Marble cuts sanctions-matching latency 50% with open-source upgrade

KUALA LUMPUR, 25 JULY 2026 —

Marble, the Paris-based AI decision hub for fraud and anti-money laundering screening, has built an open-source sanctions-matching engine called Motiva that cuts end-to-end latency by half compared with the original Python implementation.

The company replaced its Python-based screening layer with a Rust reimplementation of OpenSanctions’ yente matching algorithms, according to a case study published by Marble. Memory consumption now stays flat even when concurrent requests triple, while the 99th-percentile response time remains under 200 milliseconds during traffic spikes, Marble said.

Marble engineered Motiva to handle high-concurrency, real-time screening as its customer base expanded. Antoine Popineau, the Marble engineer who started Motiva as a side project, said he avoided designing a new scoring methodology from scratch.

“We didn’t want to invent another scoring wheel,” Popineau said. “OpenSanctions already had battle-tested matching logic we could extend.”

OpenSanctions, the Berlin-based open database of sanctions targets and politically exposed persons, welcomed the work. “Marble didn’t need to license a black-box engine or reverse-engineer a competitor’s method,” said Friedrich Lindenberg, founder of OpenSanctions. “They rebuilt yente’s matching layer to fit their use case and shared the result back into the ecosystem.”

The collaboration runs both ways. OpenSanctions has since rewritten its own rigour text-normalisation library and the nomenklatura scoring engine in Rust, lifting /match endpoint throughput by about 50% in the latest yente release. Motiva now mirrors OpenSanctions’ logic-v2 algorithm, improving cross-language matching and reducing false positives in Arabic, Cyrillic and Chinese name variants.

Marble adopted Motiva across its FRAML data pipeline, handling simultaneous checks against sanctions lists, politically exposed persons and watchlists. The switch lowered infrastructure costs because the same hardware now processes 1.7 times more screening requests per hour, the case study shows.

yente, the open-source API behind the matching, ships with pre-built Docker containers and is licensed under Creative Commons 4.0 Attribution Non-Commercial. It is not limited to OpenSanctions data: banks, fintechs and market infrastructures use it to screen proprietary customer and transaction datasets.

“Open infrastructure lowers the cost of building compliance systems,” Lindenberg said. “Teams can deploy, extend and validate against a transparent methodology without vendor lock-in.”

Popineau said Marble runs reference tests on every commit that compare Motiva’s scores against the original Python engine to ensure accuracy. “Without open code we couldn’t have reached this performance,” he said. “The tests keep us honest.”

OpenSanctions plans to fold Motiva’s Rust components into the next yente release, giving every downstream user an immediate performance gain. The partners expect further speed-ups as they tune logic-v2 for high-frequency screening in trading and remittance systems.

“In a traditionally siloed industry, shared open infrastructure raises the baseline for everyone,” Lindenberg said. “That’s how we collectively fight financial crime.”