Hook

Seven information points. Zero numbers. That is the complete data payload of the BNB Chain "AI native currency" briefing, and the emptiness is the story.

The setting matters more than the sentence. The sentence was delivered at a Nasdaq bell-ringing — a venue built for listing ceremonies, not protocol announcements. The speaker was Changpeng Zhao. The claim: crypto assets will become the native currency of artificial intelligence.

Then he named the candidates. Bitcoin. Existing stablecoins. Tokens issued by AI companies themselves.

Three candidates. The chain whose ecosystem he was ostensibly promoting was not among them.

Read that again as a balance sheet. On the asset side, a thesis about machine payments. On the liability side, an omission. A founder appearing at a traditional capital markets venue to argue that machines will need money — and the money he lists excludes the token his own venue sits adjacent to.

That is a data anomaly in the strict sense: an expected value that failed to print. My first rule when a series breaks is to check whether the break is an error or a signal. Here it reads as a signal, because the omission is internally consistent with everything else in the packet — a packet that contains no token mechanism, no standard proposal, no counterparty, no date, no capital commitment.

So this is not a product story. It is a positioning story. Positioning stories are auditable, if you know which ledger to open.

Context

Methodology first, protocol second.

I grade source packets before I grade claims. This one is a single second-hand retelling of a single media briefing, with seven discrete assertions and no primary transcript. That places it at source grade D on my internal scale: directional signal, not evidentiary record. Everything downstream inherits that discount. Any figure below is either labeled illustrative arithmetic or labeled background knowledge. Nothing else qualifies.

Three of the seven points are load-bearing. Point two: the native-currency claim. Point four: the hope that BNB Chain becomes the primary network, with the concession that it "still needs further development." Point five: a compute-linked token that may be brought onto BNB Chain, details withheld.

The other four are texture. A Nasdaq appearance. A candidate list. An admission that crypto payments remain inconvenient for ordinary users. A recommendation to focus on remittance corridors and underserved markets rather than competing head-on with consumer payment apps in developed economies.

Those four texture points are worth more than the three load-bearing ones. I will argue that in the back half of this piece, and I will show the arithmetic.

Background, for readers arriving from outside the ecosystem. BNB Chain is not one chain. It is a stack: BSC as the EVM-compatible execution layer, opBNB as an optimistic rollup, Greenfield as a storage layer. Consensus on BSC is Proof of Staked Authority — a small validator set elected through staking. Throughput is adequate. Fees are low. Finality is fast. Decentralization is thinner than the label "L1" implies, and that is a design tradeoff rather than a scandal.

BNB the token carries a burn architecture. Quarterly Auto-Burn, plus BEP-95, which destroys a portion of gas fees in real time. The intended trajectory has been public for years: a reduction from an original 200 million toward 100 million. I am not going to restate the parameters from memory as if they were freshly verified. Check the official ledger. That is the only defensible convention.

So the substrate exists. Cheap execution, an EVM toolchain, a large retail user base concentrated in emerging markets, and a distribution funnel through the largest centralized exchange in the world. That is a real asset. It is not a technical moat. It is a distribution moat, and the two are priced differently by people who understand both.

The competitive frame, stated plainly. Ethereum holds the standards conversation for agent payments — account abstraction, session keys, intent architectures, and the various ERC efforts around agent identity. Solana holds the narrative density for AI-adjacent crypto and DePIN, with the throughput profile to match. Dedicated AI chains hold narrative purity and little else. BNB Chain holds neither the standard nor the purity. It holds the on-ramp.

That is the correct lens for the entire packet. A distribution-first chain making a standards-first claim. Mismatch is expected. The question is whether the mismatch is fatal or cosmetic.

Core

A Term Without a Falsification Test

"AI native currency" fails the first test I apply to any technical claim: can I write down what would prove it false?

If the phrase means AI agents will denominate invoices in a specific asset, it is testable — you look at invoicing data. If it means agents will hold balances on-chain, testable — you cluster wallets. If it means AI systems require a blockchain to function, it is a category error, because most AI systems settle in databases and always will.

The briefing supports none of these. It supports a mood.

Compare with how a real standard claim reads. A real one names a specification, a version, a reference implementation, a test suite, and a migration path. "ERC-4337 changes how accounts sign" is a claim you can audit. "Crypto will be AI's native currency" is a claim you can only repeat.

This matters for one practical reason. Non-falsifiable claims cannot be priced. Markets price distributions, not adjectives. When a narrative has no testable predicate, the only thing that trades is attention — and attention in this sector has a half-life measured in days, occasionally hours.

I have watched this pattern from the inside. In 2018 I spent six weeks tracing the Zcash shielded transaction consensus rules. What made those findings actionable was not rhetoric. It was that a proof either verified or it did not, and the verification was reproducible by anyone holding the circuit description. Three implementation flaws in the zero-knowledge proof path, filed through GitHub, patched in two weeks. The patch was possible because the claim was falsifiable.

Nothing in this packet admits a patch, because nothing in it admits a test.

Confidence: high. And that high confidence is itself the output — an unverifiable claim is not a weak claim, it is a non-claim with a media format.

The Candidate List Is the Finding

Most coverage reads the three-candidate list as an inclusive gesture. It is the opposite. It is a specification.

Consider what each candidate implies about machine settlement.

Bitcoin: a high-variance, hard-capped bearer asset with slow finality and no native programmability. Excellent as a reserve. Poor as a medium for a machine pricing a contract term shorter than a block time.

Stablecoins: low-variance claims on fiat liabilities, transferable at par, with existing regulatory wrappers in most major jurisdictions. Designed, whether intentionally or not, for exactly the use case described.

Tokens issued by AI companies: equity-like instruments with a payment function bolted on. The most legally exposed of the three, and the one whose economics most resemble a security.

Now run the variance test. A machine-to-machine contract is short. An inference job might run four hundred milliseconds. A GPU rental might run six hours. The shorter the contract, the more the counterparty cares about nominal variance, because there is no time for a position to mean-revert.

This is not ideology. It is arithmetic. If an agent quotes a price in an asset with four percent daily realized volatility, it must add a spread to cover the settlement window. That spread is a tax. Multiply it across millions of micro-settlements and the tax dominates the transaction.

Stablecoins win the denomination question before any chain is chosen. That is not a controversial claim in 2026. It is the observed behavior of every serious agent-payment pilot I have reviewed, and it is the conclusion my own 2024 institutional-flow work kept producing from a different direction: the capital that behaves institutionally is the capital that refuses to take nominal risk on the settlement leg.

Which means the briefing's headline thesis, taken seriously, argues against its own venue. If stablecoins are the native currency of AI, the strategic question is not which chain. The question is which dollar rail, and who clears it. The chain becomes a transport detail.

That is the first real fracture line. The phrase "crypto assets" is doing enormous hiding work. It bundles a bearer reserve asset, a fiat claim, and a corporate equity instrument into one noun, then lets the audience assume the noun means the token they already hold.

It does not. The noun means: not the token you already hold.

Compute Tokens — The Oracle Problem Wearing a Hardware Mask

Point five is the only technically interesting item in the packet, and the only one with no detail.

A compute-linked token means tokenizing the usage rights or the revenue rights of GPU capacity. That sits in the DePIN category. The consensus layer is not the hard part. BSC can mint a token in an afternoon.

The hard part is verification of the physical asset. Three problems, in ascending order of unpleasantness.

Start with existence. Does the GPU exist? Not does the operator say it exists — does an independent party have a hardware attestation chain, a serial number registry, a datacenter power-draw record, and a plausible colocation contract? A token claiming to be backed by H100 capacity and actually backed by a spreadsheet is not a DePIN asset. It is an unsecured note with a logo.

Then liveness. Is it running? Idle GPUs generate no revenue. Utilization claims must reconcile against billing records from actual workloads. The reconciliation surface — job identifiers, timestamps, customer identities — is where most DePIN projects quietly stop publishing, and the quietness is the tell.

And the worst of the three: double-pledge. The same physical unit can be pledged to three lenders if the lenders do not share a registry. This is a classic failure in commodity finance — the same warehouse receipt for the same rice, sold four times. Crypto does not solve it. Crypto accelerates it, because pledging is instant and auditing is quarterly.

This is an oracle problem, and I spent the last year inside a very specific version of it. In my 2026 work on AI-agent data integrity, roughly 30 percent of agent-driven trading errors traced back to manipulated or stale oracle inputs. We built a standardized verification protocol — zero-knowledge proofs validating feeds before agent execution — and deployed it across three lending protocols, which cut oracle-related losses by about 45 percent. That worked because the inputs were digital and the validity condition was expressible in a circuit.

Physical compute is harder. The validity condition involves a machine that a proof cannot inspect. You need the cryptographic layer and the legal layer and a physical audit regime, and the chain contributes nothing to any of the three.

So when a briefing says "compute-linked token, details withheld," the honest translation is: the hardest part is unresolved and possibly not resolvable at the current stage. Details are not withheld because they are secret. Details are withheld because they do not exist yet in publishable form.

The Transmission Coefficient Nobody Calculates

Here is where I part company with most of my peers. They argue about whether the narrative is bullish. I want to know the diameter of the pipe.

The claimed transmission path for BNB runs like this: narrative attracts developers, developers deploy, deployment generates activity, activity generates gas, gas burns, burn reduces supply, reduced supply supports price.

Six links. Assign each a generous probability and multiply.

Narrative converting to developer deployment: 0.25. Most narratives do not convert. The base rate for "announced direction becomes shipped product" in this industry is brutal, and the 2022 cycle taught an entire cohort what it costs to assume conversion.

Deployment converting to meaningful activity: 0.30. Most deployed contracts go unused. The long tail of BSC contracts with zero lifetime calls is the empirical proof, and it is public.

Activity converting to gas: high, call it 0.90, because gas is mechanical.

Gas converting to net burn: this is where the chain collapses. BEP-95 burns a fraction of fees, but the fee pool is shared with validators, and the marginal contribution of one new application category to a global fee pool measured in tens of millions of dollars per quarter is a rounding error against the denominator.

Burn converting to price support: not a mechanical relationship at all. It is a flow argument whose magnitude depends on burn size relative to secondary market volume.

Multiply 0.25 by 0.30 by 0.90 and you land near 0.07 before touching the burn. Seven percent, and that is the optimistic branch.

Now the denominator. What is the cost structure of an AI compute business? GPU depreciation, power, cooling, networking, datacenter rent, engineering, sales. Gas is not on that list. Gas is not one basis point of that list. A tokenized GPU cloud processing a million dollars of compute billing might generate a few hundred dollars of chain fees, and a fraction of those get burned.

The transmission coefficient from "AI narrative" to "BNB supply reduction" is not merely small. It is asymptotic to zero.

This is not a criticism of BNB specifically. It is a structural property of every general-purpose L1 trying to capture value from an activity whose cost base lives off-chain. The chain is a receipt printer. Receipts are cheap, and receipt printers do not capture the margin of the business they serve.

The Granularity Floor

Here is the insight I have not seen stated plainly in any coverage of this briefing, and it reframes the whole debate.

Machine payments have a granularity problem, and the problem is set by the fee floor, not the fee level.

Consider inference. A single small-model inference call in 2026 prices somewhere below a cent, and falling. An embedding lookup is cheaper still. An agent orchestrating a task might make four hundred billable calls in a minute.

Now attach a settlement fee to each one. Even on the cheapest general-purpose chain, a transaction costs a nontrivial fraction of a cent once you include the real cost of getting it included — priority, retry, nonce management, key custody overhead, the operational surface of failover.

Multiply by four hundred.

Settling each call on-chain costs more than the call. Not slightly more. Categorically more, by orders of magnitude.

This kills the naive version of the thesis. "AI agents paying each other on-chain in micro-increments" is not a chain-scaling problem. It is a clearing problem. The architecture that survives contact with arithmetic looks like this: agents maintain credit relationships off-chain, accumulate net positions, and settle periodically — hourly, daily, whatever the risk tolerance permits — through a netting layer that touches the chain a handful of times per cycle.

That is a clearing house. It has been the correct answer for payment systems since the eighteenth century, and the blockchain does not change the underlying economics. It changes the settlement leg.

Once you accept the clearing-house frame, the competitive question inverts. The winning chain is not the fastest one. It is the one with the best netting primitives, the deepest stablecoin liquidity, the most credible escrow and dispute resolution, and the lowest cost of final settlement. Block time becomes nearly irrelevant, because the chain is touched rarely and the netting layer is where the engineering lives.

Every gas fee tells a story of intent, and the story here is that agents intend to settle net, not gross. Anyone building gross settlement for machine payments is building a bridge with one end in the air.

Confidence: high on the structural argument, medium on the specific fee magnitudes, which readers should re-derive from live data rather than trust from a briefing about a briefing.

Standards Vacuum

Points two and four share a defect. Neither names a specification.

An agent payment standard is not one thing. It is at least five things, and all five are in contested development across the industry right now.

Payment authorization is the first. An agent needs a mandate: maximum spend per period, permitted counterparties, revocation semantics, delegation depth. Session keys and scoped allowances approximate this today. Nothing in the packet references it.

Machine identity is the second. An agent needs a verifiable identifier that is not a wallet address, because an address says nothing about who the agent acts for. Decentralized identifiers and attestation frameworks exist. Nothing in the packet references them.

Verifiable inference is the third. If an agent pays for a computation, it needs to know the computation happened and happened correctly. Proof-of-inference is an open research problem with partial commercial solutions. Nothing in the packet references it.

Machine-readable receipts and reconciliation are the fourth, and this is the one that actually gates enterprise adoption. An agent's accounting system needs structured settlement records that reconcile against invoices automatically. It sounds tedious. It is the reason no treasury function has yet signed off on an agent-operated payment rail, because the audit trail currently terminates in a block explorer.

Dispute resolution is the fifth and the least discussed. When an agent pays for a service that was not delivered, what happens? Escrow, arbitration, insurance, slashing. Nothing in the packet.

Five missing specifications. The briefing offered a sentiment and a venue.

To be fair, a bell-ringing is not a standards committee, and demanding a spec from a photo opportunity is unreasonable. But that is exactly the point: the event was a photo opportunity, and the coverage treated it as a directional statement about technology. The mismatch between format and interpretation is the actual error, and it is an error made by readers, not by the speaker.

Code does not lie, only developers do — and the corollary is that absence of code is not a neutral observation. It is a datum with a sign.

The Cheapest Chain Wins the Micro-Payment and Loses the Custody

BNB Chain's genuine advantage here is cost. Its genuine disadvantage is trust minimization.

The PoSA validator set is small. That is a deliberate engineering choice buying fast finality and low fees, which is precisely what high-frequency small-value settlement needs. For an agent moving three cents, you do not need a thousand validators. You need cheap and fast and predictable.

But the same architecture disqualifies the opposite use case. An agent custodying institutional balances — a treasury agent rebalancing a nine-figure book — needs the property that no small coalition can rewrite its state. PoSA does not offer that property at the same level as a large, geographically dispersed proof-of-stake network with a costly attack surface.

So the addressable market splits cleanly. BNB Chain is well positioned for the retail-facing, high-frequency, small-ticket half of machine payments. It is poorly positioned for the institutional custody half, and the custody half is where the fees are, because custody is priced on risk aversion and risk aversion is priced rich.

This is the same fragmentation pathology I have written about at the rollup layer, recurring one level down. Dozens of execution environments compete for the same finite pool of actual users and actual capital. That is not scaling. That is slicing an already thin liquidity base into thinner slices and presenting the arithmetic as a roadmap.

Liquidity is the current of truth. It flows to depth, depth accumulates where users already are, and users are not where block times are shortest. They are where the on-ramp is and where the counterparties are and where the exit is.

The Bell as a Balance Sheet Event

The most under-analyzed element of the packet is the venue.

A Nasdaq bell-ringing is not decorative. It is a capital-markets ritual. It signals that an entity has listed, or is listing, or wishes to be read as listing-adjacent. Combined with persistent ecosystem references to an entity styled BNB Standard, the natural inference is a Digital Asset Treasury structure — a listed vehicle whose core asset strategy is holding BNB.

Flagging this as inference, not fact. I have not seen the filings. But the structure is familiar and the mechanics are well understood, and I have spent enough time on pre-mortems to know that familiar structures fail in repetitive ways.

A DAT has a reflexivity problem. Its equity trades at a premium or discount to net asset value. When the underlying token rises, NAV rises, the premium often expands, the entity issues shares accretively, buys more of the token, and the loop reinforces. When the token falls, the loop inverts: NAV falls, premium flips to discount, accretive issuance becomes dilutive issuance, financing dries up, and any leverage in the structure begins to force sales.

That is a negative feedback path connecting a crypto order book to a public equity order book. Once it exists, the volatility is no longer contained inside the asset class.

Why does this matter for a narrative briefing? Because if such an entity exists and the speaker holds economic exposure to it, then a public statement about the token's future utility function has a second audience: equity holders and prospective equity holders. The statement is not solely a technological opinion. Functionally, it is marketing for a levered wrapper.

No allegation of misconduct. An observation about an incentive gradient, which is the minimum standard of care before reading a promotional statement as analysis.

Standardization survives the chaos of collapse — and the standard that will matter here is disclosure. If the entity exists and files, the filings are the primary source and the briefing is noise. If the entity does not file, the briefing is the only source, and that is a materially worse epistemic position for everyone downstream.

Howey, Applied to Compute

The four-part test is not complicated. Money invested. A common enterprise. An expectation of profit. Profits derived from the efforts of others.

Run a compute-linked token through it.

Money invested: yes, by construction. Purchasers exchange capital for the token.

Common enterprise: yes where the token represents a share of a pooled compute fleet whose economics depend on aggregate utilization and market pricing for GPU hours.

Expectation of profit: this is where design intent decides the outcome. A pure utility token consumed and burned on use carries a weaker profit narrative. A token distributing rental income carries an unambiguous one.

Efforts of others: decisive. If the holder's return depends on a team procuring GPUs, negotiating power contracts, signing customers, and managing utilization, then the holder is passive and the return is derivative of managerial effort. That is the classic investment contract shape, and it is the shape most compute tokens take because that is the only way the business is operable.

The honest verdict on a revenue-sharing compute token is that it is presumptively a security in the United States and plausibly so in several other major jurisdictions. Not certainly. Presumptively. And the presumption is rebuttable only by structure — a legal opinion, a jurisdictional choice, genuinely consumptive token rights, a distribution that stays inside an exemption, and a lockup schedule that does not look like an underwriting.

The briefing's phrase "details withheld" sits exactly on top of this. The most legally consequential attribute of the asset is the one attribute we were told nothing about.

Second-order point, and it deserves its own paragraph. The briefing also endorsed the idea of AI companies issuing their own tokens. That is not a neutral observation. It is encouragement for non-crypto-native firms to raise capital through token issuance, and if that capital is raised from US persons, the securities analysis is not ambiguous. It is a registration question with a well-documented enforcement history.

Distribution Beats Standardization, Until It Doesn't

Let me steelman the bullish read, because it deserves one and because a one-sided audit is a bad audit.

BNB Chain's actual superpower is converting narrative into tradeable assets faster than anyone. Hand it a category and it will produce fifty tokens in that category within a month, seed liquidity, list several on the largest exchange in the world, and generate the price discovery that makes the category legible to the broader market. That is a genuine function. It is how categories get born, and sneering at it is the habit of people who have never had to bootstrap a market.

For compute tokens specifically, the match is strong. A compute token is an issuance product. It needs a launch venue, a liquidity venue, and a retail audience with appetite for duration risk. BNB Chain has all three, better than any competitor.

So the bullish case is not that BNB Chain will define the AI payment standard. The bullish case is that BNB Chain will be where compute assets get issued and traded, and that issuance volume is a real business with real fees.

Legitimate business. Also a business with a cyclical dependency on the narrative it feeds. Issuance revenue peaks at the top of the cycle and disappears at the bottom, which is the exact moment you need it.

And the standard-setting business — the one generating durable rents, the one where the specification becomes the default and the default becomes the toll — goes to whoever writes the documents. Historically that is Ethereum, occasionally Solana, and it is a slow game measured in years of committee work and reference implementations, not quarters of bell-ringings.

Efficiency is the only permanent alpha, and efficiency in issue-and-distribute is a different species from efficiency in define-and-standardize. Confusing the two is how ecosystems overestimate their own durability right before a cycle turns.

The One Falsifiable Line in the Packet

At the end of the briefing, two sentences everyone skimmed.

One: crypto payments remain inconvenient for ordinary users. Two: focus on remittance corridors and underserved markets rather than trying to displace consumer payment apps head-on in developed economies.

These are self-limiting statements. In a promotional venue, self-limiting statements are rare and therefore informationally dense. A speaker optimizing purely for narrative would not concede inconvenience, and would not concede that the largest consumer payment markets are out of reach.

So I weight those two sentences above the headline. They describe a business.

Cross-border remittance is a market with a measurable inefficiency: corridor costs running from low single digits to double digits as a percentage of principal, with the worst corridors concentrated in exactly the regions where BNB Chain's user base lives. Stablecoin rails compress the settlement leg of that cost dramatically. What remains is the fiat on-ramp and off-ramp, which is a licensing and partnership problem, not a blockchain problem, and licensing problems are solvable with lawyers and time rather than with throughput.

A chain hosting deep stablecoin liquidity, cheap settlement, and a large diaspora user base is structurally advantaged in that business. That is a real, bounded, falsifiable claim. You can measure corridor volume. You can measure cost per remittance. You can measure whether stablecoin supply on the chain grows.

The graph clarifies what sentiment confuses. The graph to watch is not the price chart of any AI-adjacent token. It is stablecoin supply on BSC, quarterly, and corridor-level remittance volume beneath it.

Contrarian

The consensus reading of this briefing is that it was a bullish narrative event that disappointed on specifics. That framing is backwards. Specificity was never the deliverable. The deliverable was the boundary.

Three boundaries were drawn, explicitly or implicitly, and each narrows the addressable claim.

Boundary one: the currency is not the chain. The candidates named were bitcoin, stablecoins, and AI-company tokens. BNB was named as a network, not as a monetary unit. The distinction is not pedantic. It means the thesis, if true, generates demand for fiat-pegged instruments and does not generate demand for a volatile exchange token. Anyone who read the briefing as a BNB demand thesis imported an assumption the speaker did not supply, and imported it at the exact moment when imports are most expensive.

Boundary two: the technology is not ready. "Still needs further development" is an admission, in the speaker's own words, in the speaker's own venue. It is the single most quotable sentence in the packet and it was quoted almost nowhere, because it does not fit the frame the coverage needed.

Boundary three: the market is not the developed consumer. The remittance recommendation concedes that the high-value consumer payment market is defended. That narrows the addressable market from all payments to corridors with structural friction.

Narrowing three times in one briefing is not weakness. It is, oddly, the most credible thing about it. Promotional rhetoric usually inflates. This deflated, twice, unprompted, in front of a capital-markets audience.

Which produces an uncomfortable conclusion for both sides. For bulls, the briefing does not support the trade they want to make. For bears, the briefing is more honest than the coverage of it, and the honest portion describes a business that could work.

One more cut. The absence of any token mechanism is not oversight. It is information. An organization holding a designed token model would describe it, because description is free and the market pays for it immediately. Silence on mechanism, in a venue optimized for market reception, implies the mechanism does not exist in publishable form. The constraint might be legal. It might be design incompleteness. Either way, the silence is the entry.

Ledger lines reveal what noise obscures. The line missing from the ledger is frequently the most informative entry on the page, and it is the one nobody reconciles.

Takeaway

Nothing in the packet justifies a position. That is not a bearish call. It is a statement about information sufficiency. You cannot size a trade against an adjective.

What the packet justifies is a watchlist, held to five items with explicit triggers.

A numbered proposal is the first. If BNB Chain's governance venues produce a BEP or EIP touching agent payments, machine identity, or compute attestation, the narrative converts from speech into engineering and the credibility discount compresses. Trigger: a proposal number and an author list.

An issuer identity is the second. Not the token — the issuer. Legal entity, jurisdiction, the identity of the compute auditor, the rights schedule. Trigger: a published term sheet or offering document.

A filing is the third. If BNB Standard exists and is listed, there are documents. Holdings, leverage, lockups, fee structure, related-party transactions. Trigger: a disclosure naming a BNB position.

Stablecoin supply on BSC is the fourth, and the cleanest proxy for whether the payment thesis is real, because stablecoin supply is the working capital of a corridor. Trigger: two consecutive quarters of meaningful net growth not attributable to a single issuer's promotional program.

The standards race elsewhere is the fifth. If Ethereum or Solana lands a widely integrated agent payment standard, the question of which chain clears machine payments is answered, and BNB Chain's role reverts to issuance venue — a fine business, a smaller one.

Five signals. None of them is a post. All of them are verifiable inside a quarter.

The larger question this briefing raised and then declined to answer: when machines transact at a scale where human review is no longer per-transaction, who keeps the ledger?

An exchange with a listing business will answer: the chain. A dollar issuer will answer: the rail. A regulator will answer: the licensee. The data answers differently. The ledger will be kept by whoever absorbs the variance, because variance absorption is what a settlement layer actually sells. Everything else is plumbing with a brand.

Chains do not absorb variance. They record it.

That is the distance between a currency and a receipt. The briefing described a currency. The engineering underneath it points at a clearing house — and clearing houses have never cared which chain you settle on, only how cheaply you can prove you did.