The Timeout That Releases Cash Already Committed

A participant timeout must not release reserved wholesale CBDC after that participant has durably prepared a settlement. Configuring reservation expiry to follow an RPC deadline quietly gives a transport setting authority over money. The defect survives code review because the expiry handler looks reasonable in isolation, while ordinary integration tests never lose the reply after commitment.

The configuration crosses a protocol boundary

Consider a wholesale CBDC settlement design in which a coordinator collects durable prepare acknowledgements before recording a shared commit decision, an architectural example rather than a claim about WCBDCVault’s implementation. A participant configuration that maps request expiry to reservation cancellation is unsafe once preparation promises that the participant can honour the eventual decision. The coordinator may have durably committed while its response remains undelivered, so local expiry can make cash spendable even though the corresponding currency or security leg is already entitled to settle. The binding condition is the dependency between transfers, consistent with the New York Fed’s explanation of atomic settlement, and a missing response supplies no evidence that this dependency has been cancelled. Review therefore has to trace the effective timeout configuration into the participant state machine, including deployment overrides, rather than merely confirm that the cancellation handler is authenticated and idempotent.

A valid signature cannot resolve an unknown outcome

Post-quantum transaction authentication using ML-DSA can establish who authorised a settlement instruction, but it cannot turn an unanswered request into proof that the settlement aborted.

Run the lost reply test today

In an isolated settlement environment using the deployment’s rendered configuration, set the client request deadline to 500 ms, a test stimulus rather than a production recommendation. Submit a funded settlement through a controllable proxy, use the coordinator’s durable decision event as the trigger, and suppress the commit response and subsequent decision delivery to the affected participant until the client deadline has elapsed. While delivery remains blocked, query that participant’s authoritative reservation state and attempt a competing debit against the same reserved balance, because an application status page alone cannot establish whether the funds became spendable. Restore delivery and resubmit the original operation with its original settlement identifier, then inspect the decision record and ledger postings to verify that recovery applies the existing decision without duplicating the transfer. Fail the release if the competing debit consumes the reservation or recovery produces duplicate postings, and treat a run that never reaches the durable commit boundary as inconclusive rather than passing.

Keep the uncertainty contained

Separate the client deadline from the protocol authority to release funds: after durable preparation, release requires a valid abort decision under the settlement protocol, while commitment requires completion of the reserved transfer. Where a protocol supports expiry, its rules must make expiry mutually exclusive with a valid commit, rather than letting each participant infer abort from its own clock. A surviving coordinator must recover decisions from durable replicated state and reject commands from a superseded coordinator, while participants preserve reservations and deduplicate decision application across restart. Financial infrastructure standards such as the Principles for Financial Market Infrastructures frame operational resilience and settlement finality as obligations; the engineering consequence here is that unaffected funds can remain usable while an unresolved reservation stays unavailable. The remaining decision is how much liquidity to admit into unresolved settlements before throttling new work, because releasing uncertainty back into the spendable balance only converts a recovery delay into a settlement integrity failure.

Back to WCBDCVault · Domains

← Back to WCBDCVault
Engineering & Architecture Brief

The Timeout That Releases Cash Already Committed

A participant timeout must not release reserved wholesale CBDC after that participant has durably prepared a settlement. Configuring reservation expiry to follow an RPC deadline quietly gives a transport setting authority over money. The defect survives code review because the expiry handler looks reasonable in isolation, while ordinary integration tests never lose the reply after commitment.

The configuration crosses a protocol boundary

Consider a wholesale CBDC settlement design in which a coordinator collects durable prepare acknowledgements before recording a shared commit decision, an architectural example rather than a claim about WCBDCVault’s implementation. A participant configuration that maps request expiry to reservation cancellation is unsafe once preparation promises that the participant can honour the eventual decision. The coordinator may have durably committed while its response remains undelivered, so local expiry can make cash spendable even though the corresponding currency or security leg is already entitled to settle. The binding condition is the dependency between transfers, consistent with the New York Fed’s explanation of atomic settlement, and a missing response supplies no evidence that this dependency has been cancelled. Review therefore has to trace the effective timeout configuration into the participant state machine, including deployment overrides, rather than merely confirm that the cancellation handler is authenticated and idempotent.

A valid signature cannot resolve an unknown outcome

Post-quantum transaction authentication using ML-DSA can establish who authorised a settlement instruction, but it cannot turn an unanswered request into proof that the settlement aborted.

Run the lost reply test today

In an isolated settlement environment using the deployment’s rendered configuration, set the client request deadline to 500 ms, a test stimulus rather than a production recommendation. Submit a funded settlement through a controllable proxy, use the coordinator’s durable decision event as the trigger, and suppress the commit response and subsequent decision delivery to the affected participant until the client deadline has elapsed. While delivery remains blocked, query that participant’s authoritative reservation state and attempt a competing debit against the same reserved balance, because an application status page alone cannot establish whether the funds became spendable. Restore delivery and resubmit the original operation with its original settlement identifier, then inspect the decision record and ledger postings to verify that recovery applies the existing decision without duplicating the transfer. Fail the release if the competing debit consumes the reservation or recovery produces duplicate postings, and treat a run that never reaches the durable commit boundary as inconclusive rather than passing.

Keep the uncertainty contained

Separate the client deadline from the protocol authority to release funds: after durable preparation, release requires a valid abort decision under the settlement protocol, while commitment requires completion of the reserved transfer. Where a protocol supports expiry, its rules must make expiry mutually exclusive with a valid commit, rather than letting each participant infer abort from its own clock. A surviving coordinator must recover decisions from durable replicated state and reject commands from a superseded coordinator, while participants preserve reservations and deduplicate decision application across restart. Financial infrastructure standards such as the Principles for Financial Market Infrastructures frame operational resilience and settlement finality as obligations; the engineering consequence here is that unaffected funds can remain usable while an unresolved reservation stays unavailable. The remaining decision is how much liquidity to admit into unresolved settlements before throttling new work, because releasing uncertainty back into the spendable balance only converts a recovery delay into a settlement integrity failure.

Related Architecture Specifications (PQCECOSYSTEM): mlkemguard.com · mlkemshield.com · pqcecosystem.com

More Technical Articles on wcbdcvault.com

Central Bank Digital Currency Cybersecurity: Threat Modeling Sovereign Digital Money Against Quantum Adversaries →Project Agorá and the Unified Ledger: Reimagining Global Cross-Border Payments with PQC →