The Final Review: One Question per Chapter

Main-line wrap-up Builds on: all main-line chapters

The main line lives in the twenty-eight chapters behind you. One question per chapter, plus one synthesis question to close. Answer them all, and the main line is done.

(The set is built to find gaps: whichever question stops you, that is the chapter to go back to. So it runs in chapter order, not difficulty order. The answers sit below the questions — answer for yourself before reading on.)

Beyond the main line are two optional chapters. The next chapter is about the payer turning into an AI agent; the one after it covers the generational replacement of four base layers — message standards, rail interconnection, central bank digital currencies, and Pay by Bank. Neither introduces a new judgment tool — they run on exactly the toolkit you just finished. For the cell-by-cell comparison of every rail covered, see appendix A at the end.

Questions¶

What is the nature of payment? Why does a book transfer cost the bank almost nothing?
What is clearing, and what is settlement? Why does the gap between them decide both the risk and the cost?
To judge how hard any form of "money" is, what two questions do you ask?
What is the wall the three-party model ran into?
What does a card network actually provide? You may not use the word "network" in your answer.
A $100 purchase carries $2.30 in fees. How does it split? Why does the biggest cut go to the issuer?
The instant the screen says "payment successful," whose hands is the money in?
Why does a liability shift work better than sending notices?
What is the difference between push and pull? Why must every pull leave the payer a return window?
Why can't you have both speed and reversibility?
Every country's instant rail is seconds-fast on the user's end. How do you tell which ones are final on arrival? In the ones that aren't, who is fronting the money?
Why does every country run its own domestic clearing rail, while cross-border is down to just two? How does each of the two get around "unify every country's clearing"?
In one cross-border remittance, where does the money go — and where does the information go?
OUR, SHA, BEN — choosing one changes what, exactly? Does the total fee change?
What does "we support SEPA Instant" leave unsaid, and why does that distinction matter to whoever builds the integration?
China's payments stack carries one layer most countries don't have. What capability did severing direct bank links strip from that layer, and what did centralized custody of customer funds strip from it?
How many clearing rails a market grows depends on which four variables? Taking that rule to an unfamiliar market, what do you look at first?
Of the four cross-border cost sources, which is the largest and which the heaviest? Why do the last two never show up on a single receipt?
Where did Wise move the cost from, and where did it move it to?
What are the three questions for judging any payments company?
Why does Revolut hold its own banking license while Chime doesn't? Answer with institutional structure — "strategic preference" is not allowed.
A user's balance at a neobank is whose liability? How many layers of risk is the user actually carrying?
A cross-border company has ample total assets, yet users in one country can't withdraw. Which of the five risks is this? Which tools apply?
Why is compliance a moat? What does the value of a license depend on?
A balance column on the users table, incremented and decremented directly on each transfer — why is that unacceptable in a payment system?
Where do stablecoins sit in the hierarchy of money? If there is no innovation in the hierarchy, where is the innovation?
Which leg do stablecoins replace, and which leg do they leave untouched?
What is the most fundamental difference between CPN and SWIFT? Why does the contest turn on neutrality?
Synthesis: someone brings you a "zero-cost" payment scheme. What are the three places you go looking for the hidden cost?

Answers¶

Answer for yourself before reading on.

Payment is editing a ledger, not moving money — what transfers is purchasing power, and the means is changing two lines of the ledger in step. A book transfer costs almost nothing because payer and payee are recorded on the same ledger: one line down, one line up, done — no claim on any other institution is created, so no clearing and no settlement are needed.
Clearing is working out who owes whom how much; settlement is actually extinguishing that debt with money both sides accept unconditionally. In the gap between them, the debt exists only on the books — the longer the gap, the bigger the risk that one side's default drags the other down (settlement risk); squeezing the gap to zero means having funds ready, in full, for every single payment (capital tied up). Risk and cost are two faces of the same gap.
Ask two questions: whose liability is this money? What is it redeemed with, and where do the reserves sit? Central bank liabilities (cash, reserves) are hardest; commercial bank deposits come next; wallet balances and stablecoins are the liabilities of private companies. The higher up the hierarchy, the less you have to trust that some company won't fail.
One company doing both jobs — vetting cardholders nationwide and signing merchants nationwide — eats the acquisition cost on both sides itself: every added city, every added industry stacks the cost linearly. The Diners Club-style three-party model holds in a small circle; it cannot carry national scale.
A card network sells enforceable rules: who may join, how messages travel, how disputes are judged, in how many days the money settles, how violators are punished — plus the power to make thousands of mutually distrusting banks all run on that one set. It issues no cards, signs no merchants, holds no transaction funds. What it stands on is a rulebook everyone is willing to obey.
Roughly $1.80 to the issuer (interchange), $0.13 to the card network (scheme fees), $0.37 to the acquiring side. The biggest cut goes to the issuer because the money follows the risk: bad debt and fraud losses land mostly on the issuer, and interchange is the compensation for carrying them — which is also why issuers bother to issue cards and pay cashback.
Still in the hands of the payer's issuer. Authorization is only the issuer answering "this one may proceed" while freezing a slice of the limit — an exchange of information, with neither clearing nor settlement. The money moves only after the merchant captures at day's end, the network nets the positions, and issuer and acquirer settle — usually a day or two later.
Because a liability shift hangs the loss on the action: whoever doesn't upgrade (no 3DS support, no chip cards) is the one the fraud loss gets awarded to. Each institution runs its own numbers and upgrades on its own — nobody has to stand over it. A notice has no such mechanism; the cost never lands on the laggard.
The difference is who initiates: in a push, the payer sends the money out; in a pull, the payee takes it from the payer's account on the strength of an authorization. In every pull, the money is taken off the account by someone else — deny the payer a window to change their mind after the fact (60 days on ACH consumer accounts, 120-plus for card chargebacks) and nobody would dare hand over their account. That same window is the payee's risk exposure.
Because reversibility presumes the money hasn't truly moved yet. Batch netting defers settlement to end of day, leaving room in the middle for returns; a rail that settles each payment in real time is final on arrival, and reversal would mean clawing money back out of the payee's account — which can't be done. So the faster, the less reversible — and fraud control is forced forward, from recovering after the fact to blocking before it.
Look at whether the money for the leg between institutions is present. Pix, SCT Inst, and FedNow move it payment by payment, in full, on the central bank's ledger; RTP moves money prefunded into a pool — either way the money is there, so arrival is final. UPI and China's IBPS record first, net, and settle later — at the moment the user sees the money arrive, the receiving institution is fronting its own money, carrying intraday exposure until the settlement window.
Domestic rails end on their own central bank's ledger and grow around domestic rules — one per country, none interoperable. The bar for "global" is touching no country's monetary sovereignty, and only two ever cleared it. SWIFT plus correspondent banking gets around it by not touching the money: funds move locally inside each country; only the information crosses the border. The card networks get around it by not touching the local: they write their own rulebook and outsource issuing and acquiring to each country's local banks.
The money never crosses the border at any point: the dollar leg moves between accounts inside the US and finally lands on the Fed's ledger; the receiving country's leg settles locally in local currency; the middle rides on the balances banks pre-place with one another in correspondent accounts. The only thing that crosses the border is information — the payment message relayed hop by hop down the correspondent chain.
The only thing that changes is which side of the principal the same intermediary fees land on: with OUR the payer pays them all on top (the payee receives in full); with SHA each side bears its own leg; with BEN everything comes out of the principal. The total fee does not change by a cent. The product consequence: under SHA and BEN, what the payee will actually receive cannot be promised at send time.
What it leaves unsaid is the clearing and settlement mechanism. The sentence establishes only that the bank has joined the EPC's SCT Inst scheme and accepted the obligations in that rulebook; whether it clears through TIPS, RT1, or a national clearing house, the scheme layer does not say.

It matters to the integrator because the pipe decides three things: what money settles it (TIPS is central bank money, RT1 is prefunded — different counterparty risk), how wide the blast radius is when it breaks, and which party the reconciliation files come from. SEPA is a rulebook, not a system — flatten the two layers into one, and you will mistake a scheme-level promise for a pipe-level guarantee.

The extra layer is non-bank payment institutions: money moves between stored-value account balances, and on a given day the two banks behind them may exchange nothing at all.

Severing direct links stripped the ability to bypass the clearing system — every cross-institution flow must now go through a clearing institution recognized by the PBoC, instead of dozens of private bilateral connections to banks. Centralized custody of customer funds stripped the ability to earn a spread on customer money — 100% of it sits at the PBoC, and no interest may be paid to customers.

One governs the flow, the other governs the stock. Only with both does that layer get pushed back from quasi-bank to pipe.

Four variables: amount, urgency, predictability, and refund rights. However many combinations a market cuts out of them is however many rails it grows — each combination has a different risk structure, the risk structure dictates the settlement mode, and the settlement mode dictates what participants owe each other.

On an unfamiliar market, look at two things first: whether bulk payroll and retail instant share one pipe (if they do, predictability was never used as a dividing line, and high-volume payroll gets none of the batch-netting cost advantage there), and whether large-value and small-value share one netted rail (if they do, that market tolerates more intraday credit exposure than most).

The largest is the FX spread — it takes the most from the user while the receipt reads "zero fees." The heaviest is prefunded liquidity — it grows linearly with the number of countries covered, independent of volume, a burden on the industry itself. Of the last two, one is smeared into the institution's overall pricing and the other shows up as a 1–5 day delay; neither is priced per transaction, which is why no single receipt ever shows them.
Wise used "local-in, local-out plus internal netting" to trade the user side's slow and expensive for tied-up capital and license burden on its own books: both ends run local clearing, and the cross-border leg becomes Wise periodically moving money to refill its pools. The weak point: a corridor must have volume both ways — on a one-way corridor the netting dies, and every payment means actually moving money.
Does it hold a license; does money cross its books; what losses does it eat. The three answers decide how much of the fee it can take, which box the regulator puts it in, and who its competitors are.
An institutional difference, and it comes down to two lines. The revenue side: the EU caps interchange at 0.2%, so issuing alone earns nothing — Revolut must do deposits, lending, and wealth products, which take a banking license; US credit card interchange is uncapped, and Chime can live on its issuing cut alone. The license side: the EU has a clear application path, and one Lithuanian banking license passports across the entire European Economic Area; the US has no newcomer-friendly federal path — getting your own means grinding state by state, and renting a sponsor bank's license is far cheaper. The same arithmetic under two rulebooks yields two answers.
Nominally the partner bank's liability (the money sits in the bank's FBO account) — but which slice is his is proven only by the neobank's internal ledger. So he carries at least two layers: the risk of the bank failing (backstopped by FDIC pass-through insurance, provided the ledger is clean), and the risk of the internal ledger going wrong. Synapse proved the second layer is the deadlier one: when the ledger breaks, the money might as well not exist.
Liquidity risk — not the absence of money, but money not being where it is needed. The tools: deepen that country's prefunded pool, arrange standby credit with local banks, or find offsetting flow in the other direction; a step further ahead, use cash forecasting so the rebalancing happens before the pool runs dry.
Because a license is time, and time can't be bought: applications run in years, on top of which come capital, a local team, and regulator trust that only accumulates — and a violation record follows the entity. The value of a license depends on how deep it lets you touch the money (passing instructions only, collecting and paying out on customers' behalf, or taking deposits) and which jurisdiction it covers — depth and geography; missing either one puts you back in the queue.
Because it stores only the result, never the process, and so cannot vouch for itself. When the balance is written wrong, there is no reference to reveal that it is wrong or which entry made it so. Double-entry books a debit and a credit for every change and forces them equal: a bad entry cannot be written, and any historical balance can be recomputed from the stream at any time — only what can be audited and reconstructed deserves the name ledger.
On the same layer as wallet balances: a private issuer's liability, redeemed against reserve assets. In the hierarchy there is no innovation at all — it even sits one rung below bank deposits. The innovation is the shared ledger: clearing and settlement complete in the same instant, 24/7, straight between any two addresses on earth. It traded a downgrade of the settlement asset for an upgrade of the settlement process.
What gets replaced is the middle leg — the funds transfer that used to hop through a chain of correspondent banks, a fee at every hop, days in transit, becomes a single on-chain transfer. The two ends stand untouched: fiat on- and off-ramps, licenses, compliance, FX — not one of them goes away. Which is why the competition is at the two ends, not on the chain.
SWIFT only carries messages; the funds are booked separately across correspondent accounts — message and money travel two roads. CPN builds the settlement asset (USDC) into the network itself: instruction and funds, same road, same instant. The contest turns on neutrality because Circle both writes the rules and issues the network's settlement asset: joining means handing the standard to a competitor. So the banks build their own tokenized-deposit networks, and the payment giants set up Open USD — the thing a network business fears most is fragmentation, and non-neutrality is exactly where fragmentation comes from.
Three places: the FX spread (divide what was received by what was paid and compare against the mid-market rate — ignore the fee column); speed and finality (is someone fronting money; has the return window actually closed); someone's balance sheet (prefunding, rolling reserves, risk exposure — the cost never disappeared, it moved to whoever in the chain can carry it). A scheme that survives all three checks with no cost found has not appeared in this course.

Previous: Chapter 28 · Where CPN Stands on the Grid Next (optional reading): Chapter 30 · Agentic Payments: When the Payer Is No Longer Human