The problem it removes
Settlement across two institutions has at least two legs: value leaves one party and arrives at the other, and both sets of books have to record it. Sequenced execution can fail between those legs. One side is debited and the other is never credited, or value moves on chain and one party’s ledger never records it. Every mechanism built to survive that gap, whether a settlement window, a reconciliation process, or a correspondent holding the balance in between, exists because the gap exists. Atomic execution removes the gap that other mechanisms manage.How it works on a shared ledger
Where both legs are recorded on the same ledger, they can execute in a single transaction. The transaction either commits, in which case every leg is recorded, or it fails, in which case none are. On a settlement rail, the paying-side decrement and receiving-side credit are one event. The books balance by construction instead of by agreement. The same property applies to an asset leg and a cash leg executed together. See Delivery vs Payment.What it depends on
Atomicity is the all-or-nothing guarantee within a single ledger’s transaction boundary. Where settlement crosses two ledgers, the guarantee comes instead from the messaging protocol: value is committed on the sending side, and a verified acknowledgement confirms the receiving side acted. If the receiving side rejects, or the message times out, the sending side reverses its commitment. That guarantee is not the same as single-transaction atomicity, and the difference matters operationally. Cross-ledger settlement has an interval during which the outcome is pending. Single-ledger settlement does not. See How IBC Works for the timeout and acknowledgement model.This page is a placeholder for the operational detail. The atomic settlement
combinations, behaviour when a cross-ledger leg times out, and operator surface
for a pending settlement still need to be documented.