What an RWA actually is

The definition that can fail a product

"Tokenization of real-world assets" decides nothing. It admits a fund share, a gold bar, a loan book and a bet on a stock price, because all four reference something outside the chain. A definition earns its keep when a product can fail it.

An RWA is a token whose value comes from an enforceable claim against an identified party that controls an asset existing off the chain. Every part of that sentence does work, and the middle part does the most.

An enforceable claim is a right the holder can pursue in court against a named party, which is a different thing from exposure to a price. A BUIDL holder has been a shareholder in a British Virgin Islands fund company that owns Treasury bills since the fund opened on 20 March 2024, and that company is the party the claim runs against. The buyer of a perpetual futures contract on a tokenized share holds a position against a venue and a counterparty, with nothing at the far end of the chain. The two move with the same number on a screen. One of them survives the counterparty.

The word identified carries weight too. A claim against nobody in particular is not a claim, and a product that cannot name the party it owes has answered the question already.

The three conditions: claim, counterparty, redemption

The definition unfolds into three conditions, and a product has to meet all three. An asset outside the chain generates the token's value. An identified party bears the claim. A defined path exists for turning the token back into the asset or into cash.

Redemption is that third path: the process by which the holder hands the token back to the issuer and gets paid. Who may use it is a separate question, and the answer is usually narrower than the product page suggests. Until 22 August 2025 a primary-market client of Mountain Protocol could hand USDM back to the issuer and be paid 1 USD, while holders who had bought the token on the open market were already dependent on whatever liquidity they could find. After that date the primary market closed and what remained of the reserves sat in a Uniswap pool against another stablecoin, which is a market rather than a redemption.

The issuer is the party that sells the token and carries the obligation written into the offering documents, which is rarely the famous name on the product page. BlackRock manages the money inside BUIDL, and the party that owes a BUIDL holder anything is the fund company registered in the British Virgin Islands. USDY, which dates from August 2023, is described as a secured note issued by a bankruptcy-remote entity with a trustee rather than by Ondo Finance in its own name. Paxos is the exception that shows the pattern: Paxos Trust Company issues PAXG under its own name and owes the holder delivery of gold.

BUIDL, PAXG and USDC meet all three conditions. A perpetual on a tokenized stock meets none of them.

Tokenization moves the register, not the asset

The Treasury bill does not go onto a blockchain. It stays where it was, with the custodian, under the same law it was under yesterday. What moves is the list of who owns what.

A custodian is the regulated firm that holds the asset and may not treat it as its own. In BUIDL, which opened on 20 March 2024, that firm is BNY Mellon, and it held securities this way long before any of them were tokenized.

A transfer agent is the licensed firm that keeps the official list of a fund's shareholders and handles subscriptions, redemptions and distributions. For BUIDL that firm is Securitize, registered with the SEC, and the secondary literature on the product describes the register it keeps as running on-chain since the same date rather than in a book-entry system. Both jobs are older than the technology, and chapter 2 covers how they worked before any of this.

So tokenization took one layer out of a working stack and moved it. Everything above that layer and everything below it stayed in place, together with its prices.

Three levels of the same sentence: register, mirror, claim

"The token represents the asset" covers three different legal constructions, and the difference between them is invisible until it matters.

At level A the token is the register. The blockchain is the official shareholder list rather than a copy of it, so the entry in the holder's wallet is the title. Franklin Templeton and the Stellar Development Foundation described FOBXX in April 2026 as the first US-registered investment fund using a public blockchain as its official system of record for transactions and share ownership, which makes it the one level A in this material documented by the issuer instead of inferred from secondary reporting. The fund started on 6 April 2021. Reading the same fund's 485BPOS filing, the author's Polish edition lands somewhere narrower: the prospectus describes a hybrid, with book-entry record keeping alongside public chains under the transfer agent's control, which is not a clean level A. The issuer's announcement and the prospectus describe the same fund and do not agree, so both readings stand here.

At level B the token mirrors a register that lives in a traditional system, synchronized by the issuer. The mirror is a convenient interface and the original is the law: where the two disagree, the original binds.

BUIDL sits at A or at B, and the author's knowledge base records that as an open question rather than an answer. Secondary sources describe a share register kept on-chain instead of in a book-entry system, and only the offering documents and the transfer agent agreement settle which of the two it is. That is the sharpest question anyone can ask about the reference product of this market, and nobody in these sources has read the documents that close it.

At level C the token is a claim on the issuer and gives no title to the underlying asset. Ondo describes USDY in those terms, a debt instrument rather than a share in a fund.

The letters are a reading aid from this series. No regulator classifies products as A, B or C, and no filing uses the terms, so the question to put to a product is the one the documents answer: is this token a share, a debt, or a certificate?

BUIDL's own terms show how much of a product is set by the exemption it is sold under, and how much of what circulates about it comes from aggregators rather than filings.

Bankruptcy is where the levels stop looking alike

In normal operation the levels are indistinguishable. The price is right, redemption settles, and the yield arrives. Everything a dashboard shows looks the same at A, at B and at C.

The rule in the author's knowledge base is short. At levels A and B the holder owns a share in a fund whose assets sit with the fund's own custodian, so the claim runs against segregated fund assets. At level C the holder owns a promise from an operating company, and in that company's bankruptcy the holder joins the queue of general creditors.

USDY shows why that rule is a starting point rather than a verdict. The construction described for it runs through a bankruptcy-remote entity with a trustee, which separates the collateral from Ondo's operating estate and leaves the holder better placed than the holder of an ordinary corporate bond and worse placed than a shareholder in a fund. That description comes from an aggregator, Coinpaprika, in August 2026, not from an offering document. Two products can hold the same Treasury bills and hand their buyers different outcomes on the day the issuer goes bankrupt, and the paperwork is the only place the difference is written down.

The market has three sizes, and they differ by an order of magnitude

Distributed assets are tokens that can be moved to a wallet outside the issuing platform and sent between wallets. Represented assets cannot, whether by design or by regulation, and exist to put a record on a chain rather than to put an asset in an investor's hands. RWA.xyz published this split on 21 November 2025, made distributed the default view of its own dashboard, and warned at the time that headline numbers would look smaller as assets moved into the represented column.

In 2019 the entire security-token industry traded 2,410,608 USD on the secondary market for the year, an average of 7,156 USD a day. A direct reading of RWA.xyz on 10 September 2026 put distributed value at 39.15 bn USD. Between the start of 2025 and August 2026 that number grew roughly six- to sevenfold on the author's own readings, the later of which is relayed through an aggregator rather than taken from the source. Growth of that shape is not a reclassification artifact, because the November 2025 change cut the reported figures rather than inflating them.

On a reading relayed by an aggregator, and one that a research pass on 10 September 2026 could not reproduce from either RWA.xyz or DefiLlama, most of the represented category sits on a single network. That network is Canton, a permissioned institutional network where a retail investor has no account and is not going to get one, and its own aggregate figures describe activity inside it rather than a market anyone can trade. This chapter quotes no value for its share.

Stablecoins meet all three conditions and are counted separately anyway. They are roughly eight times the size of everything else in the category, and the Treasury bills inside their reserves would otherwise be counted twice, once in the stablecoin and once in the tokenized fund that holds them. RWA.xyz and McKinsey's forecasts both keep them apart, and so does this series.

The author's own model arranges these sizes as four circles, represented outermost, then distributed, then what changes hands, then what is lent against in DeFi, with about 2,300 to 1 between the outermost and the innermost. Under RWA.xyz's definitions the first two circles are disjoint rather than nested, since an asset is classified as one or the other. Read the circles as sizes on one scale and the ratio holds, and every argument about this market sits somewhere inside it.

The failure mode that actually happens

What matters about RWA failures is the shape they share, which a list of dead projects hides. Five things can end a product: the asset defaults, the smart contract is hacked, the issuer goes bankrupt, the custodian goes bankrupt, or the issuer decides to stop. The second lane is empty in the record behind this series, and the fifth is the one that keeps happening.

USDM is the cleanest case. From 22 August 2025 the token was no longer backed by Treasury bills and no longer redeemable at 1 USD from the issuer, and its backing became USDC sitting in a Uniswap pool. The bills were fine the whole way through. What disappeared was the product, not the asset. About 10,820 addresses held the token on 12 May 2025, the day the acquisition was announced and minting was switched off, and most of them belonged to holders who were not primary-market clients, so their way out ran through whatever a decentralized exchange would pay. No legal structure protects against this, because nothing broke: the issuer used a right it held from the beginning.

Harbor's failure runs the other way and teaches as much. The company tokenized a student housing block, Hub at Columbia: 260 units, a 20 mn USD deal, split into 955 tokens of 21,000 USD each. It collapsed on 11 April 2019, when the mortgage lender refused to permit the change of ownership structure. The cause was not a regulator, not a shortage of investors and not the contract, and what the investors recovered is not recorded in the sources behind this chapter. A legal link three levels above the token vetoed the deal, and that link is still in place in every deal shaped like this one.

Both of those breaks happened off the chain, at the legal wrapper and at the issuer. That is the practical value of the chain-of-claims picture: it puts the blockchain in its actual position, sixth of nine as this series counts the stops, and puts the attention on the two ends where products have come apart.