If T+1 was hard, why is same-day or atomic settlement not simply one more day of compression?

Netting needs a window, and the window is what you are removing

Multilateral netting is the largest single efficiency in post-trade, and it is bought entirely with time. Trades accumulate over a period, the clearing house nets each member's buys and sells per line of stock and per currency, and what actually moves at the depository is one net delivery and one net payment. Gross activity of millions of shares becomes a delivery of a few hundred thousand, and the cash follows the same collapse.

That only works if there is a period to accumulate over. If a trade settles the moment it is agreed, there is nothing left to net it against, because the offsetting sale has not happened yet. So the interesting property of same-day and atomic settlement is not the calendar; it is that the shorter the cycle, the smaller the netting set, and at the limit the netting set is one trade.

One member, one line of stock, one day

  Buys      1,000,000 shares      Sells      900,000 shares

Netted at the end of the day
  Receive     100,000 shares      Pay one net cash amount
  Stock needed on hand to deliver: none

Settled gross, trade by trade
  Deliver     900,000 shares across the sells, as they occur
  Receive   1,000,000 shares across the buys, as they occur
  Stock needed on hand to deliver: whatever has not yet arrived
  Cash needed: the gross value of every purchase, before any sale pays you

The arithmetic is deliberately simple, and the last two lines are the whole argument. Under netting the member never had to own the 900,000 shares it sold, because the purchases covered them by the time anything moved. Settled gross, each sale needs stock in the account at that moment, and it does not care that the matching purchase settles ten minutes later.

Sequencing suddenly matters

Under netting, the order of the day's trades is irrelevant to settlement, because only the net result is instructed. Settle gross and the order becomes load-bearing. Selling stock you are about to receive fails if the delivery arrives after your own obligation falls due, and a chain of such dependencies, where A owes B what B owes C, can lock up entirely if the first delivery in the chain is missing.

Depositories handle this today with machinery netting mostly makes unnecessary: partial settlement, so an available fraction moves rather than nothing; shaping, where a large obligation is split into smaller ones that can settle as inventory allows; recycling of failed instructions through later cycles; and optimisation runs that look for circular chains and settle them together. In a market that settles trade by trade in real time, those tools have less to work with, because there is no batch to reorder.

What has to be pre-positioned, and what that costs

The consequence is that both legs have to be there in advance. Cash must be pre-funded, so the intraday liquidity a firm needs stops being the net figure it funds each morning and becomes the peak gross outflow during the day. Securities must be pre-positioned, which is a genuine constraint for anyone whose inventory is out on loan, held at a different custodian, or expected from an unsettled purchase in another market.

Neither is free. Pre-funded cash is a balance sheet item earning nothing at the moment it is idle, and pre-positioned stock is stock not lent, which removes a real revenue line and, at scale, reduces the supply that makes short selling and market making possible. The efficiency that netting produced was not accounting; it was the reason the same activity could be supported with a fraction of the liquidity.

Two constraints deserve naming separately because they are on different clocks. The currency leg of a cross-border purchase is bought in a market whose own conventional spot settlement is not necessarily aligned with the securities cycle, so the funding trade can settle after the delivery it was meant to pay for unless it is arranged earlier or funded another way; this is exactly the problem the move to T+1 created for non-domestic buyers of US equities and it gets sharper, not milder, as the cycle shortens. Securities lending recall is the mirror image: a lender must recall stock in time for the borrower to return it before the delivery is due, and each compression removes hours from a process that involves another firm's operations team.

You are trading one risk for another, not reducing risk

The case for compression is sound and worth stating precisely. A shorter interval between execution and settlement means less time during which a counterparty can fail while the trade is open, so the potential future exposure the clearing house must collateralise is smaller and initial margin falls with it. Less unsettled exposure sitting in the system is a real reduction in the thing that transmits a default.

What replaces it is liquidity risk and operational risk. Liquidity risk, because a firm that cannot fund a gross obligation at 11:00 fails a settlement even though it is solvent and would comfortably have met the net figure. Operational risk, because the repair window disappears: under T+2 a mismatched allocation could be fixed the following morning, and every hour removed from the cycle converts an exception that a person used to handle into one a system must handle or escalate within minutes. Fail rates rise when cycles shorten for exactly this reason, and the penalties and buy-in regimes that price fails then bite more often.

A candidate who presents atomic settlement as strictly better has missed that settlement risk and liquidity risk are substitutes. The reason markets settle on a cycle at all is that somebody decided the second risk was more expensive than the first.

Where this lands in the systems you would build

Three design consequences follow, and they are the reason this question is asked of engineers rather than of operations managers.

Inventory has to become a real-time service rather than an overnight position file. If a delivery can be due at any moment, the question "what is available to deliver right now, net of loans, pledges and pending instructions" has to be answerable in milliseconds and has to be right, which is a different system from a batch that produces a position as of last night.

Exception handling moves onto the critical path. A break queue worked by people between 07:00 and 09:00 is an architecture that assumes slack. Without slack, matching, enrichment and repair need automated resolution with an escalation clock, and the meaningful metric becomes the age of the oldest unresolved item rather than the count.

And end-of-day stops being a boundary. Valuation, risk and reporting all inherit a cut-off from the settlement cycle, and a market that settles continuously has no natural moment when yesterday is finished. Choosing that moment becomes a policy your systems must implement consistently, rather than a fact the market hands you.

Every hour taken out of the settlement cycle buys a reduction in counterparty exposure and pays for it in netting, funding and repair time. The engineering question is never how to go faster; it is what you will do with the slack you have just removed.