rApp Observability: Causal Attribution of KPI Changes to rApp Actions in O-RAN Non-RT RIC

Concurrent rApps (third-party applications running atop the Non-RT RIC) introduce a fundamental observability gap: when a Key Performance Indicator (KPI) changes, operators cannot systematically determine which rApp action caused it. Existing O-RAN conflict mitigation operates proactively or reactively within the control loop but offers no post-hoc causal attribution.We propose an observability architecture for the Non-RT RIC that captures timestamped rApp actions across the R1, A1, and O1 interfaces, aligns them with KPI time series, and applies a causal attribution methodology combining changepoint detection, Granger causality, conditional intervention effect, and temporal precedence scoring. Evaluation with three concurrent rApps over a simulated multi-cell 5G setup, across three scenarios of increasing difficulty, shows that the combined method achieves high accuracy when actions are temporally or spatially separated and consistently places the causal rApp within the top-2 candidates under concurrent overlapping actions on shared cells.