The eleven-step sequence this piece is built on is not mine. It was laid out by @_jamico on X, from deposit to total loss.
I am not going to re-walk it. Restating a good argument in more words is not an extension of it. What follows is three things the sequence invites and does not do: an overlay of what has already been observed at each step, a measurement of the gap the whole structure rests on, and an extrapolation of where this lands if offchain valuations are never independently attested.
The short version of the extension is that the hypothetical has a name, a manager, and a live deployment; the thing at the center of it can be sized using a control group that already exists; and the direction of travel is toward more of this rather than less, for reasons that are structural rather than speculative.
I maintain a public registry of reserve attestation reports for stablecoins and tokenized funds. It holds 187 records, of which 125 are verified against a report I can link to directly. The remainder fail verification mostly because the report does not exist, cannot be reached, or turns out to be a self-published disclosure rather than a practitioner's report. That distinction is the subject here.
The architecture stopped being hypothetical
In April 2025, Securitize and Gauntlet brought Apollo's Diversified Credit Fund onchain as ACRED and wired it into a levered strategy, as CoinDesk and Markets Media reported at the time. The mechanics are the outline's second step, executed properly: a vault deposits ACRED as collateral on Compound Blue, a Morpho-powered market on Polygon, borrows USDC against it, buys more ACRED, and repeats, with the loop bounded by Gauntlet's optimization engine. Unchained's coverage of the deployment was headlined "DeFi Looping Comes to Apollo's $1.3 Billion Credit Fund: What Could Go Wrong?", which is the correct question asked early.
The underlying is not a Treasury bill. Apollo's Diversified Credit Fund invests across corporate direct lending, asset-backed lending, and dislocated and structured credit. The collateral in that loop is a claim on illiquid private corporate debt.
Notice what this is not. It is not a scandal, an exploit, or an anonymous team. Every party is named and reputable: one of the world's largest private credit managers, a registered transfer agent in Securitize, a quantitative risk manager in Gauntlet, an established oracle provider in RedStone, a widely used lending protocol in Morpho. The disclosure is public. Nobody is hiding anything.
That is the finding, not an aside. The sequence does not require bad actors. It is what you get when competent parties each do their own job correctly and no party owns the question of whether the price is true.
The oracle has a name now, and the name is the concession
The outline calls the price feed a "Verified NAV Oracle" and treats it as the load-bearing component. That was generous. The real thing is more honest and more revealing.
Securitize selected RedStone as its primary oracle partner in March 2025, and the two piloted a primitive they called the Trusted Single Source Oracle, or TSSO, described as designed specifically for real-world assets whose valuation is inherently single-source. For ACRED, the NAV data is obtained directly from the fund administrator and pushed onchain on a daily heartbeat, per RedStone's own case study.
Read the name again. Trusted. Single. Source. The engineering is candid in a way the marketing around it usually is not: it tells you there is one origin for the number and that the system's job is to carry it faithfully. That is a real property and it is worth having. Tamper-evident delivery from a fund administrator on a daily cadence is a genuine improvement over a manager tweeting a figure.
It is also not verification, and nobody involved claims it is. The oracle transports. The transfer agent records. The curator sizes the position. The protocol enforces the threshold. The verification question gets handed one step upstream at every layer until it reaches the fund administrator, and then it stops, because an administrator computes NAV from inputs the manager supplies. Administrators reconcile, they calculate, and for illiquid positions they take the manager's valuation as an input rather than independently testing it.
So the chain that establishes what your collateral is worth begins and ends inside the party whose solvency is the question. Every component is excellent. There is no component whose job is to disagree.
The industry has already conceded the liquidation problem
The strongest evidence that step ten is real is that the same oracle provider shipped a product to address it.
RedStone launched Settle on 28 April 2026, an on-demand liquidation settlement layer built specifically for tokenized RWAs used as collateral in lending protocols. Its stated premise is that RWA liquidation does not work: DEX pools are too shallow, compliance restrictions eliminate permissionless liquidation, and redemption windows of 60 to 180 days make forced exits impossible.
That is the ninth and tenth steps of the outline, published as product rationale by a vendor with every commercial reason to be optimistic about this asset class. When the infrastructure providers are building workarounds for the failure mode, the failure mode is not a thought experiment.
The control group
Here is the part the outline gestures at without measuring, and it is measurable.
Tokenized private credit is not the only place illiquid loan books get marked by their managers. Business development companies hold substantially the same asset class, value it on the same quarterly cadence with the same model-driven methods, and then trade on public exchanges. The spread between a BDC's share price and its reported net asset value is therefore a continuously updated, market-implied estimate of how wrong the manager's mark is.
That estimate is currently large. BDCs trade at nearly a 20% discount to net asset value, with some large names approaching 50%. VanEck reports the MVIS US Business Development Companies Index at roughly 0.83x price-to-book as of 27 February 2026, against a long-term average near 0.97x. Mercer Capital published an analysis in 2026 titled "Public Prices, Private Marks: What BDC Discounts Are Signaling", which is the question stated plainly.
Now put that number next to the protocol parameters.
Morpho Blue computes the liquidation incentive factor from the market's LLTV. The formula appears in a comment above the code that implements it in Morpho.sol: the minimum of the maximum incentive factor and 1 divided by (1 minus cursor times (1 minus LLTV)). The constants are set in ConstantsLib.sol at a maximum of 1.15 and a cursor of 0.3. At an LLTV of 91.5%, 1 divided by (1 minus 0.3 times 0.085) equals 1.0262. A 2.62% bonus. A position sitting at 90% LTV against that threshold liquidates on a 1.64% decline in collateral value.
So: the public market's estimate of the error in private credit marks is roughly 20%. The distance to liquidation is 1.64%. The entire economic budget available to whoever steps in is 2.62%. The suspected error is about twelve times the trigger distance and about eight times the incentive.
Two honest caveats, because the argument does not need overstatement. A BDC discount is not a pure measurement of mark error. It also carries liquidity preference, vehicle leverage, rate expectations, tax treatment, and plain sentiment, and closed-end structures trade away from NAV for reasons that have nothing to do with valuation accuracy. And ACRED is a feeder into a diversified fund rather than a single BDC, so the composition differs. Cut the 20% by three quarters to strip out everything that is not mark error and you still have 5%, which is three times the distance to liquidation and roughly double the liquidator's entire budget.
This is the sentence the whole structure turns on. The liquidation incentive is the market's budget for verification, and at 91.5% LLTV that budget is 2.62% of the position, against a suspected valuation error an order of magnitude larger. The protocol is not mispricing risk through carelessness. It is offering a bounty calibrated for collateral that can be valued in a block and applying it to collateral where valuation is a quarterly exercise that the public market discounts by a fifth.
The inputs are deteriorating on a published schedule
The valuation gap would be academic if the underlying credit were stable. It is not.
Proskauer's Private Credit Default Index, which tracked 716 loans representing $195.6 billion of original principal in its most recent reading, recorded a default rate of 2.51% for Q2 2026, down from 2.73% in Q1 2026, which was itself up from 2.46% in Q4 2025 and 1.84% in Q3 2025. The rate rose by roughly half across three quarters and then eased.
Around the same period, Moody's noted that private credit inflows shifted to the first-ever quarterly outflow in Q1 2026, that publicly traded BDCs have maximized leverage, and that as of March 2026 the average default-probability-implied risk for public BDCs exceeded that of Baa-rated public corporates by the widest margin since post-pandemic normalization, with payment-in-kind income rising as a share of investment income. PIMCO has written on what BDC redemptions and NAV pressure signal for the asset class.
Payment-in-kind income deserves its own line. A borrower paying interest in more debt rather than in cash is a borrower whose cash position is deteriorating while its carrying value keeps compounding upward. That is the outline's fourth step stated as an accounting policy, and it shows up in the NAV as growth.
The IMF made the general case in its April 2024 Global Financial Stability Report, naming fragile borrowers, layered leverage, and stale and potentially subjective valuations, and observing that stale valuations create a first-mover advantage. Based on historical price-to-NAV patterns it takes at least four quarters for price and NAV to converge after a shock, and the gap between the most optimistic and most conservative manager's mark on the same loan exceeds five percentage points.
Note what that last figure means in this context. Five percentage points of legitimate disagreement between two honest managers, with no fraud and no distress, is already double the liquidator's budget.
Why the exposure grows rather than shrinks
The natural assumption after Stream Finance was that this would self-correct. The data says the opposite, and the reasons are structural.
Regulation is channeling yield demand into these instruments. The GENIUS Act prohibits permitted payment stablecoin issuers from paying interest or yield to holders based solely on holding the token, in any form. The demand for a digital dollar that earns something did not evaporate; it relocated to products that are legally free to pay. An Oxford Law Blogs analysis published in July 2026 framed this precisely as regulatory arbitrage by legal form, with tokenized Treasury funds, DeFi lending protocols, and offshore issuers positioned to absorb it. The Congressional Research Service has its own treatment of the stablecoin yield debate. The effect of the cleanest piece of stablecoin legislation to date is to push yield-seeking dollars out of the most heavily attested instrument class in crypto and into instruments priced by single-source NAV.
The collateral base is shifting toward these assets while crypto-native activity softens. Tokenized RWA deposits in DeFi grew from $2.3 billion in Q2 2025 to $7.4 billion in Q2 2026, a 200% year-over-year increase, per CoinShares. That is not just growth; it is a compositional change in what backs onchain credit. The Block's research desk has been tracking RWAs as collateral as a distinct primitive.
The allocation decision is concentrating in a small number of curators. Curated vaults now sit between depositors and markets, and a recent paper on institutionalizing risk curation in decentralized credit treats the role as an emerging financial institution rather than a technical convenience. Industry surveys of the vault landscape put Gauntlet alone at roughly $1.5 billion of curated TVL, with the largest curators controlling most curated assets. Concentration of the allocation decision is the same shape as concentration of the valuation decision, one layer down.
The distribution end now terminates in consumer products. Coinbase's USDC lending routes customer deposits through a Morpho vault curated by Steakhouse Financial, as Morpho and CoinDesk both describe. That is a legitimate, disclosed structure. It is also a demonstration of how few steps now separate an exchange balance from a permissionless market whose collateral is priced by a single upstream source.
The Financial Stability Board flagged the general pattern in its October 2024 report on the financial stability implications of tokenisation, identifying vulnerabilities in the underlying reference asset and noting that functions kept separate in traditional finance, such as issuance, custody, and secondary trading, become blurred and intermingled in tokenized systems.
The overlay
Each step of the sequence, and where it has already been observed.
Deposit into an onchain credit facility for a yield-bearing token. ACRED for the institutional version, xUSD and deUSD for the version that failed. All three are claims on offchain books priced by their sponsors.
A high-LLTV market that enables a loop. Live and disclosed in the ACRED strategy on Compound Blue. Present in its uncontrolled form at Stream, where onchain researchers traced the same 1.9 million USDC used to mint roughly 14.5 million xUSD, an amplification of about 7.6 times, and found roughly $170 million of onchain collateral against roughly $530 million borrowed.
An oracle relaying the issuer's own NAV. TSSO, daily, from the fund administrator. Named as single-source by the people who built it.
Borrowers deteriorating before the number moves. The Proskauer index rising by roughly half across three quarters while marks are struck quarterly, and PIK income rising as a share of investment income.
A markdown that liquidates on a small move. 1.64% at 91.5% LLTV. Q1 2026 brought NAV declines and redemption pressure across the BDC complex.
A run met from the cash sleeve first. The first-ever quarterly outflow in private credit in Q1 2026, and the IMF's first-mover advantage operating at redemption-window speed in traditional vehicles and at block speed onchain.
Redemptions pause; the secondary trades far below NAV. BDCs at a 20% discount are the orderly version. deUSD at roughly $0.015, down about 98%, and Stables Labs' USDX below $0.60 are the disorderly one.
Suppliers withdraw and rates spike. Roughly $1 billion left DeFi yield products within a week of the Stream disclosure.
The liquidator discovers the token may confer no usable claim. RedStone Settle's own premise: compliance restrictions eliminate permissionless liquidation and redemption windows of 60 to 180 days make forced exits impossible.
The bonus is too small to attract anyone. 2.62%, computed from the deployed contracts.
Retail loses without ever taking a credit decision. Curators allocated depositor funds into xUSD without depositors understanding the exposure, and public warnings circulated before the failure without changing allocations. The Coinbase and Steakhouse structure shows the same distribution chain operating correctly.
Two precedents predate all of it. Orthogonal Trading defaulted on $36 million across eight loans on Maple in December 2022, roughly 30% of active loans, weeks after telling lenders its FTX exposure was about $2.5 million; investors in the affected pool faced an 80% loss on remaining capital. Goldfinch accumulated more than $18 million of cumulative losses, including a $1.9 million diversion by one borrower in breach of its agreement and a facility where the borrower repaid $4.25 million of $10.15 million, and has since shut its credit platform down. Stream Finance announced a $93 million loss in November 2025 and engaged Perkins Coie; xUSD fell 77%, Elixir had lent roughly 65% of deUSD reserves to Stream, and researchers traced roughly $285 million of interconnected debt. Rekt and The Defiant walk the mechanics.
In every case the onchain accounting was current, precise, and wrong.
Where this goes if the mark is never attested
This is the extrapolation, and it is a claim about mechanism rather than a date.
The next failure will not look like Stream. Stream was fast, small, opaque, and adjacent to fraud, which made it legible and easy to dismiss as someone else's bad behavior. The structure now being built is slow, large, transparent, and staffed by people with reputations to protect. The failure mode that follows is a legitimate quarterly markdown at a real manager, disclosed on schedule, that happens to be four percentage points larger than the buffer beneath the leverage stacked on it. Nobody will have done anything wrong at any step.
Liquidation will not clear; the loss will sit. This is the specific prediction and it follows from arithmetic rather than sentiment. At a 2.62% bonus against collateral requiring an offchain workout, there is no bid. Positions do not get resolved at a bad price; they fail to get resolved at all. Bad debt accrues to the market, which means it accrues to the suppliers, which means it accrues to whichever curated vault was allocating there, which means it accrues to depositors who selected a yield product. The mechanism that is supposed to transfer the loss to a risk-taker instead distributes it to people who thought they were lending.
The first credible signal will be a pause, not a disclosure. A BDC holder learns about deterioration from a quarterly filing and, imperfectly and continuously, from a market price. A tokenized feeder holder has neither. There is no examiner who can demand the file and no auditor positioned to qualify an opinion on the mark that the oracle is transporting. So the first hard information most holders receive about a bad valuation will be the redemption queue stopping. Suspension will be doing the work that disclosure does everywhere else, and it will arrive after the exit has closed.
The response will be restriction rather than verification, which caps the market. The available levers after an incident are permissioned markets, whitelists, lower LLTVs, longer redemption notice, and institutional-only access. Every one of those shrinks the blast radius without answering whether the number is right. That is a rational response and it is also a ceiling: an asset class that can only be made safe by restricting who touches it does not reach the multi-trillion-dollar projections attached to it. The forecasts assume tokenized private credit becomes usable collateral in open markets. Usable collateral requires a price that a counterparty who distrusts the issuer can rely on. Without attested valuation, the growth path terminates in a walled garden that is functionally the existing private credit market with faster settlement.
And the gap widens before it closes. Onchain settlement is instant and getting faster. Offchain valuation is quarterly and, per the IMF, takes at least four quarters to converge after a shock. Every improvement to settlement, collateral mobility, and automated leverage increases the ratio between how fast the position can move and how fast the truth about it can. The oracle's daily heartbeat does not close that gap; it reports a quarterly judgment more often.
What attested valuation would actually require
It is worth being precise about the thing that does not exist, because "audited" and "verified" are doing enormous unearned work in this market.
An examination engagement has five parts. Management writes an assertion, typically that as of a stated date the fair value of assets held was equal to or greater than some obligation. The assertion is measured against stated criteria. A practitioner tests it against evidence: confirmations, holdings records, valuation support. The practitioner expresses an opinion in writing. The practitioner's name and their firm's liability stand behind that opinion. A NAV relay has none of these. There is no assertion, no criteria, no evidence testing, no practitioner, and no opinion. There is a number and a signature on the number's delivery.
I built the Reserve Attestation Registry partly because that gap is invisible from the outside, where a monthly examination under AT-C 205 by a PCAOB-registered firm and a quarterly agreed-upon-procedures report expressing no opinion at all are both reported as "attestations."
Traditional funds solved a version of this and the solution is worth naming, because it shows what is missing rather than what is impossible. The SEC adopted Rule 2a-5 on 3 December 2020, with a compliance date of 8 September 2022. It permits a fund board to designate a valuation designee, which is the adviser, while requiring the board to actively oversee it: assess valuation risk periodically, test the fairness of methodologies, oversee pricing services, scrutinize the information received, manage the designee's conflicts of interest, and receive quarterly reporting, with recordkeeping behind all of it. The SEC's compliance guide sets out the mechanics.
That is not a technology. It is a party with standing, a duty to look, and consequences for not looking. A tokenized offshore feeder wrapped around a fund share carries the settlement rail and leaves that apparatus behind. The delta is the entire subject.
Applied here, attested offchain valuation would mean at minimum: an assertion by the manager about the fair value of the book as of a date; a declared scope, so that a position left out is a finding rather than an absence; testing by a practitioner independent of the manager, with an opinion and liability attached; and a cadence matched to the leverage rather than to the fund's reporting calendar, since a daily-liquidatable position priced on a quarterly judgment is not a pricing problem but a maturity mismatch in the evidence.
The principles I have been writing up in the Attested Execution Framework apply without modification, which was not the plan when I wrote them for machine execution. A control is not evidence until it has been observed enforcing, and the NAV oracle has never rejected a NAV and has no mechanism to. Verification must not depend on the operator, and here every link passes through the party whose solvency is the question. Scope must be declared, and absence within it is a finding, which is why an undeclared NAV boundary makes an omitted deteriorating position undetectable rather than merely unreported. An untested backup or attestation is a claim rather than a control, and a redemption channel that has never absorbed a run is a hypothesis about liquidity.
One limit, stated plainly so the recommendation is not oversold. A valuation attestation would not make illiquid credit liquid, would not make a 91.5% LLTV prudent, would not create a legal claim where the documents do not, and would not have prevented Stream, which was a disclosure and custody failure before it was a valuation one. What it does is convert an unpriceable unknown into a priceable one. A market cannot charge for a risk it cannot see. Today the mark carries no signal about its own reliability, so every participant is forced to assume the best case, and the assumption is capitalized at up to 11.8 times.
The number at the center
Every layer of this stack is now genuinely good. Settlement is instant. Custody is institutional. The transfer agent is registered. The oracle is tamper-evident and honest about being single-source. The curator is quantitative. The manager is one of the largest in the world. The disclosure is public.
The number at the center is still produced by one interested party and independently tested by nobody, and every one of those excellent layers is a machine for making that number matter more.
The market built the transport and skipped the assertion. Until someone with standing can compel production and put an opinion behind the answer, the most sophisticated infrastructure in onchain finance is carrying an unverified figure very quickly, very accurately, to more places.
The eleven-step sequence is @_jamico's; the extension, arithmetic, and any errors are mine. Other sources are linked inline. Morpho's liquidation arithmetic is computed from the formula and constants in Morpho Blue's deployed source rather than from a secondary description of them. Figures on Stream Finance, Elixir, Maple, and Goldfinch come from contemporaneous reporting cited above. Valuation, default, and discount figures come from the IMF, Proskauer, Moody's, VanEck, and the cited market commentary, and carry those sources' definitions and limitations. This is a personal research piece. It is informational only, is not investment advice, is not an assurance opinion, and is not affiliated with or endorsed by any employer. Nothing here asserts wrongdoing by any named firm; the argument is about structure, not conduct.