A Protocol-Level, Reproducible Testbed for Post-Quantum Migration of RTGS: Engineering Realization and Validation of the Modular Decoupling Scheme
Abstract
Project Leap Phase 2 showed that replacing the classical RSA signature of the Eurosystem’s euro real-time gross settlement (RTGS) system (T2/ESMIG, in a test environment configured like production) with post-quantum cryptography (PQC) is feasible but triggers integration friction: a mean signature-verification time of 209.9 ms and a hybridisation barrier, since the existing verifier accepts only one signature algorithm. Companion analytical and architecture papers supply a cost-model envelope and a modular decoupling scheme, but neither has been run. This paper contributes a publicly released, protocol-level testbed for post-quantum migration of RTGS: every node is a real process exchanging real ISO 20022 pacs.008 messages over a real TCP envelope protocol, with nested ML-DSA-87 + RSA-2048 signatures and a hybrid key encapsulation mechanism (KEM; ML-KEM-768 + RSA-OAEP) at the Data Exchange Protocol (DEP) layer. The testbed enforces the Watchdog inequality online with exponentially weighted moving averages and reproduces Leap’s 209.9 ms regime under calibration, a consistency check whose sensitivity is quantified: native nested verify measures <inline-formula> <tex-math notation="LaTeX">$123~{\mathrm {\mu \text {s} }}$ </tex-math></inline-formula>, so the figure is protocol-shell cost. Under four fault scenarios the decoupled scheme holds the settlement-decision (provisional pre-credit) latency flat at <inline-formula> <tex-math notation="LaTeX">${\approx }{\,}4.4$ </tex-math></inline-formula> ms while the inline baseline grows to 212.9 ms (<inline-formula> <tex-math notation="LaTeX">$48\times $ </tex-math></inline-formula>), whereas final-settlement latency is unchanged (<inline-formula> <tex-math notation="LaTeX">${\approx }{\,}219$ </tex-math></inline-formula> ms vs. 214.8 ms). A sustained-load experiment with equal verifier budgets shows that final-settlement throughput is bounded by verifier capacity for both pipelines and quantifies pending exposure, queue growth and rollbacks. Ledger consistency is preserved throughout; the artifact is released for third-party reproduction.