Stream processing systems compute financial microstructure metrics in real time, yet existing observability monitors only infrastructure indicators and cannot detect semantic correctness violations. This work proposes a three-level observability model and demonstrates that formal definitions of metrics such as Order Flow Imbalance inherently define runtime-checkable invariants detecting data corruption invisible to infrastructure monitoring. A taxonomy of streaming operators for financial metrics is introduced, classifying them by state requirements and order sensitivity. Fault injection experiments show that standard monitoring misses realistic data quality failures while semantic invariants reliably detect them with O(1) per-event overhead.

Колычев Е.И. (науч. рук. Перл И.А.) Semantic observability in stream processing: runtime correctness verification for online market microstructure feature engineering // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2026]. URL: https://kmu.itmo.ru/digests/article/18138