A peg is a mechanism, not a property

"The stablecoin holds a peg" is a claim about a mechanism. It is not a property of the token, in the way that a metal's density or a bond's seniority is a property of the instrument. Nothing in USDC, USDT, DAI or crvUSD guarantees the dollar peg. Each one assembles a set of machines that have historically produced it, and each machine has a clock.

The useful question is not whether a stablecoin is pegged. It is four questions, and each issuer's own documentation answers them differently.

One: what actually backs the token

Start with the asset, because "backed" hides a spread of instruments that behave nothing alike under stress.

Tether's reserves are the widest of the four, and its own filing says so. In the BDO Italia report on the Consolidated Reserves Report as of 30 September 2023, management asserted that "The Group's consolidated total assets amount to at least US$ 86,384,653,832," against "The Group's consolidated total liabilities amount to US$ 83,176,997,409 of which US$ 83,153,363,663 relates to digital tokens issued."

The report breaks that asset base into six lines, and the interesting ones are not the Treasury bills. US Treasury bills sit at US$56,611,569,049. But the same table carries US$3,149,722,956 in precious metals, US$1,661,948,724 in bitcoin, US$2,257,992,725 in "Other Investments" and US$5,168,269,890 in secured loans. Cash and short-term deposits total US$74,058,148,982, which is 85.7% of the group's total consolidated assets, a denominator that also includes corporate bonds rather than only assets backing the token.

That last line is the one to watch. A secured loan is a receivable. It is a promise by a borrower, valued at amortised cost in this report, and it settles when the borrower pays. It is not a bank deposit, and it does not have deposit insurance. Tether's transparency page states the point plainly in its own FAQ: "Every Tether token is 100% backed by Reserves, which includes traditional currency, cash equivalents, and other assets, including receivables from loans made by Tether to third parties."

The BDO report also states what kind of assurance this is. "We carried out our work in accordance with the criteria established in the International Standard on Assurance Engagements 3000 (Revised)", and it draws the boundary explicitly: "Our opinion is limited solely to the CRR and the corresponding consolidated total assets and consolidated total liabilities as of 30 September 2023. Activity prior to and after this time and date was not considered when testing the balances and information described above." Its own valuation policy note is the sentence a risk reader should keep: "The valuation of the assets of the Group is based on normal trading conditions and does not reflect unexpected and extraordinary market conditions, or the case of key custodians or counterparties experiencing substantial illiquidity, which may result in delayed realisable values."

Read that last clause as the definition of a depeg window. A point-in-time assertion, produced under normal trading conditions, that explicitly excludes the conditions in which the peg is actually tested.

Circle's composition is narrower, and Circle's filings let you read it security by security. Circle's Transparency page describes the structure: "Treasuries and overnight reverse repurchase agreements (Overnight Reverse Treasury Repo) held as reserves for the benefit of USDC holders may be held by the Issuer in custodial accounts, as part of a separately managed account, or within the Circle Reserve Fund, a 2a-7 money market fund managed by BlackRock." The page also states the cadence of verification: "USDC reserve holdings are fully disclosed on a weekly basis, along with associated mint/burn flows. Additionally, a Big Four accounting firm provides monthly third-party assurance that the value of USDC reserves are greater than the amount of USDC in circulation."

The monthly report for August 2026 is a document worth actually opening rather than summarising. Its USDC Reserve Report gives two dates, 11 August and 31 August 2026 at 11:59pm Coordinated Universal Time. As of 31 August 2026, USDC in circulation was 73,320,757,305 and "Fair Value of Assets Held in USDC Reserve" was $73,398,532,613. The gap is US$77,775,308, which is 0.11% of circulation. The excess is real, and at this size it is a rounding-scale buffer rather than a shock absorber.

Composition at that date, read directly off the report's own subtotals: the Circle Reserve Fund held US$62,317,374,425, or 84.9% of total USDC reserve assets. Inside the fund, US Treasury repurchase agreements were US$49,584,000,000 (67.6% of the whole reserve) and US Treasuries US$11,918,146,195 (16.2%). Outside the fund, "Cash held at regulated financial institutions" was US$9,591,096,978, or 13.1% of the reserve, with US$1,301,292,779 in Treasuries (1.8%). Circle Reserve Fund assets total US$73,398,532,613 across both dates' structure, and "OTHER USDC RESERVE ASSETS" total US$11,081,158,188.

Circle's reserve report also names the accountant's standard. The report is an examination under AICPA attestation standards: "Our examination was conducted in accordance with attestation standards established by the American Institute of Certified Public Accountants (AICPA)." That is a reasonable assurance engagement on a point-in-time assertion, not a full audit of financial statements, and Circle is explicit about who audits what: "Independently audited Deloitte & Touche LLP is Circle's independent auditor and has audited Circle's financials since fiscal 2022. Prior to Deloitte, Grant Thornton LLP served as Circle's independent auditor from 2015."

One detail in Circle's report is quietly load-bearing. Of "Cash held at regulated financial institutions" at 31 August 2026, the reserve funds are described as "unencumbered accounts held on behalf of USDC holders that are segregated from other accounts of Circle, including general corporate funds." Segregation means corporate creditors do not reach the reserve. It does not mean the depositor bank cannot fail. Those are two different risks, and conflating them is how readers get surprised.

Two: who can redeem, and at what speed

This is where most peg claims stop being mechanisms and start being marketing.

USDC's own Terms define two different holders. "Users with a Circle Mint account are referred to herein as 'User Type A.'" Everyone else, "referred to herein as 'User Type B'," and: "For the avoidance of doubt, Users Type B are not customers of Circle, as Users Type B do not have a Circle Mint account." The redemption right is explicit and gated: "The following only applies to User Type B: You may not redeem USDC with Circle unless and until you open a Circle Mint account." The Terms then make the commitment itself: "Circle (or an affiliate designated by Circle) commits to redeem 1 USDC for 1 USD, subject to these Terms, applicable law, and any fees where applicable." And the gate is repeated: "Only Users Type A can redeem USDC directly with Circle."

The USDC Services, including redemption, are listed as one of three things Circle makes available "in accordance with the Circle Mint account User Agreement," and Circle's legal page notes that "Circle Mint and the Services are currently only available to institutions located in supported jurisdictions."

Tether's gate is even more explicit because it is priced. From the Relevant Information Document dated 20 February 2026, submitted to El Salvador's CNAD: "The current minimum redemption amount of Tether Tokens through the Tether Website is Tether Tokens with a value of US$100,000. Redemptions of Tether Tokens are subject to a fee of the greater of US$1,000 or 0.1% of the Tether Tokens to be redeemed." The fees page lists the same threshold: "Minimum Tether Token acquisition or redemption amount 100,000 USD" and "Fee per redemption The greater of $1,000 or 0.1%".

The US$100,000 floor means USDT's 1:1 claim is a contractual right held by a subset of holders, not by the holder sitting in a liquidity pool. The same document says so about scope: "Holders of Tether Tokens are not entitled to any increases in value of the Reserves in excess of the face value of the Tether Tokens (less fees)." And it reserves discretion in a way that is worth reading closely, because a redemption window is exactly this clause: "Tether may delay or suspend redemption under various circumstances, in accordance with its Terms," including "when Tether is directed to do so by any government (including any court, law enforcement agency, tribunal, or regulatory body)" and "if a KYC Verified Customer's account or wallet is subject to pending litigation, investigation, or government proceedings."

Circle's gate is an account and a jurisdiction rather than a ticket size, but it resolves to the same thing: a redemption queue served by a bank, during bank hours. Circle said so plainly during the only serious depeg test its reserve has faced. In the 11 March 2023 post "An Update on USDC and Silicon Valley Bank," the summary reads: "While USDC can be used 24/7/365 on chain, issuance and redemption is constrained by the working hours of the U.S. banking system." That sentence is the mechanism in full. The token trades continuously; the primary market that mints and retires it runs on a weekday schedule.

Three: what the oracle reads when the market is thin

A fiat-backed token's peg is enforced by arbitrage against a redemption right. A CDP token's peg is enforced by liquidation arithmetic against a price feed. Those are different oracles and they fail in different ways.

Sky's documentation on Dai describes the first mechanism as a contract, not a hope. "The Maker Protocol uses contracts called PSMs (Peg Stability Modules) that allow Dai to be directly redeemed for other cryptocurrencies pegged to the U.S. dollar." The mechanics are unconditional for anyone: "At any time, any user can receive minted Dai in exchange for these cryptocurrencies. If a Peg Stability Module contains sufficient stock, any user can destroy Dai in exchange for receiving these cryptocurrencies."

The second word in that sentence is the mechanism, and it is also the boundary. "If a Peg Stability Module contains sufficient stock." A PSM is a swap against a stock of a whitelisted stablecoin. When that stock runs low, the 1:1 swap stops, and DAI's short-term peg defence becomes a statement about inventory rather than a mechanism. The documentation itself ranks it by horizon: "This mechanism is the most effective at maintaining the Dai peg in the short term." Stability fees are ranked elsewhere: "This mechanism is most effective at maintaining the Dai peg in the medium/long term."

Curve builds crvUSD on the second mechanism and documents the oracle explicitly. The design note: "Curve incorporates specialized on-chain Exponential Moving Average (EMA) oracles built into Stableswap-NG, Tricrypto-NG, and Twocrypto-NG Curve pool implementations. crvUSD markets derive price information from a select number of high TVL Curve pools." The smoothing is the manipulation defence, and Curve states its limits: "Despite the manipulation-resistant design specification, Curve pool oracles may exhibit price distortions in certain scenarios that have the potential to result in missed or excessive liquidations. This may be a result of liquidity and volume migration to alternate venues that increase the risk of oracle manipulation."

The specific hazard the article should name is thin liquidity moving the reading venue, not a broken feed. The oracle reads Curve's own pools. If volume migrates elsewhere, the pools that remain are thinner than the volume they are being asked to price. The oracle degrades exactly when it is needed most, and the window is set by how long the EMA smooths.

Curve then bolts an external bound onto the internal oracle, and the design of the bound is the lesson. "The oracle contracts have the option to utilize Chainlink prices, which serve as safety limits. When enabled, these limits are triggered if the Chainlink price deviates by more than 1.5% (represented by BOUND_SIZE) from the internal price oracles." Note "option" and "when enabled." Curve's documentation continues: "Chainlink limits can be turned on and off by calling set_use_chainlink(do_it: bool), which can only be done by the admin of the Factory contract." So the external check is admin-toggleable, and Curve's own developer forum carries a proposal titled "Disable Chainlink limits for price oracles in crvUSD," which argues that when the market is very volatile Chainlink-based limits "can cause more losses than necessary."

Then Curve documents that crvUSD's own peg machinery depends on other stablecoins staying stable, which is the cleanest admission of recursive peg risk anywhere in this set of documents. From the crvUSD risk disclaimer: "crvUSD makes use of a Peg Stabilization Reserve (PSR) which consists of contracts authorized to deposit and withdraw crvUSD from a whitelisted Curve crvUSD Stableswappool up to a predefined debt cap. These contracts reference a subset of whitelisted stablecoins as a proxy for the intended USD price. Instability affecting any counterparty Reserve assets (e.g. USDT, USDC), which are also used to aggregate a USD price for crvUSD, may cause the Reserve to deposit all of its crvUSD into the pool in an attempt to rebalance. This creates a dependency on the Reserve counterparty assets that determines the stability of the crvUSD peg."

So crvUSD's defence against a $1.00 breach is a pool denominated in USDT and USDC. A depeg in one of those is a depeg in the machinery built to prevent one.

Four: the redemption window is the whole ballgame

The March 2023 USDC event is the cleanest documented example available, and it is documented by the issuer rather than by a commentator.

The trigger was a bank. Circle disclosed on 11 March 2023 that "USDC is currently collateralized 77% ($32.4B) with US Treasury Bills (with a three month or less maturation period), and 23% ($9.7B) with cash held at a variety of institutions, of which SVB is only one." Within that cash, "$3.3bn of USDC's cash reserves remain with SVB." Circle also stated that "$1bn of the USDC reserves is held with Customers Bank" and that it "maintains transaction and settlement accounts for USDC with Signature Bank."

The depeg was not a reserve shortfall. Circle's account of what followed is unambiguous: "However, it is also possible that SVB may not return 100% and that any return might take some time, as the FDIC issues IOUs (i.e., receivership certificates) and advanced dividends to deposit holders. In such a case Circle, as required by law under stored-value money transmission regulation, will stand behind USDC and cover any shortfall using corporate resources, involving external capital if necessary."

What broke was the timing. USDC traded on a blockchain continuously through a weekend in which the primary market was closed, and the redemption queue could not clear. Circle's own 15 March update, published 16 March, records the resolution: "As of close of U.S. banking operations Wednesday, March 15, we have cleared substantially all of the backlog of minting and redemption requests for USDC. Since Monday morning, Circle has redeemed $3.8 billion USDC and minted $0.8 billion USDC."

Three facts from that sequence are the whole argument of this article. The reserve was intact. The redemption path existed and was contractual. And for approximately sixty hours, the token's market price was not a statement about backing, because the arbitrageur who enforces the 1:1 relationship could not reach the mechanism that enforces it. The 12 March press release put the resolution in one line: "The $3.3B USDC reserve deposit held at Silicon Valley Bank, about 8% of the USDC total reserve, will be fully available when U.S. banks open tomorrow morning."

Note what fixed it: not a new reserve policy, not a new attestation, not an oracle improvement. A deposit guarantee from a government regulator. The peg was restored by the banking system's opening hours and the FDIC's guarantee, which is to say by the redemption window becoming short enough to trust again.

That is the test to apply to any stablecoin, including the ones that have not been tested. Ask four questions and read the answers in the issuer's or protocol's own documents, with the dates attached:

What backs it. If the answer includes receivables, loans, or volatile assets, count them separately from cash and short-dated government paper. Tether's September 2023 report had US$5,168,269,890 in secured loans and US$1,661,948,724 in bitcoin inside a reserve whose assertion was dated to a single day.

Who can redeem. USDC: only User Type A, with a Circle Mint account. USDT: only KYC Verified Customers, at US$100,000 minimum, paying the greater of US$1,000 or 0.1%.

How fast. Circle's own words, 11 March 2023: "While USDC can be utilized 24/7/365 on-chain, issuance and redemption is constrained by the working hours of the U.S. banking system." For a CDP token, the equivalent question is whether the redemption stock exists, because "If a Peg Stability Module contains sufficient stock."

What the oracle reads in a thin market. For crvUSD, EMA prices from a "select number of high TVL Curve pools," with an external 1.5% Chainlink bound that Curve's own docs say can be turned off by the Factory admin.

A peg that cannot be enforced at the speed of the panic is not a peg. It is a queue.

The wider point

Stablecoin design is not a solved problem with a few remaining rough edges. It is a set of arrangements between four clocks that rarely run together: the clock of the reserve's legal claim on its assets, the clock of the banking system that serves redemption, the clock of the oracle that prices collateral, and the clock of the market that repricing the token in seconds.

The March 2023 USDC depeg is instructive because it was, by the issuer's own account, a failure of timing and not of backing. That is the most common shape a stablecoin break takes, and it is also the shape that a reserve attestation cannot see. An attestation is a photograph. A peg is a clock. The two are different objects, and the gap between them is where the risk lives.

Risk note: cryptocurrency trading, leveraged perpetual futures, and automated algorithmic strategies carry significant risk of rapid and total financial loss. Never risk funds you cannot afford to lose completely. A stablecoin's dollar peg is the intended output of a mechanism with redemption gates, banking-hours dependency, oracle assumptions and issuer discretion, not a guarantee; a depeg can and has occurred within hours, and reserve attestations are point-in-time assertions produced under the conditions described in the reports cited above. Nothing in this article is investment advice, a recommendation, or an offer to sell any product.