday night has no banking rail to clear against until Monday morning. The queue absorbs the shock in the meantime, and the queue is where trust either holds or breaks. This is not a tail risk. It is a design property, and it has never been stress-tested at systemic scale.
Verification is not the same as trust, and the difference is the entire industry.
I learned the difference in 2017, at twenty, when I found integer overflow vulnerabilities in an implementation of the ERC-20 standard. I did not wait for a patch. I audited fifty thousand lines and submitted a fix. What that experience taught me is the axiom I have built everything on since: decentralized trust is not philosophical. It is mathematical. You do not trust a system because it is well-intentioned. You trust it because you can verify it, line by line, and because the verification does not require permission.
The FSB's warning is, at its core, a verification gap. The regulator is not saying stablecoins are dishonest. It is saying they are unverified โ that the claim of par redemption under stress has never been independently, adversarially tested. This is the same gap that has kept another idea frozen for three years.
Soulbound tokens were supposed to give us portable, non-transferable, on-chain reputation. The concept has existed for three years and has not shipped at scale for a reason that is almost never stated in the papers: no issuer wants its full liability ledger permanently on-chain and auditable by anyone. The stablecoin industry faces the same resistance in a different costume. It will disclose a reserve attestation. It will not disclose a live, permissionless, continuously verifiable reserve feed, because that would surrender the discretion that a fractional-liquidity structure depends on. The resistance to transparency is not a technical limitation. It is an incentive. This is where the FSB's real leverage lives. It can force the denominator into the light.
The interest rate on the reserve is the hidden fault line.
There is a mechanism in this system that almost no one prices, and it is the one I would watch most closely. The issuer of a fiat-reserve stablecoin earns interest on the reserves backing the token. In a high-rate environment, that interest is substantial. The question the FSB's framing forces is simple and uncomfortable: who owns that yield, and who absorbs the loss when the reserves must be sold at a discount?
The current answer is a structural incentive mismatch. The issuer captures the upside of the reserve portfolio. The holder bears the downside of the redemption risk. There is no deposit insurance to socialize the loss and no central bank to backstop the run. The yield is privatized and the tail risk is externalized. That is not a stablecoin-specific flaw. It is the defining feature of every fractional structure that has ever existed, and it is precisely what stress testing is designed to expose.
And this connects to a deeper problem I have written about before: the interest rate models that govern stablecoin lending markets are, in large part, arbitrary. Utilization curves set the borrow rate as a function of how much of a pool is borrowed, with parameters chosen by governance rather than derived from any real measure of supply and demand. In normal conditions, these curves behave plausibly. In a crisis, they behave pathologically. The rate spikes toward its maximum, borrowing becomes impossible, and the incentive to repay collateralizes into an incentive to exit. The untested scenario the FSB worries about is not only a stablecoin run. It is a run colliding with a rate model that was never designed to price a panic.
The stack wars are a distraction from the settlement question.
There is a debate in this industry about which rollup stack will win โ OP Stack against ZK Stack โ and it is conducted as if it were a technical question. It is not. The difference that decides it is which ecosystem can convince more projects to deploy chains first. Distribution beats architecture when the architecture is good enough, and both are good enough.
Stablecoins sit above this debate. They settle wherever the liquidity is, and the liquidity follows the integrations, not the proving system. The FSB's warning is indifferent to which chain a stablecoin chooses. It is not indifferent to whether the stablecoin's redemption path crosses a bridge with its own trust assumptions, or a sequencer with its own liveness assumptions. Every layer a redemption has to traverse is a layer where the stress test has not been run.
Contrarian: The tests already happened. The regulator missed what they proved.
Here is the part of the FSB's framing that deserves pushback.
The claim that stablecoins are untested is, strictly, false. They have been tested twice, in public, with real money. An algorithmic stablecoin was tested to zero, and it took a multi-billion-dollar ecosystem with it. A custodial stablecoin depegged to eighty-seven cents on a bank failure and recovered within days. Both events are data. The regulator's word untested is doing something subtler than describing a vacuum: it is describing the absence of a systemic test. The failures we have seen were idiosyncratic โ one design, one bank, one weekend. The FSB is worried about the correlated version, where every stablecoin faces the same shock at the same time.
And here is the blind spot that the regulator's own remedy could create. If the FSB pushes all compliant stablecoins toward the same reserve composition โ short-dated sovereign paper, held at the same handful of custodians, cleared through the same banking rails โ it will have solved the transparency problem and manufactured a correlation problem. Diversification is safety. Homogenization is the new systemic risk. The safest design, adopted universally, becomes the most fragile structure in the system, because a shock to the shared reserve asset or the shared custodian is now a shock to everyone simultaneously. A regulator that demands a single standard must accept that it is also demanding a single point of failure.
Untested risk is not only the stablecoin's private problem. It is the regulator's design problem, and the design is not yet finished.
Takeaway: The stress test will not be announced.
The test the FSB wants will not arrive as a scheduled exercise with a published report. It will arrive as a redemption wave on a weekend, when the Treasury market is thin, the banks are closed, and the queue is longer than the reserves are liquid. It will not care which chain the token settles on. It will care whether the maturity ladder was real and whether the redemption path was verified end to end.
Watch the maturity ladder. Watch the redemption queue. Watch who owns the yield on the reserve, and who is left holding the loss. The word untested is not a verdict. It is an invitation โ to verify, in advance, the one claim this entire market rests on. In a world of noise, code is the only quiet truth.