The Unsigned Claim: Taking Apart 'Own Your Payment Rails' at the Settlement Layer

"Signature invalid."

That is the first gate every transaction hits at my node. No valid ECDSA signature, no mempool inclusion. No inclusion, no state transition. The rule is deterministic and unforgiving: an unverified claim does not propagate. The mempool is a filter, and the filter is the point.

Journalism has no such filter. Over the past seven days, a claim moved through crypto media at near-instant speed — a "Bastion CEO" saying large enterprises want to own their payment rails rather than rent them from Visa or Mastercard. No name attached. No registered entity. No named customer. No number. No timestamp. The claim was unsigned, and it replicated anyway, through every aggregator in the industry, inside a single news cycle.

I spent an afternoon trying to verify it. There was nothing to verify. So I did the only thing I know how to do: I pulled the claim apart at the settlement layer to test whether the assertion is even mechanically coherent. The interesting part of this story is not the CEO. It is the fee stack that never makes it onto a slide.

I have to fix the mechanics before the argument, because "payment rails" is a phrase that does enormous unexamined work.

A card payment is a four-party model. Four parties: the cardholder, the issuing bank (which gave the cardholder the credential), the acquiring bank (which holds the merchant's account), and the merchant. Visa and Mastercard are not banks. They are switches and rulebooks — they route authorization messages and set operating rules, but they do not hold funds at the moment of authorization. That distinction matters later, because "challenging Visa" is not the same as "challenging the banks."

A card transaction runs in three phases. Authorization: an ISO 8583 message goes from merchant to acquirer to network to issuer, and the issuer returns an approve or decline in under two seconds. Clearing: at the end of the day, the network nets all the authorized transactions and computes who owes whom. Settlement: the net obligations move between banks over interbank rails, typically T+1. The card network is the switch and the netting engine. The money is a bank balance the whole time.

When a merchant accepts a card, it pays a merchant discount rate, the MDR. The MDR decomposes into three layers: interchange, the fee that flows to the issuing bank, typically 1.5% to 3% on consumer credit, and the dominant component; assessment or network fees, what flows to Visa or Mastercard, usually a fraction of a percent plus fixed per-transaction cents; and acquirer or PSP markup, what the merchant's processor charges for gateway, risk, and settlement services.

The first line is the one everyone skips. Interchange is the largest cost, and it does not go to the network. It goes to the bank that issued the card. Hold that.

On the other side sits the stablecoin rail. A stablecoin transfer is not a message; it is a signed state transition. On an EVM chain, an ERC-20 transfer is a call that mutates a balance mapping in a contract's storage. On Solana, it is an instruction executed against a token account. The rail is the chain, the settlement asset is the token, and the finality is whatever the chain's consensus guarantees. There is no netting engine, no T+1 cycle, no ISO 8583. There is a transaction, a signature, and a new state root.

The news itself was thin. Five information points, four of them restatements of a single opinion — enterprises want to own rails, stablecoins may take share from card networks, stablecoins may lower costs — sourced to a "Bastion CEO" with no name, published as a brief on Crypto Briefing. No data. No product. No token. No customer. That is the entire input. And a brief that thin is a sample point for narrative density, not a datum for decision-making. I treat it as a thermometer, not a signal.

Two rails, one ledger

Start with the obvious. Both systems are ledgers. The card network maintains a claim ledger — who owes whom — settled net at the end of a cycle through interbank rails. The stablecoin system maintains a balance ledger — who holds what — settled continuously on-chain.

The difference is not "old versus new." It is where the ledger lives and who is allowed to write to it.

On the card side, the ledger is permissioned and bilateral. The issuing bank writes to it. The acquiring bank writes to it. Visa arbitrates. You cannot append a balance unless you are a member of the club, and the club has a rulebook, a brand, and a compliance department.

On the stablecoin side, the ledger is permissionless and global. Anyone with a key can write, subject to gas. The token contract is the rulebook, and the rulebook is open source. There is no membership. There is no arbiter. There is no brand policing who may transact.

This is the real content of "own versus rent." When an enterprise says it wants to own its rail, what it is actually saying is: I want write access to my own settlement ledger. That is a coherent desire. It is also a much smaller claim than "replace Visa." Write access to a private ledger is not the same as membership in a global network. The two get conflated in every retelling of this story.

The cost stack, itemized

Here is where "stablecoins lower costs" must be tested with arithmetic, not adjectives. Let me lay both stacks side by side.

Card stack, per transaction: - Interchange: 1.5%–3.0%, to the issuing bank - Network/assessment: ~0.10%–0.20%, to the network - PSP/acquirer markup: ~0.30%–1.00% - Total MDR: roughly 2.0%–4.0% for card-present consumer credit

Stablecoin stack, per transaction: - On-chain gas: variable, often cents on L2s, dollars on L1 during congestion - On/off-ramp spread: ~0.10%–0.50%, fiat to token and back - Custody/MPC fees: ~0.05%–0.30%, key management and treasury operations - KYC/AML orchestration: fixed plus per-verification cost - Reconciliation and accounting: operational, per settlement batch - Reserve float opportunity cost: implicit, non-zero - Total: highly variable, and the headline "near-zero" ignores everything after gas

The claim "stablecoins lower costs" is true only if you compare on-chain gas to interchange and stop counting. That is not a cost comparison. That is a comparison of one line item to another line item's replacement. The moment you net the full stack — ramp spread, custody, compliance, reconciliation, and the float you give up — the delta compresses hard, and for small transactions it can invert.

I built a small notebook for this, the way I do for every claim I care about. A single function, two cost models, one solve-for-breakeven. It is reproducible, and that is the point. If you cannot reproduce a cost claim in code, you are reading marketing.

The honest version of the claim is narrower: stablecoin settlement can lower the settlement and float layer for large, low-frequency B2B transfers, while offering little to nothing for high-frequency retail, where the cost is dominated by fraud risk and dispute handling, not clearing.

A pseudocode interlude

I reduce every cost claim to a function, because prose hides assumptions and code does not. Here is the shape of the model I ran.

def card_net_cost(amount, interchange=0.022, network=0.0015, psp=0.005, dispute_rate=0.001, dispute_cost=25):
    variable = amount * (interchange + network + psp)
    disputes = dispute_rate * dispute_cost
    return variable + disputes

def stable_net_cost(amount, gas=0.02, ramp=0.003, custody=0.0015, kyc=0.35, reconcile=0.05, float_bps=0.0005): variable = amount * (ramp + custody + float_bps) fixed = gas + kyc + reconcile return variable + fixed

def breakeven(dispute_rate=0.001): # solve for amount where card == stable ... ```

Two things jump out. First, the card model is almost entirely variable — it scales with amount and barely with count, except for the dispute term, which scales with count and is amortized across the network. Second, the stablecoin model carries a fixed cost per transaction — gas, KYC, reconciliation — that does not scale down with amount. That fixed term is why small stablecoin payments are expensive and large ones are cheap. The crossover is a ticket-size threshold, not a universal win.

The entire "stablecoins are cheaper" debate reduces to one number: the ticket size at which the fixed compliance cost per transfer is amortized below the variable interchange it replaces. Below that number, cards win. Above it, stablecoins win. There is no global answer, and anyone who gives you one is selling something.

Where the money actually goes

The narrative gets this wrong, and it is a two-line fix.

The merchant pays the MDR. The biggest slice, interchange, goes to the issuing bank. The network takes a small assessment fee. So when a headline says "stablecoins challenge Visa and Mastercard," it points at the wrong balance sheet.

If stablecoin rails genuinely displace card volume, the first casualty is not the network. It is the issuing bank's interchange revenue, and behind it the independent sales organizations — the ISOs — that live on interchange splits. The network's network fee and brand fee are the stickiest parts of the stack, because they are tied to acceptance, and acceptance is a network property, not a cost property.

I learned this the hard way in 2020, during my opcode autopsy of the SushiSwap fork. I spent six weeks disassembling the constant-product AMM and mapping every SLOAD and SSTORE to gas cost. The lesson was not that gas was cheap or expensive. The lesson was that the cost you can see is rarely the cost that dominates. In that contract, the dominant cost was not the swap math — it was the storage writes in the fee accounting. Same structure here. The visible fee (network assessment) is not the dominant fee (interchange). You have to read the whole stack, not the headline line item.

Opcode leaked. Liquidity drained. Same pattern, different layer. The cost that leaks is never the cost you were watching.

The float income nobody names

Now the hidden transfer. Stablecoin issuers do not earn from transactions. They earn from reserves. When you hold USDC, the issuer holds a dollar of short-dated Treasuries and keeps the yield. That is float income — a de facto issuance tax on every token in circulation. The issuer's margin is not a spread on payments. It is the risk-free rate applied to a float it did not have to earn.

This reframes "stablecoins lower costs." Lower for whom? If the merchant pays less interchange but the value accrues to the stablecoin issuer's reserve income and to the middleware provider's markup, then the cost did not disappear. It moved. The merchant discount became an issuance spread plus a custody fee plus a ramp spread. Three line items replaced one, and none of them are printed on the receipt.

The source material never asks this question. It presents cost reduction as a one-way benefit, with no beneficiary named and no counterparty named. That is not analysis. That is a press release wearing an analysis costume.

I modeled this in Python during my 2025 work on modular DA layers. The exercise was the same shape: take a volume, run it through both systems, and solve for the break-even where the new system's net cost beats the old system's net cost. For DA layers, the answer depended on validator consolidation and slashing assumptions. For payments, the answer depends on ticket size and dispute rate. The result here was unsurprising and unhelpful to the narrative: the break-even sits at high ticket sizes with low dispute rates. Below that line, the card network wins on total cost, because fraud and dispute handling are priced into interchange and amortized across the whole network.

Settlement finality and the missing chargeback

Here is the trade-off no one in the "own your rail" camp wants to discuss.

Card settlement is reversible by design. A chargeback is a first-class primitive. The cardholder can dispute, the issuing bank can reverse, and the network arbitrates. That reversibility is not a bug. It is the product. It is why a consumer trusts a credential they cannot see, and why a merchant accepts it despite the cost. Reversibility is what makes a stranger's promise credible.

Stablecoin settlement is irreversible by default. A confirmed transfer is final. There is no chargeback primitive in the ERC-20 standard. There is no arbitration layer. There is no issuing bank to call. Once the state root updates, the write is permanent, and permanence is a feature, not an oversight.

This is the structural asymmetry that determines where each rail wins: - Reversible, high-trust, retail, consumer-facing → card rail. - Final, low-trust-needed, B2B, large-ticket → stablecoin rail.

The moment you see it this way, "stablecoins challenge card networks" collapses into a much narrower claim: stablecoins challenge card networks only in the segment where reversibility is worthless — large B2B settlement between parties who already trust each other and who prefer finality to protection.

For retail, the reversibility is worth more than the fee. You cannot buy your way out of that with cheaper clearing. The consumer does not pay the MDR directly; the merchant does, and the merchant passes it through in prices. The consumer experiences reversibility as protection, and protection is not a cost they will trade away for a fraction of a basis point they never see.

State root mismatch. Trust updated. The ledger that looks cheaper is the ledger that dropped a primitive you still need.

A rail is a consensus mechanism

Now the deepest problem, and the one that sinks the "everyone builds their own rail" thesis.

A payment rail is not software you install. It is a consensus mechanism. Its value is a function of how many counterparties accept writes to it. Visa's rail is valuable because roughly four billion credentials can write to it and tens of millions of merchants can read from it. That is the network effect, and it is the actual moat. The fee is the price of admission to the network, not the price of the software.

If an enterprise builds its own rail, it controls the ledger. Good. But a ledger with one writer and a handful of counterparties has no network. It settles the enterprise against its own suppliers. It cannot reach the consumer, because the consumer holds no key, has no wallet, and will not adopt one to buy a coffee. A private rail is a private consensus, and a private consensus is just a database.

This is the internal contradiction the source material hides. "Own your rail" and "network effect" are in direct tension. The reason enterprises rent the card network is precisely that the rental price includes access to a network they could never build alone. You do not escape the network by building a private rail. You escape the network by shrinking your reach to the counterparties who will bother to connect to you.

So the realistic scope of "own your rail" is B2B, large-ticket, low-frequency, between parties with existing commercial trust. It is not a consumer rail. The narrative sells the consumer version and ships the B2B version. That gap is the story.

I traced exactly this kind of gap in early 2024, when I audited the L2 standard bridge contracts after the Arbitrum NFT bridge exploit. Fifteen thousand lines of Rust and Solidity, and the contracts were clean. The vulnerability was in the dApp wrapper — a race condition that allowed double-spend under specific network latency. The lesson generalized: the protocol is rarely the weak point. The integration layer is. Here, the protocol claim ("own your rail") is clean; the integration claim ("and reach consumers") is where it breaks.

The reversibility you cannot have both ways

Let me be precise about the trade-off, because it is the whole game.

Every settlement system must choose how it handles the interval between "funds committed" and "funds final." Card networks handle it with reversibility plus fraud liability allocation. Stablecoins handle it with finality plus a market of escrow and arbitration bolted on top, if any exists at all.

You cannot have both cheap finality and free reversibility. Reversibility has a cost: it requires a trusted arbiter — the issuing bank, the network — and it requires capital held against disputes. Finality has a cost: it requires you to accept that a mistake is permanent, which forces escrow, insurance, or pre-existing trust.

When the source says stablecoins lower costs, it is comparing final settlement to reversible settlement and pretending the reversibility was free. It was not. It was priced into interchange, and it bought the consumer the ability to dispute a charge. Remove the fee, remove the protection. That is the trade, stated honestly. The rail that costs less is the rail that protects less, and the market will price that difference once the novelty fades.

Latency at the aggregation layer

There is a second-order cost the narrative ignores: latency. I spent three months in 2022 reverse-engineering the Cairo VM's constraint system for a piece on StarkNet's proof aggregation, and the finding was that the bottleneck was not the proof itself but the aggregation of many proofs into one. Payments have the same shape. The on-chain transfer is fast. The aggregation — netting, reconciliation, treasury sweeps, off-ramp batching — is slow. A rail is only as fast as its slowest aggregation step, and for enterprise settlement that step is off-chain and human.

The dirty work is off-chain

One more thing the narrative skips entirely: the technical difficulty is not on the chain. It is off it.

A stablecoin payment rail needs fiat on/off-ramps, which means money transmission licenses in every jurisdiction it touches. It needs KYC/AML orchestration — identity verification, sanctions screening, transaction monitoring. It needs accounting reconciliation between on-chain balances and off-chain books, which is harder than it sounds because on-chain finality and accounting-period finality are different clocks. And it needs a chargeback equivalent, which, as established, does not exist natively, so it must be rebuilt as escrow or insurance or pre-funded dispute reserves.

None of that is glamorous, and none of it is in the source material. But it is what determines whether "own your rail" is a viable operating business or a slide. The compliance layer is where the cost lives and where the failure modes concentrate. A rail that is cheap on-chain and expensive off-chain is not a cheap rail. It is an expensive rail with a cheap-looking top layer.

Regulation is the rail

Let me state the thing the source never says: in payments, the license is the rail. Not the chain, not the token, not the wallet. The license.

A rail that moves value between fiat and token needs money transmission licenses in each jurisdiction, or a partner who has them. It needs sanctions screening against OFAC lists, and it needs to freeze and report. The regulatory perimeter, not the technical stack, decides who can operate. This is why the deepest moat in payments is not code. It is a license portfolio, and license portfolios are accumulated slowly and defended with legal budgets.

This cuts both ways for the "own your rail" thesis. On one hand, if regulation tightens and licenses become the barrier, the enterprise that builds its own rail still needs a licensed partner for on/off-ramps, which means it is not truly owning the rail — it is renting the compliance layer. On the other hand, if a large enterprise already holds banking relationships and licenses, the marginal cost of adding a settlement rail is lower, and the "own" thesis gets stronger for exactly those firms. The thesis is not wrong. It is conditional, and the condition is licensing, not technology.

The source presents "own versus rent" as a binary about technology preference. It is actually a question about regulatory capital. Most enterprises rent because renting the compliance layer is cheaper than owning it. That is the honest reason, and it has nothing to do with whether the technology works.

Now the part the narrative cannot see.

The contrarian claim is not that stablecoins will fail. It is that the real threat to Visa and Mastercard is not replacement. It is co-optation — and the networks are already executing it.

Both Visa and Mastercard have been routing stablecoin settlement into their own networks, settling obligations in USDC while keeping the acceptance layer and the rulebook intact. Read that move carefully. The network does not need to win the rail war. It needs to own the orchestration layer above whichever rail settles. If card networks become the routing and dispute layer on top of stablecoin clearing, then "stablecoins challenge card networks" is the wrong frame entirely. The networks absorb the rail and keep the fees that matter — the network and brand fees, which, as established, are the stickiest part of the stack.

Meanwhile, the actors the narrative never names are the ones actually exposed: issuing banks and ISOs, whose interchange economics are what a stablecoin B2B rail erodes first. And the fragmentation paradox cuts against the whole thesis. If every large enterprise builds a proprietary rail, each supplier must integrate N rails, and total friction rises rather than falls. A single shared network is more efficient than N private ledgers. That is why networks exist, and it is why the "everyone owns their own rail" world is less efficient, not more.

So the blind spot is this: "own your rail" is told as a story about cost, but it is really a story about control and reach. And on reach, the private rail loses by construction.

What would falsify the claim

I keep a rule: if a claim cannot be falsified, it is not a claim, it is a mood. So here is what would falsify "enterprises want to own their rails."

Falsifier one: no named enterprise ever publicly operates a proprietary settlement rail at scale within the next several quarters. If the first mover never names itself, the thesis is a slide, not a deployment.

Falsifier two: card networks disclose rising stablecoin settlement volume in their own earnings, while proprietary enterprise rails do not appear. That would show the demand is real but flowing into the networks, not around them — co-optation, not replacement.

Falsifier three: on-chain B2B settlement volume grows but is dominated by PSPs and acquirers, not by enterprises. That would show the rail is being built by the incumbents one layer down, which again is co-optation.

If any one of these lands, the "own your rail" story is reframed from disruption to absorption. If none lands, the story stays a mood. Either way, the deciding evidence is a named entity and a disclosed number, not a CEO's adjective.

The claim is unsigned, and unsigned claims should not move capital. But the trend underneath is real and worth tracking, because it will be decided by three hard metrics, not by a CEO's soundbite:

• Named enterprise customers publicly running a proprietary settlement rail.
• Stablecoin settlement volume disclosed in card network earnings calls.
• On-chain B2B settlement volume, net of float and compliance costs.

Watch those. Ignore the adjectives. If the first stays at zero and the second keeps rising, the story is not replacement. It is co-optation. And the fee stack nobody put on the slide will still be there, one layer down, waiting to be read.

⚠️ Deep article forbidden